Compare commits

..

33 Commits

Author SHA1 Message Date
Alexander Whitestone
ce3822bb5f feat: quality gate — validate all pipeline outputs (#623)
Some checks are pending
Architecture Lint / Linter Tests (pull_request) Waiting to run
Architecture Lint / Lint Repository (pull_request) Blocked by required conditions
PR Checklist / pr-checklist (pull_request) Waiting to run
Smoke Test / smoke (pull_request) Waiting to run
Validate Config / Python Test Suite (pull_request) Blocked by required conditions
Validate Config / Shell Script Lint (pull_request) Waiting to run
Validate Config / Cron Syntax Check (pull_request) Waiting to run
Validate Config / Deploy Script Dry Run (pull_request) Waiting to run
Validate Config / Playbook Schema Validation (pull_request) Waiting to run
Validate Config / YAML Lint (pull_request) Waiting to run
Validate Config / JSON Validate (pull_request) Waiting to run
Validate Config / Python Syntax & Import Check (pull_request) Waiting to run
Validates pipeline outputs before saving. Rejects bad entries,
tracks quality scores per pipeline.

Checks:
- Training pairs: prompt/response non-empty, response != prompt
- Scene descriptions: all required fields, description min length
- Knowledge entries: no placeholders (TODO, FIXME), min length
- Prompt enhancement: rich > terse length, min 20 chars
- Adversary entries: id/family/prompt present, min prompt length
- SOUL.md compliance: no human life valuation, no weapon/child content
- Deduplication: detects duplicate entries by key fields

Features:
- Auto-reject bad outputs with reasons
- Quality score per entry (0.0-1.0)
- Batch mode (--dir) for processing all JSONL at once
- Stats tracking (~/.hermes/pipeline/quality_stats.json)
- --status to view historical quality metrics

Usage:
  python3 pipeline/quality_gate.py --input data.jsonl --type training_pairs
  python3 pipeline/quality_gate.py --dir pipeline/output/
  python3 pipeline/quality_gate.py --status

Closes #623
2026-04-15 08:20:18 -04:00
817785d763 Merge pull request 'feat: training data augmentation — paraphrase and translate pairs (#695)' (#732) from fix/695 into main 2026-04-15 11:56:28 +00:00
Alexander Whitestone
3603030235 feat: training data augmentation — paraphrase and translate pairs (#695)
Some checks are pending
Architecture Lint / Linter Tests (pull_request) Waiting to run
Architecture Lint / Lint Repository (pull_request) Blocked by required conditions
PR Checklist / pr-checklist (pull_request) Waiting to run
Smoke Test / smoke (pull_request) Waiting to run
Validate Config / Playbook Schema Validation (pull_request) Waiting to run
Validate Config / YAML Lint (pull_request) Waiting to run
Validate Config / JSON Validate (pull_request) Waiting to run
Validate Config / Python Syntax & Import Check (pull_request) Waiting to run
Validate Config / Python Test Suite (pull_request) Blocked by required conditions
Validate Config / Shell Script Lint (pull_request) Waiting to run
Validate Config / Cron Syntax Check (pull_request) Waiting to run
Validate Config / Deploy Script Dry Run (pull_request) Waiting to run
augment_pairs.py: generates paraphrases and translations for any
JSONL training file.

Features:
- Auto-detects text field (rich, terse, text, content, lyric_line, etc.)
- N paraphrases per entry (template-based, or LLM with --llm-endpoint)
- Translations to ES, FR, DE (template dictionary, or LLM)
- Outputs augmented JSONL alongside originals
- Marks each augmented entry with _augmentation, _original, _language

Usage:
  python3 augment_pairs.py --input data.jsonl
  python3 augment_pairs.py --input data.jsonl --paraphrases 5 --langs es,fr
  python3 augment_pairs.py --input data.jsonl --llm-endpoint http://localhost:11434/v1

Closes #695
2026-04-15 07:51:38 -04:00
35a191f7b1 Merge PR #725: feat: Provider health monitor with auto-switch (#509) 2026-04-15 06:10:45 +00:00
e987e1b870 Merge PR #726: feat: Pre-flight provider check for session launch (#508) 2026-04-15 06:10:41 +00:00
19278513b4 Merge PR #727: feat: Three.js-specific glitch detection patterns (#543) 2026-04-15 06:10:38 +00:00
1088bf8983 test: add Three.js pattern tests and update assertions (#543)
Some checks are pending
Architecture Lint / Linter Tests (pull_request) Waiting to run
Architecture Lint / Lint Repository (pull_request) Blocked by required conditions
PR Checklist / pr-checklist (pull_request) Waiting to run
Smoke Test / smoke (pull_request) Waiting to run
Validate Config / YAML Lint (pull_request) Waiting to run
Validate Config / JSON Validate (pull_request) Waiting to run
Validate Config / Python Syntax & Import Check (pull_request) Waiting to run
Validate Config / Python Test Suite (pull_request) Blocked by required conditions
Validate Config / Shell Script Lint (pull_request) Waiting to run
Validate Config / Cron Syntax Check (pull_request) Waiting to run
Validate Config / Deploy Script Dry Run (pull_request) Waiting to run
Validate Config / Playbook Schema Validation (pull_request) Waiting to run
- Added TestThreeJsPatterns class with 14 tests
- Tests cover: pattern existence, severity inference, vision prompt
- Updated pattern count assertion (14+ patterns now)
- Updated demo test (6 glitches: 4 original + 2 Three.js)
2026-04-15 05:37:17 +00:00
94f0a132d4 feat: add get_threejs_patterns() filter function (#543) 2026-04-15 05:34:17 +00:00
279356bed6 feat: add --threejs flag and Three.js-aware severity inference (#543)
- Added --threejs flag for focused Three.js pattern scanning
- Updated _infer_severity with shader_failure, texture_placeholder,
  uv_mapping_error, frustum_culling, shadow_map_artifact categories
- Added Three.js demo detections (shader failure, shadow map)
- Bumped detector version to 0.2.0
2026-04-15 05:34:16 +00:00
511ff863c2 feat: add Three.js-specific glitch detection patterns (#543)
Adds 6 new Three.js-specific glitch categories and patterns:
- SHADER_FAILURE: Solid black materials from shader compilation errors
- TEXTURE_PLACEHOLDER: 1x1 white pixel stretched surfaces
- UV_MAPPING_ERROR: BufferGeometry UV coordinate errors
- FRUSTUM_CULLING: Objects popping at screen edges
- SHADOW_MAP_ARTIFACT: Pixelated/blocky shadow maps
- BLOOM_OVERFLOW: Excessive post-processing bloom bleed

Closes #543
2026-04-15 05:32:25 +00:00
b6e3a647b0 feat: add pre-flight provider check script (#508)
Some checks are pending
Architecture Lint / Linter Tests (pull_request) Waiting to run
Architecture Lint / Lint Repository (pull_request) Blocked by required conditions
PR Checklist / pr-checklist (pull_request) Waiting to run
Smoke Test / smoke (pull_request) Waiting to run
Validate Config / Deploy Script Dry Run (pull_request) Waiting to run
Validate Config / Playbook Schema Validation (pull_request) Waiting to run
Validate Config / YAML Lint (pull_request) Waiting to run
Validate Config / JSON Validate (pull_request) Waiting to run
Validate Config / Python Syntax & Import Check (pull_request) Waiting to run
Validate Config / Python Test Suite (pull_request) Blocked by required conditions
Validate Config / Shell Script Lint (pull_request) Waiting to run
Validate Config / Cron Syntax Check (pull_request) Waiting to run
- Checks OpenRouter balance via /api/v1/auth/key
- Tests Nous and Anthropic API keys
- Verifies Ollama is running
- Pre-flight check before session launch
- Returns exit code for automation

Closes #508
2026-04-15 03:55:04 +00:00
e14158676d feat: add provider health monitor script (#509)
Some checks are pending
Smoke Test / smoke (pull_request) Waiting to run
Architecture Lint / Linter Tests (pull_request) Waiting to run
Architecture Lint / Lint Repository (pull_request) Blocked by required conditions
PR Checklist / pr-checklist (pull_request) Waiting to run
Validate Config / Cron Syntax Check (pull_request) Waiting to run
Validate Config / Deploy Script Dry Run (pull_request) Waiting to run
Validate Config / Playbook Schema Validation (pull_request) Waiting to run
Validate Config / YAML Lint (pull_request) Waiting to run
Validate Config / JSON Validate (pull_request) Waiting to run
Validate Config / Python Syntax & Import Check (pull_request) Waiting to run
Validate Config / Python Test Suite (pull_request) Blocked by required conditions
Validate Config / Shell Script Lint (pull_request) Waiting to run
- Tests all configured providers
- Maintains health map in tmux-state.json
- Auto-switches profiles to working providers
- Supports --daemon and --status modes

Closes #509
2026-04-15 03:48:37 +00:00
26e39d8949 feat: add autonomous cron supervisor job (#513)
- Runs every 7 minutes
- Checks dev and timmy sessions
- Loads tmux-supervisor skill
- Telegram only on actionable events
- Silent when all agents busy
2026-04-15 03:33:43 +00:00
d120526244 fix: add python3 shebang to scripts/visual_pr_reviewer.py (#681) 2026-04-15 02:57:53 +00:00
8596ff761b fix: add python3 shebang to scripts/diagram_meaning_extractor.py (#681) 2026-04-15 02:57:40 +00:00
7553fd4f3e fix: add python3 shebang to scripts/captcha_bypass_handler.py (#681) 2026-04-15 02:57:25 +00:00
71082fe06f fix: add python3 shebang to bin/soul_eval_gate.py (#681) 2026-04-15 02:57:14 +00:00
6d678e938e fix: add python3 shebang to bin/nostr-agent-demo.py (#681) 2026-04-15 02:57:00 +00:00
ad751a6de6 docs: add pipeline scheduler README 2026-04-14 22:47:12 +00:00
130fa40f0c feat: add pipeline-scheduler cron job 2026-04-14 22:46:51 +00:00
82f9810081 feat: add nightly-pipeline-scheduler.sh 2026-04-14 22:46:38 +00:00
2548277137 cleanup test
Some checks failed
Architecture Lint / Lint Repository (push) Blocked by required conditions
Validate Config / Python Test Suite (push) Blocked by required conditions
Architecture Lint / Lint Repository (pull_request) Blocked by required conditions
Validate Config / Python Test Suite (pull_request) Blocked by required conditions
Architecture Lint / Linter Tests (push) Successful in 22s
Smoke Test / smoke (push) Failing after 21s
Validate Config / YAML Lint (push) Failing after 13s
Validate Config / JSON Validate (push) Successful in 14s
Validate Config / Python Syntax & Import Check (push) Failing after 1m9s
Validate Config / Shell Script Lint (push) Failing after 31s
Validate Config / Cron Syntax Check (push) Successful in 5s
Validate Config / Deploy Script Dry Run (push) Successful in 7s
Validate Config / Playbook Schema Validation (push) Successful in 16s
Architecture Lint / Linter Tests (pull_request) Successful in 14s
Smoke Test / smoke (pull_request) Failing after 13s
Validate Config / YAML Lint (pull_request) Failing after 12s
Validate Config / JSON Validate (pull_request) Successful in 13s
Validate Config / Python Syntax & Import Check (pull_request) Failing after 54s
Validate Config / Shell Script Lint (pull_request) Failing after 21s
Validate Config / Cron Syntax Check (pull_request) Successful in 5s
Validate Config / Deploy Script Dry Run (pull_request) Successful in 7s
Validate Config / Playbook Schema Validation (pull_request) Successful in 18s
PR Checklist / pr-checklist (pull_request) Failing after 3m27s
2026-04-14 22:39:03 +00:00
2b234fde79 test: verify API works
Some checks failed
Architecture Lint / Linter Tests (push) Has been cancelled
Architecture Lint / Lint Repository (push) Has been cancelled
Validate Config / Python Test Suite (push) Blocked by required conditions
Smoke Test / smoke (push) Failing after 12s
Validate Config / YAML Lint (push) Failing after 11s
Validate Config / JSON Validate (push) Successful in 11s
Validate Config / Python Syntax & Import Check (push) Failing after 47s
Validate Config / Shell Script Lint (push) Failing after 33s
Validate Config / Cron Syntax Check (push) Successful in 10s
Validate Config / Deploy Script Dry Run (push) Successful in 10s
Validate Config / Playbook Schema Validation (push) Successful in 14s
2026-04-14 22:39:02 +00:00
04cceccd01 feat: add rock scene generator (#607)
Some checks failed
Architecture Lint / Lint Repository (push) Has been cancelled
Architecture Lint / Linter Tests (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
2026-04-14 22:35:43 +00:00
1ad2f2b239 feat: 100 rock lyrics-to-scene sets (#607)
Some checks failed
Architecture Lint / Linter Tests (push) Has been cancelled
Architecture Lint / Lint Repository (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
2026-04-14 22:35:11 +00:00
04dbf772b1 feat: Visual Smoke Test for The Nexus #490 (#558)
Some checks failed
Architecture Lint / Linter Tests (push) Successful in 15s
Smoke Test / smoke (push) Failing after 14s
Validate Config / YAML Lint (push) Failing after 13s
Validate Config / JSON Validate (push) Successful in 13s
Validate Config / Shell Script Lint (push) Failing after 40s
Validate Config / Python Syntax & Import Check (push) Failing after 58s
Validate Config / Python Test Suite (push) Has been skipped
Validate Config / Cron Syntax Check (push) Successful in 11s
Validate Config / Deploy Script Dry Run (push) Successful in 11s
Validate Config / Playbook Schema Validation (push) Successful in 18s
Architecture Lint / Lint Repository (push) Failing after 13s
Architecture Lint / Linter Tests (pull_request) Successful in 26s
Smoke Test / smoke (pull_request) Failing after 17s
Validate Config / YAML Lint (pull_request) Failing after 12s
Validate Config / JSON Validate (pull_request) Successful in 12s
PR Checklist / pr-checklist (pull_request) Failing after 3m36s
Validate Config / Shell Script Lint (pull_request) Failing after 40s
Validate Config / Python Syntax & Import Check (pull_request) Failing after 1m4s
Validate Config / Python Test Suite (pull_request) Has been skipped
Validate Config / Cron Syntax Check (pull_request) Successful in 9s
Validate Config / Deploy Script Dry Run (pull_request) Successful in 9s
Validate Config / Playbook Schema Validation (pull_request) Successful in 20s
Architecture Lint / Lint Repository (pull_request) Failing after 16s
Merge PR #558
2026-04-14 22:18:23 +00:00
697a273f0f fix: a11y R4 - <time> elements for relative timestamps (closes #554) (#569)
Some checks failed
Architecture Lint / Linter Tests (push) Successful in 10s
Smoke Test / smoke (push) Failing after 7s
Validate Config / YAML Lint (push) Failing after 7s
Validate Config / JSON Validate (push) Successful in 7s
Architecture Lint / Lint Repository (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Merge PR #569

Co-authored-by: Timmy Time <timmy@alexanderwhitestone.ai>
Co-committed-by: Timmy Time <timmy@alexanderwhitestone.ai>
2026-04-14 22:17:39 +00:00
9651a56308 feat: Glitch Detector HTML Report with annotated screenshots #544 (#567)
Some checks failed
Architecture Lint / Lint Repository (push) Has been cancelled
Architecture Lint / Linter Tests (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Merge PR #567
2026-04-14 22:17:32 +00:00
a84e6b517f [a11y] Visual Accessibility Audit — Foundation Web (#492) (#556)
Some checks failed
Architecture Lint / Lint Repository (push) Has been cancelled
Architecture Lint / Linter Tests (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Merge PR #556
2026-04-14 22:17:17 +00:00
31313c421e feat: 3D World Glitch Detection in The Matrix (#491) (#535)
Some checks failed
Architecture Lint / Linter Tests (push) Has been cancelled
Architecture Lint / Lint Repository (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Merge PR #535
2026-04-14 22:17:06 +00:00
063572ed1e feat: Visual Accessibility Audit of Foundation Web #492 (#531)
Some checks failed
Architecture Lint / Lint Repository (push) Has been cancelled
Architecture Lint / Linter Tests (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Merge PR #531
2026-04-14 22:16:58 +00:00
46b4f8d000 feat: pane-watchdog — stuck pane detection + auto-restart (#515) (#526)
Some checks failed
Architecture Lint / Linter Tests (push) Has been cancelled
Architecture Lint / Lint Repository (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Validate Config / YAML Lint (push) Has been cancelled
Validate Config / JSON Validate (push) Has been cancelled
Merge PR #526
2026-04-14 22:16:52 +00:00
e091868fef feat: auto-commit-guard — 4-layer work preservation (#511) (#525)
Some checks failed
Architecture Lint / Lint Repository (push) Has been cancelled
Architecture Lint / Linter Tests (push) Has been cancelled
Smoke Test / smoke (push) Has been cancelled
Validate Config / Python Syntax & Import Check (push) Has been cancelled
Validate Config / Python Test Suite (push) Has been cancelled
Validate Config / Shell Script Lint (push) Has been cancelled
Validate Config / Cron Syntax Check (push) Has been cancelled
Validate Config / Deploy Script Dry Run (push) Has been cancelled
Validate Config / Playbook Schema Validation (push) Has been cancelled
Validate Config / YAML Lint (push) Has started running
Validate Config / JSON Validate (push) Has started running
Merge PR #525
2026-04-14 22:16:49 +00:00
34 changed files with 6887 additions and 34 deletions

View File

@@ -202,6 +202,19 @@ curl -s -X POST "{gitea_url}/api/v1/repos/{repo}/issues/{issue_num}/comments" \\
REVIEW CHECKLIST BEFORE YOU PUSH:
{review}
COMMIT DISCIPLINE (CRITICAL):
- Commit every 3-5 tool calls. Do NOT wait until the end.
- After every meaningful file change: git add -A && git commit -m "WIP: <what changed>"
- Before running any destructive command: commit current state first.
- If you are unsure whether to commit: commit. WIP commits are safe. Lost work is not.
- Never use --no-verify.
- The auto-commit-guard is your safety net, but do not rely on it. Commit proactively.
RECOVERY COMMANDS (if interrupted, another agent can resume):
git log --oneline -10 # see your WIP commits
git diff HEAD~1 # see what the last commit changed
git status # see uncommitted work
RULES:
- Do not skip hooks with --no-verify.
- Do not silently widen the scope.

View File

@@ -161,6 +161,14 @@ run_worker() {
CYCLE_END=$(date +%s)
CYCLE_DURATION=$((CYCLE_END - CYCLE_START))
# --- Mid-session auto-commit: commit before timeout if work is dirty ---
cd "$worktree" 2>/dev/null || true
# Ensure auto-commit-guard is running
if ! pgrep -f "auto-commit-guard.sh" >/dev/null 2>&1; then
log "Starting auto-commit-guard daemon"
nohup bash "$(dirname "$0")/auto-commit-guard.sh" 120 "$WORKTREE_BASE" >> "$LOG_DIR/auto-commit-guard.log" 2>&1 &
fi
# Salvage
cd "$worktree" 2>/dev/null || true
DIRTY=$(git status --porcelain 2>/dev/null | wc -l | tr -d ' ')

159
bin/auto-commit-guard.sh Normal file
View File

@@ -0,0 +1,159 @@
#!/usr/bin/env bash
# auto-commit-guard.sh — Background daemon that auto-commits uncommitted work
#
# Usage: auto-commit-guard.sh [interval_seconds] [worktree_base]
# auto-commit-guard.sh # defaults: 120s, ~/worktrees
# auto-commit-guard.sh 60 # check every 60s
# auto-commit-guard.sh 180 ~/my-worktrees
#
# Scans all git repos under the worktree base for uncommitted changes.
# If dirty for >= 1 check cycle, auto-commits with a WIP message.
# Pushes unpushed commits so work is always recoverable from the remote.
#
# Also scans /tmp for orphaned agent workdirs on startup.
set -uo pipefail
INTERVAL="${1:-120}"
WORKTREE_BASE="${2:-$HOME/worktrees}"
LOG_DIR="$HOME/.hermes/logs"
LOG="$LOG_DIR/auto-commit-guard.log"
PIDFILE="$LOG_DIR/auto-commit-guard.pid"
ORPHAN_SCAN_DONE="$LOG_DIR/.orphan-scan-done"
mkdir -p "$LOG_DIR"
# Single instance guard
if [ -f "$PIDFILE" ]; then
old_pid=$(cat "$PIDFILE")
if kill -0 "$old_pid" 2>/dev/null; then
echo "auto-commit-guard already running (PID $old_pid)" >&2
exit 0
fi
fi
echo $$ > "$PIDFILE"
trap 'rm -f "$PIDFILE"' EXIT
log() {
echo "[$(date '+%Y-%m-%d %H:%M:%S')] AUTO-COMMIT: $*" >> "$LOG"
}
# --- Orphaned workdir scan (runs once on startup) ---
scan_orphans() {
if [ -f "$ORPHAN_SCAN_DONE" ]; then
return 0
fi
log "Scanning /tmp for orphaned agent workdirs..."
local found=0
local rescued=0
for dir in /tmp/*-work-* /tmp/timmy-burn-* /tmp/tc-burn; do
[ -d "$dir" ] || continue
[ -d "$dir/.git" ] || continue
found=$((found + 1))
cd "$dir" 2>/dev/null || continue
local dirty
dirty=$(git status --porcelain 2>/dev/null | wc -l | tr -d " ")
if [ "${dirty:-0}" -gt 0 ]; then
local branch
branch=$(git branch --show-current 2>/dev/null || echo "orphan")
git add -A 2>/dev/null
if git commit -m "WIP: orphan rescue — $dirty file(s) auto-committed on $(date -u +%Y-%m-%dT%H:%M:%SZ)
Orphaned workdir detected at $dir.
Branch: $branch
Rescued by auto-commit-guard on startup." 2>/dev/null; then
rescued=$((rescued + 1))
log "RESCUED: $dir ($dirty files on branch $branch)"
# Try to push if remote exists
if git remote get-url origin >/dev/null 2>&1; then
git push -u origin "$branch" 2>/dev/null && log "PUSHED orphan rescue: $dir$branch" || log "PUSH FAILED orphan rescue: $dir (no remote access)"
fi
fi
fi
done
log "Orphan scan complete: $found workdirs checked, $rescued rescued"
touch "$ORPHAN_SCAN_DONE"
}
# --- Main guard loop ---
guard_cycle() {
local committed=0
local scanned=0
# Scan worktree base
if [ -d "$WORKTREE_BASE" ]; then
for dir in "$WORKTREE_BASE"/*/; do
[ -d "$dir" ] || continue
[ -d "$dir/.git" ] || continue
scanned=$((scanned + 1))
cd "$dir" 2>/dev/null || continue
local dirty
dirty=$(git status --porcelain 2>/dev/null | wc -l | tr -d " ")
[ "${dirty:-0}" -eq 0 ] && continue
local branch
branch=$(git branch --show-current 2>/dev/null || echo "detached")
git add -A 2>/dev/null
if git commit -m "WIP: auto-commit — $dirty file(s) on $branch
Automated commit by auto-commit-guard at $(date -u +%Y-%m-%dT%H:%M:%SZ).
Work preserved to prevent loss on crash." 2>/dev/null; then
committed=$((committed + 1))
log "COMMITTED: $dir ($dirty files, branch $branch)"
# Push to preserve remotely
if git remote get-url origin >/dev/null 2>&1; then
git push -u origin "$branch" 2>/dev/null && log "PUSHED: $dir$branch" || log "PUSH FAILED: $dir (will retry next cycle)"
fi
fi
done
fi
# Also scan /tmp for agent workdirs
for dir in /tmp/*-work-*; do
[ -d "$dir" ] || continue
[ -d "$dir/.git" ] || continue
scanned=$((scanned + 1))
cd "$dir" 2>/dev/null || continue
local dirty
dirty=$(git status --porcelain 2>/dev/null | wc -l | tr -d " ")
[ "${dirty:-0}" -eq 0 ] && continue
local branch
branch=$(git branch --show-current 2>/dev/null || echo "detached")
git add -A 2>/dev/null
if git commit -m "WIP: auto-commit — $dirty file(s) on $branch
Automated commit by auto-commit-guard at $(date -u +%Y-%m-%dT%H:%M:%SZ).
Agent workdir preserved to prevent loss." 2>/dev/null; then
committed=$((committed + 1))
log "COMMITTED: $dir ($dirty files, branch $branch)"
if git remote get-url origin >/dev/null 2>&1; then
git push -u origin "$branch" 2>/dev/null && log "PUSHED: $dir$branch" || log "PUSH FAILED: $dir (will retry next cycle)"
fi
fi
done
[ "$committed" -gt 0 ] && log "Cycle done: $scanned scanned, $committed committed"
}
# --- Entry point ---
log "Starting auto-commit-guard (interval=${INTERVAL}s, worktree=${WORKTREE_BASE})"
scan_orphans
while true; do
guard_cycle
sleep "$INTERVAL"
done

439
bin/glitch_patterns.py Normal file
View File

@@ -0,0 +1,439 @@
"""
Glitch pattern definitions for 3D world anomaly detection.
Defines known visual artifact categories commonly found in 3D web worlds,
particularly The Matrix environments. Each pattern includes detection
heuristics and severity ratings.
"""
from dataclasses import dataclass, field
from enum import Enum
from typing import Optional
class GlitchSeverity(Enum):
CRITICAL = "critical"
HIGH = "high"
MEDIUM = "medium"
LOW = "low"
INFO = "info"
class GlitchCategory(Enum):
FLOATING_ASSETS = "floating_assets"
Z_FIGHTING = "z_fighting"
MISSING_TEXTURES = "missing_textures"
CLIPPING = "clipping"
BROKEN_NORMALS = "broken_normals"
SHADOW_ARTIFACTS = "shadow_artifacts"
LIGHTMAP_ERRORS = "lightmap_errors"
LOD_POPPING = "lod_popping"
WATER_REFLECTION = "water_reflection"
SKYBOX_SEAM = "skybox_seam"
# Three.js-specific categories (ref: timmy-config#543)
SHADER_FAILURE = "shader_failure"
TEXTURE_PLACEHOLDER = "texture_placeholder"
UV_MAPPING_ERROR = "uv_mapping_error"
FRUSTUM_CULLING = "frustum_culling"
SHADOW_MAP_ARTIFACT = "shadow_map_artifact"
BLOOM_OVERFLOW = "bloom_overflow"
@dataclass
class GlitchPattern:
"""Definition of a known glitch pattern with detection parameters."""
category: GlitchCategory
name: str
description: str
severity: GlitchSeverity
detection_prompts: list[str]
visual_indicators: list[str]
confidence_threshold: float = 0.6
def to_dict(self) -> dict:
return {
"category": self.category.value,
"name": self.name,
"description": self.description,
"severity": self.severity.value,
"detection_prompts": self.detection_prompts,
"visual_indicators": self.visual_indicators,
"confidence_threshold": self.confidence_threshold,
}
# Known glitch patterns for Matrix 3D world scanning
MATRIX_GLITCH_PATTERNS: list[GlitchPattern] = [
GlitchPattern(
category=GlitchCategory.FLOATING_ASSETS,
name="Floating Object",
description="Object not properly grounded or anchored to the scene geometry. "
"Common in procedurally placed assets or after physics desync.",
severity=GlitchSeverity.HIGH,
detection_prompts=[
"Identify any objects that appear to float above the ground without support.",
"Look for furniture, props, or geometry suspended in mid-air with no visible attachment.",
"Check for objects whose shadows do not align with the surface below them.",
],
visual_indicators=[
"gap between object base and surface",
"shadow detached from object",
"object hovering with no structural support",
],
confidence_threshold=0.65,
),
GlitchPattern(
category=GlitchCategory.Z_FIGHTING,
name="Z-Fighting Flicker",
description="Two coplanar surfaces competing for depth priority, causing "
"visible flickering or shimmering textures.",
severity=GlitchSeverity.MEDIUM,
detection_prompts=[
"Look for surfaces that appear to shimmer, flicker, or show mixed textures.",
"Identify areas where two textures seem to overlap and compete for visibility.",
"Check walls, floors, or objects for surface noise or pattern interference.",
],
visual_indicators=[
"shimmering surface",
"texture flicker between two patterns",
"noisy flat surfaces",
"moire-like patterns on planar geometry",
],
confidence_threshold=0.55,
),
GlitchPattern(
category=GlitchCategory.MISSING_TEXTURES,
name="Missing or Placeholder Texture",
description="A surface rendered with a fallback checkerboard, solid magenta, "
"or the default engine placeholder texture.",
severity=GlitchSeverity.CRITICAL,
detection_prompts=[
"Look for bright magenta, checkerboard, or solid-color surfaces that look out of place.",
"Identify any surfaces that appear as flat untextured colors inconsistent with the scene.",
"Check for black, white, or magenta patches where detailed textures should be.",
],
visual_indicators=[
"magenta/pink solid color surface",
"checkerboard pattern",
"flat single-color geometry",
"UV-debug texture visible",
],
confidence_threshold=0.7,
),
GlitchPattern(
category=GlitchCategory.CLIPPING,
name="Geometry Clipping",
description="Objects passing through each other or intersecting in physically "
"impossible ways due to collision mesh errors.",
severity=GlitchSeverity.HIGH,
detection_prompts=[
"Look for objects that visibly pass through other objects (walls, floors, furniture).",
"Identify characters or props embedded inside geometry where they should not be.",
"Check for intersecting meshes where solid objects overlap unnaturally.",
],
visual_indicators=[
"object passing through wall or floor",
"embedded geometry",
"overlapping solid meshes",
"character limb inside furniture",
],
confidence_threshold=0.6,
),
GlitchPattern(
category=GlitchCategory.BROKEN_NORMALS,
name="Broken Surface Normals",
description="Inverted or incorrect surface normals causing faces to appear "
"inside-out, invisible from certain angles, or lit incorrectly.",
severity=GlitchSeverity.MEDIUM,
detection_prompts=[
"Look for surfaces that appear dark or black on one side while lit on the other.",
"Identify objects that seem to vanish when viewed from certain angles.",
"Check for inverted shading where lit areas should be in shadow.",
],
visual_indicators=[
"dark/unlit face on otherwise lit model",
"invisible surface from one direction",
"inverted shadow gradient",
"inside-out appearance",
],
confidence_threshold=0.5,
),
GlitchPattern(
category=GlitchCategory.SHADOW_ARTIFACTS,
name="Shadow Artifact",
description="Broken, detached, or incorrectly rendered shadows that do not "
"match the casting geometry or scene lighting.",
severity=GlitchSeverity.LOW,
detection_prompts=[
"Look for shadows that do not match the shape of nearby objects.",
"Identify shadow acne: banding or striped patterns on surfaces.",
"Check for floating shadows detached from any visible caster.",
],
visual_indicators=[
"shadow shape mismatch",
"shadow acne bands",
"detached floating shadow",
"Peter Panning (shadow offset from base)",
],
confidence_threshold=0.5,
),
GlitchPattern(
category=GlitchCategory.LOD_POPPING,
name="LOD Transition Pop",
description="Visible pop-in when level-of-detail models switch abruptly, "
"causing geometry or textures to change suddenly.",
severity=GlitchSeverity.LOW,
detection_prompts=[
"Look for areas where mesh detail changes abruptly at visible boundaries.",
"Identify objects that appear to morph or shift geometry suddenly.",
"Check for texture resolution changes that create visible seams.",
],
visual_indicators=[
"visible mesh simplification boundary",
"texture resolution jump",
"geometry pop-in artifacts",
],
confidence_threshold=0.45,
),
GlitchPattern(
category=GlitchCategory.LIGHTMAP_ERRORS,
name="Lightmap Baking Error",
description="Incorrect or missing baked lighting causing dark spots, light "
"leaks, or mismatched illumination on static geometry.",
severity=GlitchSeverity.MEDIUM,
detection_prompts=[
"Look for unusually dark patches on walls or ceilings that should be lit.",
"Identify bright light leaks through solid geometry seams.",
"Check for mismatched lighting between adjacent surfaces.",
],
visual_indicators=[
"dark splotch on lit surface",
"bright line at geometry seam",
"lighting discontinuity between adjacent faces",
],
confidence_threshold=0.5,
),
GlitchPattern(
category=GlitchCategory.WATER_REFLECTION,
name="Water/Reflection Error",
description="Incorrect reflections, missing water surfaces, or broken "
"reflection probe assignments.",
severity=GlitchSeverity.MEDIUM,
detection_prompts=[
"Look for reflections that do not match the surrounding environment.",
"Identify water surfaces that appear solid or incorrectly rendered.",
"Check for mirror surfaces showing wrong scene geometry.",
],
visual_indicators=[
"reflection mismatch",
"solid water surface",
"incorrect environment map",
],
confidence_threshold=0.5,
),
GlitchPattern(
category=GlitchCategory.SKYBOX_SEAM,
name="Skybox Seam",
description="Visible seams or color mismatches at the edges of skybox cubemap faces.",
severity=GlitchSeverity.LOW,
detection_prompts=[
"Look at the edges of the sky for visible seams or color shifts.",
"Identify discontinuities where skybox faces meet.",
"Check for texture stretching at skybox corners.",
],
visual_indicators=[
"visible line in sky",
"color discontinuity at sky edge",
"sky texture seam",
],
confidence_threshold=0.45,
),
# --- Three.js-Specific Glitch Patterns (ref: timmy-config#543) ---
GlitchPattern(
category=GlitchCategory.SHADER_FAILURE,
name="Shader Compilation Failure",
description="Three.js shader failed to compile, rendering the material as solid black. "
"Common when custom ShaderMaterial has syntax errors or missing uniforms.",
severity=GlitchSeverity.CRITICAL,
detection_prompts=[
"Look for objects or surfaces rendered as pure black (#000000) that should have visible textures or materials.",
"Identify geometry that appears completely dark while surrounding objects are normally lit.",
"Check for objects where the material seems to 'absorb all light' — flat black with no shading gradient.",
],
visual_indicators=[
"solid black object with no shading",
"geometry rendered as silhouette",
"material appears to absorb light entirely",
"black patch inconsistent with scene lighting",
],
confidence_threshold=0.7,
),
GlitchPattern(
category=GlitchCategory.TEXTURE_PLACEHOLDER,
name="Three.js Texture Not Loaded",
description="Three.js failed to load the texture asset, rendering a 1x1 white pixel "
"stretched across the entire surface. Distinguished from missing-texture by "
"the uniform white/grey appearance rather than magenta.",
severity=GlitchSeverity.CRITICAL,
detection_prompts=[
"Look for surfaces that are uniformly white or light grey with no texture detail, even on large geometry.",
"Identify objects where the texture appears as a single solid color stretched across complex UVs.",
"Check for surfaces that look 'blank' or 'unloaded' — flat white/grey where detail should exist.",
],
visual_indicators=[
"uniform white or light grey surface",
"no texture detail on large geometry",
"stretched single-color appearance",
"1x1 pixel placeholder stretched to fill UV space",
],
confidence_threshold=0.65,
),
GlitchPattern(
category=GlitchCategory.UV_MAPPING_ERROR,
name="BufferGeometry UV Mapping Error",
description="Three.js BufferGeometry has incorrect UV coordinates, causing textures to "
"appear stretched, compressed, or mapped to the wrong faces.",
severity=GlitchSeverity.HIGH,
detection_prompts=[
"Look for textures that appear dramatically stretched in one direction on specific faces.",
"Identify surfaces where the texture pattern is distorted but other nearby surfaces look correct.",
"Check for faces where the texture seems 'smeared' or mapped with incorrect aspect ratio.",
],
visual_indicators=[
"texture stretching on specific faces",
"distorted pattern on geometry",
"smeared texture appearance",
"aspect ratio mismatch between texture and surface",
],
confidence_threshold=0.6,
),
GlitchPattern(
category=GlitchCategory.FRUSTUM_CULLING,
name="Frustum Culling Artifact",
description="Three.js frustum culling incorrectly marks objects as outside the camera "
"frustum, causing them to pop in/out of existence at screen edges.",
severity=GlitchSeverity.MEDIUM,
detection_prompts=[
"Look for objects that are partially visible at the edge of the frame — half-rendered or cut off unnaturally.",
"Identify geometry that seems to 'pop' into existence as the view angle changes.",
"Check screen edges for objects that appear suddenly rather than smoothly entering the viewport.",
],
visual_indicators=[
"half-visible object at screen edge",
"object popping into frame",
"abrupt appearance of geometry",
"bounding box visible but mesh missing",
],
confidence_threshold=0.55,
),
GlitchPattern(
category=GlitchCategory.SHADOW_MAP_ARTIFACT,
name="Shadow Map Resolution Artifact",
description="Three.js shadow map has insufficient resolution, causing pixelated, "
"blocky shadows with visible texel edges instead of smooth shadow gradients.",
severity=GlitchSeverity.MEDIUM,
detection_prompts=[
"Look for shadows with visible blocky or pixelated edges instead of smooth gradients.",
"Identify shadow maps where individual texels (texture pixels) are clearly visible.",
"Check for shadows that appear as jagged stair-stepped patterns rather than soft edges.",
],
visual_indicators=[
"blocky shadow edges",
"visible texel grid in shadows",
"stair-stepped shadow boundary",
"pixelated shadow gradient",
],
confidence_threshold=0.55,
),
GlitchPattern(
category=GlitchCategory.BLOOM_OVERFLOW,
name="Post-Processing Bloom Overflow",
description="Three.js UnrealBloomPass or similar post-processing bloom effect is too "
"intense, causing bright areas to bleed glow into surrounding geometry.",
severity=GlitchSeverity.LOW,
detection_prompts=[
"Look for bright areas that have an unusually large, soft glow bleeding into adjacent surfaces.",
"Identify scenes where light sources appear to have a 'halo' that extends beyond physical plausibility.",
"Check for bright objects whose glow color bleeds onto nearby unrelated geometry.",
],
visual_indicators=[
"excessive glow bleeding from bright surfaces",
"halo around light sources",
"bloom color tinting adjacent geometry",
"glow bleeding beyond object boundaries",
],
confidence_threshold=0.5,
),
]
def get_patterns_by_severity(min_severity: GlitchSeverity) -> list[GlitchPattern]:
"""Return patterns at or above the given severity level."""
severity_order = [
GlitchSeverity.INFO,
GlitchSeverity.LOW,
GlitchSeverity.MEDIUM,
GlitchSeverity.HIGH,
GlitchSeverity.CRITICAL,
]
min_idx = severity_order.index(min_severity)
return [p for p in MATRIX_GLITCH_PATTERNS if severity_order.index(p.severity) >= min_idx]
def get_pattern_by_category(category: GlitchCategory) -> Optional[GlitchPattern]:
"""Return the pattern definition for a specific category."""
for p in MATRIX_GLITCH_PATTERNS:
if p.category == category:
return p
return None
def build_vision_prompt(patterns: list[GlitchPattern] | None = None) -> str:
"""Build a composite vision analysis prompt from pattern definitions."""
if patterns is None:
patterns = MATRIX_GLITCH_PATTERNS
sections = []
for p in patterns:
prompt_text = " ".join(p.detection_prompts)
indicators = ", ".join(p.visual_indicators)
sections.append(
f"[{p.category.value.upper()}] {p.name} (severity: {p.severity.value})\n"
f" {p.description}\n"
f" Look for: {prompt_text}\n"
f" Visual indicators: {indicators}"
)
return (
"Analyze this 3D world screenshot for visual glitches and artifacts. "
"For each detected issue, report the category, description of what you see, "
"approximate location in the image (x%, y%), and confidence (0.0-1.0).\n\n"
"Known glitch patterns to check:\n\n" + "\n\n".join(sections)
)
# Three.js-specific category set for filtering (ref: timmy-config#543)
THREEJS_CATEGORIES = {
GlitchCategory.SHADER_FAILURE,
GlitchCategory.TEXTURE_PLACEHOLDER,
GlitchCategory.UV_MAPPING_ERROR,
GlitchCategory.FRUSTUM_CULLING,
GlitchCategory.SHADOW_MAP_ARTIFACT,
GlitchCategory.BLOOM_OVERFLOW,
}
def get_threejs_patterns() -> list[GlitchPattern]:
"""Return only Three.js-specific glitch patterns."""
return [p for p in MATRIX_GLITCH_PATTERNS if p.category in THREEJS_CATEGORIES]
if __name__ == "__main__":
import json
print(f"Loaded {len(MATRIX_GLITCH_PATTERNS)} glitch patterns:\n")
for p in MATRIX_GLITCH_PATTERNS:
print(f" [{p.severity.value:8s}] {p.category.value}: {p.name}")
print(f"\nVision prompt preview:\n{build_vision_prompt()[:500]}...")

View File

@@ -0,0 +1,586 @@
#!/usr/bin/env python3
"""
Matrix 3D World Glitch Detector
Scans a 3D web world for visual artifacts using browser automation
and vision AI analysis. Produces structured glitch reports.
Usage:
python matrix_glitch_detector.py <url> [--angles 4] [--output report.json]
python matrix_glitch_detector.py --demo # Run with synthetic test data
Ref: timmy-config#491, timmy-config#543
"""
import argparse
import base64
import json
import os
import sys
import time
import uuid
from dataclasses import dataclass, field, asdict
from datetime import datetime, timezone
from pathlib import Path
from typing import Optional
# Add parent for glitch_patterns import
sys.path.insert(0, str(Path(__file__).resolve().parent))
from glitch_patterns import (
GlitchCategory,
GlitchPattern,
GlitchSeverity,
MATRIX_GLITCH_PATTERNS,
build_vision_prompt,
get_patterns_by_severity,
get_threejs_patterns,
)
@dataclass
class DetectedGlitch:
"""A single detected glitch with metadata."""
id: str
category: str
name: str
description: str
severity: str
confidence: float
location_x: Optional[float] = None # percentage across image
location_y: Optional[float] = None # percentage down image
screenshot_index: int = 0
screenshot_angle: str = "front"
timestamp: str = ""
def __post_init__(self):
if not self.timestamp:
self.timestamp = datetime.now(timezone.utc).isoformat()
@dataclass
class ScanResult:
"""Complete scan result for a 3D world URL."""
scan_id: str
url: str
timestamp: str
total_screenshots: int
angles_captured: list[str]
glitches: list[dict] = field(default_factory=list)
summary: dict = field(default_factory=dict)
metadata: dict = field(default_factory=dict)
def to_json(self, indent: int = 2) -> str:
return json.dumps(asdict(self), indent=indent)
def generate_scan_angles(num_angles: int) -> list[dict]:
"""Generate camera angle configurations for multi-angle scanning.
Returns a list of dicts with yaw/pitch/label for browser camera control.
"""
base_angles = [
{"yaw": 0, "pitch": 0, "label": "front"},
{"yaw": 90, "pitch": 0, "label": "right"},
{"yaw": 180, "pitch": 0, "label": "back"},
{"yaw": 270, "pitch": 0, "label": "left"},
{"yaw": 0, "pitch": -30, "label": "front_low"},
{"yaw": 45, "pitch": -15, "label": "front_right_low"},
{"yaw": 0, "pitch": 30, "label": "front_high"},
{"yaw": 45, "pitch": 0, "label": "front_right"},
]
if num_angles <= len(base_angles):
return base_angles[:num_angles]
return base_angles + [
{"yaw": i * (360 // num_angles), "pitch": 0, "label": f"angle_{i}"}
for i in range(len(base_angles), num_angles)
]
def capture_screenshots(url: str, angles: list[dict], output_dir: Path) -> list[Path]:
"""Capture screenshots of a 3D web world from multiple angles.
Uses browser_vision tool when available; falls back to placeholder generation
for testing and environments without browser access.
"""
output_dir.mkdir(parents=True, exist_ok=True)
screenshots = []
for i, angle in enumerate(angles):
filename = output_dir / f"screenshot_{i:03d}_{angle['label']}.png"
# Attempt browser-based capture via browser_vision
try:
result = _browser_capture(url, angle, filename)
if result:
screenshots.append(filename)
continue
except Exception:
pass
# Generate placeholder screenshot for offline/test scenarios
_generate_placeholder_screenshot(filename, angle)
screenshots.append(filename)
return screenshots
def _browser_capture(url: str, angle: dict, output_path: Path) -> bool:
"""Capture a screenshot via browser automation.
This is a stub that delegates to the browser_vision tool when run
in an environment that provides it. In CI or offline mode, returns False.
"""
# Check if browser_vision is available via environment
bv_script = os.environ.get("BROWSER_VISION_SCRIPT")
if bv_script and Path(bv_script).exists():
import subprocess
cmd = [
sys.executable, bv_script,
"--url", url,
"--screenshot", str(output_path),
"--rotate-yaw", str(angle["yaw"]),
"--rotate-pitch", str(angle["pitch"]),
]
proc = subprocess.run(cmd, capture_output=True, text=True, timeout=30)
return proc.returncode == 0 and output_path.exists()
return False
def _generate_placeholder_screenshot(path: Path, angle: dict):
"""Generate a minimal 1x1 PNG as a placeholder for testing."""
# Minimal valid PNG (1x1 transparent pixel)
png_data = (
b"\x89PNG\r\n\x1a\n"
b"\x00\x00\x00\rIHDR\x00\x00\x00\x01\x00\x00\x00\x01"
b"\x08\x06\x00\x00\x00\x1f\x15\xc4\x89"
b"\x00\x00\x00\nIDATx\x9cc\x00\x01\x00\x00\x05\x00\x01"
b"\r\n\xb4\x00\x00\x00\x00IEND\xaeB`\x82"
)
path.write_bytes(png_data)
def analyze_with_vision(
screenshot_paths: list[Path],
angles: list[dict],
patterns: list[GlitchPattern] | None = None,
) -> list[DetectedGlitch]:
"""Send screenshots to vision AI for glitch analysis.
In environments with a vision model available, sends each screenshot
with the composite detection prompt. Otherwise returns simulated results.
"""
if patterns is None:
patterns = MATRIX_GLITCH_PATTERNS
prompt = build_vision_prompt(patterns)
glitches = []
for i, (path, angle) in enumerate(zip(screenshot_paths, angles)):
# Attempt vision analysis
detected = _vision_analyze_image(path, prompt, i, angle["label"])
glitches.extend(detected)
return glitches
def _vision_analyze_image(
image_path: Path,
prompt: str,
screenshot_index: int,
angle_label: str,
) -> list[DetectedGlitch]:
"""Analyze a single screenshot with vision AI.
Uses the vision_analyze tool when available; returns empty list otherwise.
"""
# Check for vision API configuration
api_key = os.environ.get("VISION_API_KEY") or os.environ.get("OPENAI_API_KEY")
api_base = os.environ.get("VISION_API_BASE", "https://api.openai.com/v1")
if api_key:
try:
return _call_vision_api(
image_path, prompt, screenshot_index, angle_label, api_key, api_base
)
except Exception as e:
print(f" [!] Vision API error for {image_path.name}: {e}", file=sys.stderr)
# No vision backend available
return []
def _call_vision_api(
image_path: Path,
prompt: str,
screenshot_index: int,
angle_label: str,
api_key: str,
api_base: str,
) -> list[DetectedGlitch]:
"""Call a vision API (OpenAI-compatible) for image analysis."""
import urllib.request
import urllib.error
image_data = base64.b64encode(image_path.read_bytes()).decode()
payload = json.dumps({
"model": os.environ.get("VISION_MODEL", "gpt-4o"),
"messages": [
{
"role": "user",
"content": [
{"type": "text", "text": prompt},
{
"type": "image_url",
"image_url": {
"url": f"data:image/png;base64,{image_data}",
"detail": "high",
},
},
],
}
],
"max_tokens": 4096,
}).encode()
req = urllib.request.Request(
f"{api_base}/chat/completions",
data=payload,
headers={
"Content-Type": "application/json",
"Authorization": f"Bearer {api_key}",
},
)
with urllib.request.urlopen(req, timeout=60) as resp:
result = json.loads(resp.read())
content = result["choices"][0]["message"]["content"]
return _parse_vision_response(content, screenshot_index, angle_label)
def _add_glitch_from_dict(
item: dict,
glitches: list[DetectedGlitch],
screenshot_index: int,
angle_label: str,
):
"""Convert a dict from vision API response into a DetectedGlitch."""
cat = item.get("category", item.get("type", "unknown"))
conf = float(item.get("confidence", item.get("score", 0.5)))
glitch = DetectedGlitch(
id=str(uuid.uuid4())[:8],
category=cat,
name=item.get("name", item.get("label", cat)),
description=item.get("description", item.get("detail", "")),
severity=item.get("severity", _infer_severity(cat, conf)),
confidence=conf,
location_x=item.get("location_x", item.get("x")),
location_y=item.get("location_y", item.get("y")),
screenshot_index=screenshot_index,
screenshot_angle=angle_label,
)
glitches.append(glitch)
def _parse_vision_response(
text: str, screenshot_index: int, angle_label: str
) -> list[DetectedGlitch]:
"""Parse vision AI response into structured glitch detections."""
glitches = []
# Try to extract JSON from the response
json_blocks = []
in_json = False
json_buf = []
for line in text.split("\n"):
stripped = line.strip()
if stripped.startswith("```"):
if in_json and json_buf:
try:
json_blocks.append(json.loads("\n".join(json_buf)))
except json.JSONDecodeError:
pass
json_buf = []
in_json = not in_json
continue
if in_json:
json_buf.append(line)
# Flush any remaining buffer
if in_json and json_buf:
try:
json_blocks.append(json.loads("\n".join(json_buf)))
except json.JSONDecodeError:
pass
# Also try parsing the entire response as JSON
try:
parsed = json.loads(text)
if isinstance(parsed, list):
json_blocks.extend(parsed)
elif isinstance(parsed, dict):
if "glitches" in parsed:
json_blocks.extend(parsed["glitches"])
elif "detections" in parsed:
json_blocks.extend(parsed["detections"])
else:
json_blocks.append(parsed)
except json.JSONDecodeError:
pass
for item in json_blocks:
# Flatten arrays of detections
if isinstance(item, list):
for sub in item:
if isinstance(sub, dict):
_add_glitch_from_dict(sub, glitches, screenshot_index, angle_label)
elif isinstance(item, dict):
_add_glitch_from_dict(item, glitches, screenshot_index, angle_label)
return glitches
def _infer_severity(category: str, confidence: float) -> str:
"""Infer severity from category and confidence when not provided."""
critical_cats = {"missing_textures", "clipping", "shader_failure", "texture_placeholder"}
high_cats = {"floating_assets", "broken_normals", "uv_mapping_error"}
medium_cats = {"frustum_culling", "shadow_map_artifact"}
cat_lower = category.lower()
if any(c in cat_lower for c in critical_cats):
return "critical" if confidence > 0.7 else "high"
if any(c in cat_lower for c in high_cats):
return "high" if confidence > 0.7 else "medium"
if any(c in cat_lower for c in medium_cats):
return "medium" if confidence > 0.6 else "low"
return "medium" if confidence > 0.6 else "low"
def build_report(
url: str,
angles: list[dict],
screenshots: list[Path],
glitches: list[DetectedGlitch],
) -> ScanResult:
"""Build the final structured scan report."""
severity_counts = {}
category_counts = {}
for g in glitches:
severity_counts[g.severity] = severity_counts.get(g.severity, 0) + 1
category_counts[g.category] = category_counts.get(g.category, 0) + 1
report = ScanResult(
scan_id=str(uuid.uuid4()),
url=url,
timestamp=datetime.now(timezone.utc).isoformat(),
total_screenshots=len(screenshots),
angles_captured=[a["label"] for a in angles],
glitches=[asdict(g) for g in glitches],
summary={
"total_glitches": len(glitches),
"by_severity": severity_counts,
"by_category": category_counts,
"highest_severity": max(severity_counts.keys(), default="none"),
"clean_screenshots": sum(
1
for i in range(len(screenshots))
if not any(g.screenshot_index == i for g in glitches)
),
},
metadata={
"detector_version": "0.2.0",
"pattern_count": len(MATRIX_GLITCH_PATTERNS),
"reference": "timmy-config#491, timmy-config#543",
},
)
return report
def run_demo(output_path: Optional[Path] = None) -> ScanResult:
"""Run a demonstration scan with simulated detections."""
print("[*] Running Matrix glitch detection demo...")
url = "https://matrix.example.com/world/alpha"
angles = generate_scan_angles(4)
screenshots_dir = Path("/tmp/matrix_glitch_screenshots")
print(f"[*] Capturing {len(angles)} screenshots from: {url}")
screenshots = capture_screenshots(url, angles, screenshots_dir)
print(f"[*] Captured {len(screenshots)} screenshots")
# Simulate detections for demo
demo_glitches = [
DetectedGlitch(
id=str(uuid.uuid4())[:8],
category="floating_assets",
name="Floating Chair",
description="Office chair floating 0.3m above floor in sector 7",
severity="high",
confidence=0.87,
location_x=35.2,
location_y=62.1,
screenshot_index=0,
screenshot_angle="front",
),
DetectedGlitch(
id=str(uuid.uuid4())[:8],
category="z_fighting",
name="Wall Texture Flicker",
description="Z-fighting between wall panel and decorative overlay",
severity="medium",
confidence=0.72,
location_x=58.0,
location_y=40.5,
screenshot_index=1,
screenshot_angle="right",
),
DetectedGlitch(
id=str(uuid.uuid4())[:8],
category="missing_textures",
name="Placeholder Texture",
description="Bright magenta surface on door frame — missing asset reference",
severity="critical",
confidence=0.95,
location_x=72.3,
location_y=28.8,
screenshot_index=2,
screenshot_angle="back",
),
DetectedGlitch(
id=str(uuid.uuid4())[:8],
category="clipping",
name="Desk Through Wall",
description="Desk corner clipping through adjacent wall geometry",
severity="high",
confidence=0.81,
location_x=15.0,
location_y=55.0,
screenshot_index=3,
screenshot_angle="left",
),
DetectedGlitch(
id=str(uuid.uuid4())[:8],
category="shader_failure",
name="Black Material on Portal Frame",
description="Portal frame rendered as solid black — shader compilation failed (missing uniform u_time)",
severity="critical",
confidence=0.91,
location_x=45.0,
location_y=30.0,
screenshot_index=0,
screenshot_angle="front",
),
DetectedGlitch(
id=str(uuid.uuid4())[:8],
category="shadow_map_artifact",
name="Pixelated Character Shadow",
description="Character shadow shows visible texel grid — shadow map resolution too low (512x512)",
severity="medium",
confidence=0.78,
location_x=52.0,
location_y=75.0,
screenshot_index=1,
screenshot_angle="right",
),
]
print(f"[*] Detected {len(demo_glitches)} glitches")
report = build_report(url, angles, screenshots, demo_glitches)
if output_path:
output_path.write_text(report.to_json())
print(f"[*] Report saved to: {output_path}")
return report
def main():
parser = argparse.ArgumentParser(
description="Matrix 3D World Glitch Detector — scan for visual artifacts",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
%(prog)s https://matrix.example.com/world/alpha
%(prog)s https://matrix.example.com/world/alpha --angles 8 --output report.json
%(prog)s --demo
""",
)
parser.add_argument("url", nargs="?", help="URL of the 3D world to scan")
parser.add_argument(
"--angles", type=int, default=4, help="Number of camera angles to capture (default: 4)"
)
parser.add_argument("--output", "-o", type=str, help="Output file path for JSON report")
parser.add_argument("--demo", action="store_true", help="Run demo with simulated data")
parser.add_argument(
"--min-severity",
choices=["info", "low", "medium", "high", "critical"],
default="info",
help="Minimum severity to include in report",
)
parser.add_argument("--verbose", "-v", action="store_true", help="Verbose output")
parser.add_argument(
"--threejs",
action="store_true",
help="Focus on Three.js-specific glitch patterns only (shader, texture, UV, culling, shadow, bloom)",
)
args = parser.parse_args()
if args.demo:
output = Path(args.output) if args.output else Path("glitch_report_demo.json")
report = run_demo(output)
print(f"\n=== Scan Summary ===")
print(f"URL: {report.url}")
print(f"Screenshots: {report.total_screenshots}")
print(f"Glitches found: {report.summary['total_glitches']}")
print(f"By severity: {report.summary['by_severity']}")
return
if not args.url:
parser.error("URL required (or use --demo)")
scan_id = str(uuid.uuid4())[:8]
print(f"[*] Matrix Glitch Detector — Scan {scan_id}")
print(f"[*] Target: {args.url}")
# Generate camera angles
angles = generate_scan_angles(args.angles)
print(f"[*] Capturing {len(angles)} screenshots...")
# Capture screenshots
screenshots_dir = Path(f"/tmp/matrix_glitch_{scan_id}")
screenshots = capture_screenshots(args.url, angles, screenshots_dir)
print(f"[*] Captured {len(screenshots)} screenshots")
# Filter patterns by severity and type
min_sev = GlitchSeverity(args.min_severity)
patterns = get_patterns_by_severity(min_sev)
if args.threejs:
threejs_patterns = get_threejs_patterns()
patterns = [p for p in patterns if p in threejs_patterns]
print(f"[*] Three.js-focused mode: {len(patterns)} patterns")
# Analyze with vision AI
print(f"[*] Analyzing with vision AI ({len(patterns)} patterns)...")
glitches = analyze_with_vision(screenshots, angles, patterns)
# Build and save report
report = build_report(args.url, angles, screenshots, glitches)
if args.output:
Path(args.output).write_text(report.to_json())
print(f"[*] Report saved: {args.output}")
else:
print(report.to_json())
print(f"\n[*] Done — {len(glitches)} glitches detected")
if __name__ == "__main__":
main()

View File

@@ -1,3 +1,4 @@
#!/usr/bin/env python3
"""
Full Nostr agent-to-agent communication demo - FINAL WORKING
"""

514
bin/pane-watchdog.sh Executable file
View File

@@ -0,0 +1,514 @@
#!/usr/bin/env bash
# pane-watchdog.sh — Detect stuck/dead tmux panes and auto-restart them
#
# Tracks output hash per pane across cycles. If a pane's captured output
# hasn't changed for STUCK_CYCLES consecutive checks, the pane is STUCK.
# Dead panes (PID gone) are also detected.
#
# On STUCK/DEAD:
# 1. Kill the pane
# 2. Attempt restart with --resume (session ID from manifest)
# 3. Fallback: fresh prompt with last known task from logs
#
# State file: ~/.hermes/pane-state.json
# Log: ~/.hermes/logs/pane-watchdog.log
#
# Usage:
# pane-watchdog.sh # One-shot check all sessions
# pane-watchdog.sh --daemon # Run every CHECK_INTERVAL seconds
# pane-watchdog.sh --status # Print current pane state
# pane-watchdog.sh --session NAME # Check only one session
#
# Issue: timmy-config #515
set -uo pipefail
export PATH="/opt/homebrew/bin:$HOME/.local/bin:$HOME/.hermes/bin:/usr/local/bin:$PATH"
# === CONFIG ===
STATE_FILE="${PANE_STATE_FILE:-$HOME/.hermes/pane-state.json}"
LOG_FILE="${PANE_WATCHDOG_LOG:-$HOME/.hermes/logs/pane-watchdog.log}"
CHECK_INTERVAL="${PANE_CHECK_INTERVAL:-120}" # seconds between cycles
STUCK_CYCLES=2 # unchanged cycles before STUCK
MAX_RESTART_ATTEMPTS=3 # per pane per hour
RESTART_COOLDOWN=3600 # seconds between escalation alerts
CAPTURE_LINES=40 # lines of output to hash
# Sessions to monitor (all if empty)
MONITOR_SESSIONS="${PANE_WATCHDOG_SESSIONS:-}"
mkdir -p "$(dirname "$STATE_FILE")" "$(dirname "$LOG_FILE")"
log() {
echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*" >> "$LOG_FILE"
}
# === HELPERS ===
# Capture last N lines of pane output and hash them
capture_pane_hash() {
local target="$1"
local output
output=$(tmux capture-pane -t "$target" -p -S "-${CAPTURE_LINES}" 2>/dev/null || echo "DEAD")
echo -n "$output" | shasum -a 256 | cut -d' ' -f1
}
# Check if pane PID is alive
pane_pid_alive() {
local target="$1"
local pid
pid=$(tmux list-panes -t "$target" -F '#{pane_pid}' 2>/dev/null | head -1 || echo "")
if [ -z "$pid" ]; then
return 1 # pane doesn't exist
fi
kill -0 "$pid" 2>/dev/null
}
# Get pane start command
pane_start_command() {
local target="$1"
tmux list-panes -t "$target" -F '#{pane_start_command}' 2>/dev/null | head -1 || echo "unknown"
}
# Get the pane's current running command (child process)
pane_current_command() {
local target="$1"
tmux list-panes -t "$target" -F '#{pane_current_command}' 2>/dev/null || echo "unknown"
}
# Only restart panes running hermes/agent commands (not zsh, python3 repls, etc.)
is_restartable() {
local cmd="$1"
case "$cmd" in
hermes|*hermes*|*agent*|*timmy*|*kimi*|*claude-loop*|*gemini-loop*)
return 0
;;
*)
return 1
;;
esac
}
# Get session ID from hermes manifest if available
get_hermes_session_id() {
local session_name="$1"
local manifest="$HOME/.hermes/sessions/${session_name}/manifest.json"
if [ -f "$manifest" ]; then
python3 -c "
import json, sys
try:
m = json.load(open('$manifest'))
print(m.get('session_id', m.get('id', '')))
except: pass
" 2>/dev/null || echo ""
else
echo ""
fi
}
# Get last task from pane logs
get_last_task() {
local session_name="$1"
local log_dir="$HOME/.hermes/logs"
# Find the most recent log for this session
local log_file
log_file=$(find "$log_dir" -name "*${session_name}*" -type f -mtime -1 2>/dev/null | sort -r | head -1)
if [ -n "$log_file" ] && [ -f "$log_file" ]; then
# Extract last user prompt or task description
grep -i "task:\|prompt:\|issue\|working on" "$log_file" 2>/dev/null | tail -1 | sed 's/.*[:>] *//' | head -c 200
fi
}
# Restart a pane with a fresh shell/command
restart_pane() {
local target="$1"
local session_name="${target%%:*}"
local session_id last_task cmd
log "RESTART: Attempting to restart $target"
# Kill existing pane
tmux kill-pane -t "$target" 2>/dev/null || true
sleep 1
# Try --resume with session ID
session_id=$(get_hermes_session_id "$session_name")
if [ -n "$session_id" ]; then
log "RESTART: Trying --resume with session $session_id"
tmux split-window -t "$session_name" -d \
"hermes chat --resume '$session_id' 2>&1 | tee -a '$HOME/.hermes/logs/${session_name}-restart.log'"
sleep 2
if pane_pid_alive "${session_name}:1" 2>/dev/null; then
log "RESTART: Success with --resume"
return 0
fi
fi
# Fallback: fresh prompt
last_task=$(get_last_task "$session_name")
if [ -n "$last_task" ]; then
log "RESTART: Fallback — fresh prompt with task: $last_task"
tmux split-window -t "$session_name" -d \
"echo 'Watchdog restart — last task: $last_task' && hermes chat 2>&1 | tee -a '$HOME/.hermes/logs/${session_name}-restart.log'"
else
log "RESTART: Fallback — fresh hermes chat"
tmux split-window -t "$session_name" -d \
"hermes chat 2>&1 | tee -a '$HOME/.hermes/logs/${session_name}-restart.log'"
fi
sleep 2
if pane_pid_alive "${session_name}:1" 2>/dev/null; then
log "RESTART: Fallback restart succeeded"
return 0
else
log "RESTART: FAILED to restart $target"
return 1
fi
}
# === STATE MANAGEMENT ===
read_state() {
if [ -f "$STATE_FILE" ]; then
cat "$STATE_FILE"
else
echo "{}"
fi
}
write_state() {
echo "$1" > "$STATE_FILE"
}
# Update state for a single pane and return JSON status
update_pane_state() {
local target="$1"
local hash="$2"
local is_alive="$3"
local now
now=$(date +%s)
python3 - "$STATE_FILE" "$target" "$hash" "$is_alive" "$now" "$STUCK_CYCLES" <<'PYEOF'
import json, sys, time
state_file = sys.argv[1]
target = sys.argv[2]
new_hash = sys.argv[3]
is_alive = sys.argv[4] == "true"
now = int(sys.argv[5])
stuck_cycles = int(sys.argv[6])
try:
with open(state_file) as f:
state = json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
state = {}
pane = state.get(target, {
"hash": "",
"same_count": 0,
"status": "UNKNOWN",
"last_change": 0,
"last_check": 0,
"restart_attempts": 0,
"last_restart": 0,
"current_command": "",
})
if not is_alive:
pane["status"] = "DEAD"
pane["same_count"] = 0
elif new_hash == pane.get("hash", ""):
pane["same_count"] = pane.get("same_count", 0) + 1
if pane["same_count"] >= stuck_cycles:
pane["status"] = "STUCK"
else:
pane["status"] = "STALE" if pane["same_count"] > 0 else "OK"
else:
pane["hash"] = new_hash
pane["same_count"] = 0
pane["status"] = "OK"
pane["last_change"] = now
pane["last_check"] = now
state[target] = pane
with open(state_file, "w") as f:
json.dump(state, f, indent=2)
print(json.dumps(pane))
PYEOF
}
# Reset restart attempt counter if cooldown expired
maybe_reset_restarts() {
local target="$1"
local now
now=$(date +%s)
python3 - "$STATE_FILE" "$target" "$now" "$RESTART_COOLDOWN" <<'PYEOF'
import json, sys
state_file = sys.argv[1]
target = sys.argv[2]
now = int(sys.argv[3])
cooldown = int(sys.argv[4])
with open(state_file) as f:
state = json.load(f)
pane = state.get(target, {})
last_restart = pane.get("last_restart", 0)
if now - last_restart > cooldown:
pane["restart_attempts"] = 0
state[target] = pane
with open(state_file, "w") as f:
json.dump(state, f, indent=2)
print(pane.get("restart_attempts", 0))
PYEOF
}
increment_restart_attempt() {
local target="$1"
local now
now=$(date +%s)
python3 - "$STATE_FILE" "$target" "$now" <<'PYEOF'
import json, sys
state_file = sys.argv[1]
target = sys.argv[2]
now = int(sys.argv[3])
with open(state_file) as f:
state = json.load(f)
pane = state.get(target, {})
pane["restart_attempts"] = pane.get("restart_attempts", 0) + 1
pane["last_restart"] = now
pane["status"] = "RESTARTING"
state[target] = pane
with open(state_file, "w") as f:
json.dump(state, f, indent=2)
print(pane["restart_attempts"])
PYEOF
}
# === CORE CHECK ===
check_pane() {
local target="$1"
local hash is_alive status current_cmd
# Capture state
hash=$(capture_pane_hash "$target")
if pane_pid_alive "$target"; then
is_alive="true"
else
is_alive="false"
fi
# Get current command for the pane
current_cmd=$(pane_current_command "$target")
# Update state and get result
local result
result=$(update_pane_state "$target" "$hash" "$is_alive")
status=$(echo "$result" | python3 -c "import json,sys; print(json.loads(sys.stdin.read()).get('status','UNKNOWN'))" 2>/dev/null || echo "UNKNOWN")
case "$status" in
OK)
# Healthy, do nothing
;;
DEAD)
log "DETECTED: $target is DEAD (PID gone) cmd=$current_cmd"
if is_restartable "$current_cmd"; then
handle_stuck "$target"
else
log "SKIP: $target not a hermes pane (cmd=$current_cmd), not restarting"
fi
;;
STUCK)
log "DETECTED: $target is STUCK (output unchanged for ${STUCK_CYCLES} cycles) cmd=$current_cmd"
if is_restartable "$current_cmd"; then
handle_stuck "$target"
else
log "SKIP: $target not a hermes pane (cmd=$current_cmd), not restarting"
fi
;;
STALE)
# Output unchanged but within threshold — just log
local count
count=$(echo "$result" | python3 -c "import json,sys; print(json.loads(sys.stdin.read()).get('same_count',0))" 2>/dev/null || echo "?")
log "STALE: $target unchanged for $count cycle(s)"
;;
esac
}
handle_stuck() {
local target="$1"
local session_name="${target%%:*}"
local attempts
# Check restart budget
attempts=$(maybe_reset_restarts "$target")
if [ "$attempts" -ge "$MAX_RESTART_ATTEMPTS" ]; then
log "ESCALATION: $target stuck ${attempts}x — manual intervention needed"
echo "ALERT: $target stuck after $attempts restart attempts" >&2
return 1
fi
attempts=$(increment_restart_attempt "$target")
log "ACTION: Restarting $target (attempt $attempts/$MAX_RESTART_ATTEMPTS)"
if restart_pane "$target"; then
log "OK: $target restarted successfully"
else
log "FAIL: $target restart failed (attempt $attempts)"
fi
}
check_all_sessions() {
local sessions
if [ -n "$MONITOR_SESSIONS" ]; then
IFS=',' read -ra sessions <<< "$MONITOR_SESSIONS"
else
sessions=()
while IFS= read -r line; do
[ -n "$line" ] && sessions+=("$line")
done < <(tmux list-sessions -F '#{session_name}' 2>/dev/null || true)
fi
local total=0 stuck=0 dead=0 ok=0
for session in "${sessions[@]}"; do
[ -z "$session" ] && continue
# Get pane targets
local panes
panes=$(tmux list-panes -t "$session" -F "${session}:#{window_index}.#{pane_index}" 2>/dev/null || true)
for target in $panes; do
check_pane "$target"
total=$((total + 1))
done
done
log "CHECK: Processed $total panes"
}
# === STATUS DISPLAY ===
show_status() {
if [ ! -f "$STATE_FILE" ]; then
echo "No pane state file found at $STATE_FILE"
echo "Run pane-watchdog.sh once to initialize."
exit 0
fi
python3 - "$STATE_FILE" <<'PYEOF'
import json, sys, time
state_file = sys.argv[1]
try:
with open(state_file) as f:
state = json.load(f)
except (FileNotFoundError, json.JSONDecodeError):
print("No state data yet.")
sys.exit(0)
if not state:
print("No panes tracked.")
sys.exit(0)
now = int(time.time())
print(f"{'PANE':<35} {'STATUS':<12} {'STALE':<6} {'LAST CHANGE':<15} {'RESTARTS'}")
print("-" * 90)
for target in sorted(state.keys()):
p = state[target]
status = p.get("status", "?")
same = p.get("same_count", 0)
last_change = p.get("last_change", 0)
restarts = p.get("restart_attempts", 0)
if last_change:
ago = now - last_change
if ago < 60:
change_str = f"{ago}s ago"
elif ago < 3600:
change_str = f"{ago//60}m ago"
else:
change_str = f"{ago//3600}h ago"
else:
change_str = "never"
# Color code
if status == "OK":
icon = "✓"
elif status == "STUCK":
icon = "✖"
elif status == "DEAD":
icon = "☠"
elif status == "STALE":
icon = "⏳"
else:
icon = "?"
print(f" {icon} {target:<32} {status:<12} {same:<6} {change_str:<15} {restarts}")
PYEOF
}
# === DAEMON MODE ===
run_daemon() {
log "DAEMON: Starting (interval=${CHECK_INTERVAL}s, stuck_threshold=${STUCK_CYCLES})"
echo "Pane watchdog started. Checking every ${CHECK_INTERVAL}s. Ctrl+C to stop."
echo "Log: $LOG_FILE"
echo "State: $STATE_FILE"
echo ""
while true; do
check_all_sessions
sleep "$CHECK_INTERVAL"
done
}
# === MAIN ===
case "${1:-}" in
--daemon)
run_daemon
;;
--status)
show_status
;;
--session)
if [ -z "${2:-}" ]; then
echo "Usage: pane-watchdog.sh --session SESSION_NAME"
exit 1
fi
MONITOR_SESSIONS="$2"
check_all_sessions
;;
--help|-h)
echo "pane-watchdog.sh — Detect stuck/dead tmux panes and auto-restart"
echo ""
echo "Usage:"
echo " pane-watchdog.sh # One-shot check"
echo " pane-watchdog.sh --daemon # Continuous monitoring"
echo " pane-watchdog.sh --status # Show pane state"
echo " pane-watchdog.sh --session S # Check one session"
echo ""
echo "Config (env vars):"
echo " PANE_CHECK_INTERVAL Seconds between checks (default: 120)"
echo " PANE_WATCHDOG_SESSIONS Comma-separated session names"
echo " PANE_STATE_FILE State file path"
echo " STUCK_CYCLES Unchanged cycles before STUCK (default: 2)"
;;
*)
check_all_sessions
;;
esac

View File

@@ -0,0 +1,271 @@
#!/usr/bin/env python3
"""
Pre-Flight Provider Check Script
Issue #508: [Robustness] Credential drain detection — provider health checks
Pre-flight check before session launch: verifies provider credentials and balance.
Usage:
python3 preflight-provider-check.py # Check all providers
python3 preflight-provider-check.py --launch # Check and return exit code
python3 preflight-provider-check.py --balance # Check OpenRouter balance
"""
import os, sys, json, yaml, urllib.request
from datetime import datetime, timezone
from pathlib import Path
# Configuration
HERMES_HOME = Path(os.environ.get("HERMES_HOME", Path.home() / ".hermes"))
LOG_DIR = Path.home() / ".local" / "timmy" / "fleet-health"
LOG_FILE = LOG_DIR / "preflight-check.log"
def log(msg):
"""Log message to file and optionally console."""
timestamp = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S")
log_entry = "[" + timestamp + "] " + msg
LOG_DIR.mkdir(parents=True, exist_ok=True)
with open(LOG_FILE, "a") as f:
f.write(log_entry + "\n")
if "--quiet" not in sys.argv:
print(log_entry)
def get_provider_api_key(provider):
"""Get API key for a provider from .env or environment."""
env_file = HERMES_HOME / ".env"
if env_file.exists():
with open(env_file) as f:
for line in f:
line = line.strip()
if line.startswith(provider.upper() + "_API_KEY="):
return line.split("=", 1)[1].strip().strip("'\"")
return os.environ.get(provider.upper() + "_API_KEY")
def check_openrouter_balance(api_key):
"""Check OpenRouter balance via /api/v1/auth/key."""
if not api_key:
return False, "No API key", 0
try:
req = urllib.request.Request(
"https://openrouter.ai/api/v1/auth/key",
headers={"Authorization": "Bearer " + api_key}
)
resp = urllib.request.urlopen(req, timeout=10)
data = json.loads(resp.read())
# Check for credits
credits = data.get("data", {}).get("limit", 0)
usage = data.get("data", {}).get("usage", 0)
remaining = credits - usage if credits else None
if remaining is not None and remaining <= 0:
return False, "No credits remaining", 0
elif remaining is not None:
return True, "Credits available", remaining
else:
return True, "Unlimited or unknown balance", None
except urllib.error.HTTPError as e:
if e.code == 401:
return False, "Invalid API key", 0
else:
return False, "HTTP " + str(e.code), 0
except Exception as e:
return False, str(e)[:100], 0
def check_nous_key(api_key):
"""Check Nous API key with minimal test call."""
if not api_key:
return False, "No API key"
try:
req = urllib.request.Request(
"https://inference.nousresearch.com/v1/models",
headers={"Authorization": "Bearer " + api_key}
)
resp = urllib.request.urlopen(req, timeout=10)
if resp.status == 200:
return True, "Valid key"
else:
return False, "HTTP " + str(resp.status)
except urllib.error.HTTPError as e:
if e.code == 401:
return False, "Invalid API key"
elif e.code == 403:
return False, "Forbidden"
else:
return False, "HTTP " + str(e.code)
except Exception as e:
return False, str(e)[:100]
def check_anthropic_key(api_key):
"""Check Anthropic API key with minimal test call."""
if not api_key:
return False, "No API key"
try:
req = urllib.request.Request(
"https://api.anthropic.com/v1/models",
headers={
"x-api-key": api_key,
"anthropic-version": "2023-06-01"
}
)
resp = urllib.request.urlopen(req, timeout=10)
if resp.status == 200:
return True, "Valid key"
else:
return False, "HTTP " + str(resp.status)
except urllib.error.HTTPError as e:
if e.code == 401:
return False, "Invalid API key"
elif e.code == 403:
return False, "Forbidden"
else:
return False, "HTTP " + str(e.code)
except Exception as e:
return False, str(e)[:100]
def check_ollama():
"""Check if Ollama is running."""
try:
req = urllib.request.Request("http://localhost:11434/api/tags")
resp = urllib.request.urlopen(req, timeout=5)
if resp.status == 200:
data = json.loads(resp.read())
models = data.get("models", [])
return True, str(len(models)) + " models loaded"
else:
return False, "HTTP " + str(resp.status)
except Exception as e:
return False, str(e)[:100]
def get_configured_provider():
"""Get the configured provider from global config."""
config_file = HERMES_HOME / "config.yaml"
if not config_file.exists():
return None
try:
with open(config_file) as f:
config = yaml.safe_load(f)
model_config = config.get("model", {})
if isinstance(model_config, dict):
return model_config.get("provider")
except:
pass
return None
def run_preflight_check():
"""Run pre-flight check on all providers."""
log("=== Pre-Flight Provider Check ===")
results = {}
# Check OpenRouter
or_key = get_provider_api_key("openrouter")
or_ok, or_msg, or_balance = check_openrouter_balance(or_key)
results["openrouter"] = {"healthy": or_ok, "message": or_msg, "balance": or_balance}
# Check Nous
nous_key = get_provider_api_key("nous")
nous_ok, nous_msg = check_nous_key(nous_key)
results["nous"] = {"healthy": nous_ok, "message": nous_msg}
# Check Anthropic
anthropic_key = get_provider_api_key("anthropic")
anthropic_ok, anthropic_msg = check_anthropic_key(anthropic_key)
results["anthropic"] = {"healthy": anthropic_ok, "message": anthropic_msg}
# Check Ollama
ollama_ok, ollama_msg = check_ollama()
results["ollama"] = {"healthy": ollama_ok, "message": ollama_msg}
# Get configured provider
configured = get_configured_provider()
# Summary
healthy_count = sum(1 for r in results.values() if r["healthy"])
total_count = len(results)
log("Results: " + str(healthy_count) + "/" + str(total_count) + " providers healthy")
for provider, result in results.items():
status = "HEALTHY" if result["healthy"] else "UNHEALTHY"
extra = ""
if provider == "openrouter" and result.get("balance") is not None:
extra = " (balance: " + str(result["balance"]) + ")"
log(" " + provider + ": " + status + " - " + result["message"] + extra)
if configured:
log("Configured provider: " + configured)
if configured in results and not results[configured]["healthy"]:
log("WARNING: Configured provider " + configured + " is UNHEALTHY!")
return results, configured
def check_launch_readiness():
"""Check if we're ready to launch sessions."""
results, configured = run_preflight_check()
# Check if configured provider is healthy
if configured and configured in results:
if not results[configured]["healthy"]:
log("LAUNCH BLOCKED: Configured provider " + configured + " is unhealthy")
return False, configured + " is unhealthy"
# Check if at least one provider is healthy
healthy_providers = [p for p, r in results.items() if r["healthy"]]
if not healthy_providers:
log("LAUNCH BLOCKED: No healthy providers available")
return False, "No healthy providers"
log("LAUNCH READY: " + str(len(healthy_providers)) + " healthy providers available")
return True, "Ready"
def show_balance():
"""Show OpenRouter balance."""
api_key = get_provider_api_key("openrouter")
if not api_key:
print("No OpenRouter API key found")
return
ok, msg, balance = check_openrouter_balance(api_key)
if ok:
if balance is not None:
print("OpenRouter balance: " + str(balance) + " credits")
else:
print("OpenRouter: " + msg)
else:
print("OpenRouter: " + msg)
def main():
if "--balance" in sys.argv:
show_balance()
elif "--launch" in sys.argv:
ready, message = check_launch_readiness()
if ready:
print("READY")
sys.exit(0)
else:
print("BLOCKED: " + message)
sys.exit(1)
else:
run_preflight_check()
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,411 @@
#!/usr/bin/env python3
"""
Provider Health Monitor Script
Issue #509: [Robustness] Provider-aware profile config — auto-switch on failure
Monitors provider health and automatically switches profiles to working providers.
Usage:
python3 provider-health-monitor.py # Run once
python3 provider-health-monitor.py --daemon # Run continuously
python3 provider-health-monitor.py --status # Show provider health
"""
import os, sys, json, yaml, urllib.request, time
from datetime import datetime, timezone
from pathlib import Path
# Configuration
HERMES_HOME = Path(os.environ.get("HERMES_HOME", Path.home() / ".hermes"))
PROFILES_DIR = HERMES_HOME / "profiles"
LOG_DIR = Path.home() / ".local" / "timmy" / "fleet-health"
STATE_FILE = LOG_DIR / "tmux-state.json"
LOG_FILE = LOG_DIR / "provider-health.log"
# Provider test endpoints
PROVIDER_TESTS = {
"openrouter": {
"url": "https://openrouter.ai/api/v1/models",
"method": "GET",
"headers": lambda api_key: {"Authorization": "Bearer " + api_key},
"timeout": 10
},
"anthropic": {
"url": "https://api.anthropic.com/v1/models",
"method": "GET",
"headers": lambda api_key: {"x-api-key": api_key, "anthropic-version": "2023-06-01"},
"timeout": 10
},
"nous": {
"url": "https://inference.nousresearch.com/v1/models",
"method": "GET",
"headers": lambda api_key: {"Authorization": "Bearer " + api_key},
"timeout": 10
},
"kimi-coding": {
"url": "https://api.kimi.com/coding/v1/models",
"method": "GET",
"headers": lambda api_key: {"x-api-key": api_key, "x-api-provider": "kimi-coding"},
"timeout": 10
},
"ollama": {
"url": "http://localhost:11434/api/tags",
"method": "GET",
"headers": lambda api_key: {},
"timeout": 5
}
}
def log(msg):
"""Log message to file and optionally console."""
timestamp = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S")
log_entry = "[" + timestamp + "] " + msg
LOG_DIR.mkdir(parents=True, exist_ok=True)
with open(LOG_FILE, "a") as f:
f.write(log_entry + "\n")
if "--quiet" not in sys.argv:
print(log_entry)
def get_provider_api_key(provider):
"""Get API key for a provider from .env or environment."""
env_file = HERMES_HOME / ".env"
if env_file.exists():
with open(env_file) as f:
for line in f:
line = line.strip()
if line.startswith(provider.upper() + "_API_KEY="):
return line.split("=", 1)[1].strip().strip("'\"")
return os.environ.get(provider.upper() + "_API_KEY")
def test_provider(provider, api_key=None):
"""Test if a provider is healthy."""
config = PROVIDER_TESTS.get(provider)
if not config:
return False, "Unknown provider: " + provider
headers = config["headers"](api_key or "")
try:
req = urllib.request.Request(
config["url"],
headers=headers,
method=config["method"]
)
resp = urllib.request.urlopen(req, timeout=config["timeout"])
if resp.status == 200:
return True, "Healthy"
else:
return False, "HTTP " + str(resp.status)
except urllib.error.HTTPError as e:
if e.code == 401:
return False, "Unauthorized (401)"
elif e.code == 403:
return False, "Forbidden (403)"
elif e.code == 429:
return True, "Rate limited but accessible"
else:
return False, "HTTP " + str(e.code)
except Exception as e:
return False, str(e)[:100]
def get_all_providers():
"""Get all providers from profiles and global config."""
providers = set()
# Global config
global_config = HERMES_HOME / "config.yaml"
if global_config.exists():
try:
with open(global_config) as f:
config = yaml.safe_load(f)
# Primary model provider
model_config = config.get("model", {})
if isinstance(model_config, dict):
provider = model_config.get("provider", "")
if provider:
providers.add(provider)
# Auxiliary providers
auxiliary = config.get("auxiliary", {})
for aux_config in auxiliary.values():
if isinstance(aux_config, dict):
provider = aux_config.get("provider", "")
if provider and provider != "auto":
providers.add(provider)
except:
pass
# Profile configs
if PROFILES_DIR.exists():
for profile_dir in PROFILES_DIR.iterdir():
if profile_dir.is_dir():
config_file = profile_dir / "config.yaml"
if config_file.exists():
try:
with open(config_file) as f:
config = yaml.safe_load(f)
model_config = config.get("model", {})
if isinstance(model_config, dict):
provider = model_config.get("provider", "")
if provider:
providers.add(provider)
auxiliary = config.get("auxiliary", {})
for aux_config in auxiliary.values():
if isinstance(aux_config, dict):
provider = aux_config.get("provider", "")
if provider and provider != "auto":
providers.add(provider)
except:
pass
# Add common providers even if not configured
providers.update(["openrouter", "nous", "ollama"])
return list(providers)
def build_health_map():
"""Build a health map of all providers."""
providers = get_all_providers()
health_map = {}
log("Testing " + str(len(providers)) + " providers...")
for provider in providers:
api_key = get_provider_api_key(provider)
healthy, message = test_provider(provider, api_key)
health_map[provider] = {
"healthy": healthy,
"message": message,
"last_test": datetime.now(timezone.utc).isoformat(),
"api_key_present": bool(api_key)
}
status = "HEALTHY" if healthy else "UNHEALTHY"
log(" " + provider + ": " + status + " - " + message)
return health_map
def get_fallback_providers(health_map):
"""Get list of healthy providers in priority order."""
# Priority order: nous, openrouter, ollama, others
priority_order = ["nous", "openrouter", "ollama", "anthropic", "kimi-coding"]
healthy = []
for provider in priority_order:
if provider in health_map and health_map[provider]["healthy"]:
healthy.append(provider)
# Add any other healthy providers not in priority list
for provider, info in health_map.items():
if info["healthy"] and provider not in healthy:
healthy.append(provider)
return healthy
def update_profile_config(profile_name, new_provider):
"""Update a profile's config to use a new provider."""
config_file = PROFILES_DIR / profile_name / "config.yaml"
if not config_file.exists():
return False, "Config file not found"
try:
with open(config_file) as f:
config = yaml.safe_load(f)
# Update model provider
if "model" not in config:
config["model"] = {}
old_provider = config["model"].get("provider", "unknown")
config["model"]["provider"] = new_provider
# Update auxiliary providers if they were using the old provider
auxiliary = config.get("auxiliary", {})
for aux_name, aux_config in auxiliary.items():
if isinstance(aux_config, dict) and aux_config.get("provider") == old_provider:
aux_config["provider"] = new_provider
# Write back
with open(config_file, "w") as f:
yaml.dump(config, f, default_flow_style=False)
log("Updated " + profile_name + ": " + old_provider + " -> " + new_provider)
return True, "Updated"
except Exception as e:
return False, str(e)
def check_profiles(health_map):
"""Check all profiles and update unhealthy providers."""
if not PROFILES_DIR.exists():
return
fallback_providers = get_fallback_providers(health_map)
if not fallback_providers:
log("CRITICAL: No healthy providers available!")
return
updated_profiles = []
for profile_dir in PROFILES_DIR.iterdir():
if not profile_dir.is_dir():
continue
profile_name = profile_dir.name
config_file = profile_dir / "config.yaml"
if not config_file.exists():
continue
try:
with open(config_file) as f:
config = yaml.safe_load(f)
model_config = config.get("model", {})
if not isinstance(model_config, dict):
continue
current_provider = model_config.get("provider", "")
if not current_provider:
continue
# Check if current provider is healthy
if current_provider in health_map and health_map[current_provider]["healthy"]:
continue # Provider is healthy, no action needed
# Find best fallback
best_fallback = None
for provider in fallback_providers:
if provider != current_provider:
best_fallback = provider
break
if not best_fallback:
log("No fallback for " + profile_name + " (current: " + current_provider + ")")
continue
# Update profile
success, message = update_profile_config(profile_name, best_fallback)
if success:
updated_profiles.append({
"profile": profile_name,
"old_provider": current_provider,
"new_provider": best_fallback
})
except Exception as e:
log("Error processing " + profile_name + ": " + str(e))
return updated_profiles
def load_state():
"""Load state from tmux-state.json."""
if STATE_FILE.exists():
try:
with open(STATE_FILE) as f:
return json.load(f)
except:
pass
return {}
def save_state(state):
"""Save state to tmux-state.json."""
LOG_DIR.mkdir(parents=True, exist_ok=True)
with open(STATE_FILE, "w") as f:
json.dump(state, f, indent=2)
def run_once():
"""Run provider health check once."""
log("=== Provider Health Check ===")
state = load_state()
# Build health map
health_map = build_health_map()
# Check profiles and update if needed
updated_profiles = check_profiles(health_map)
# Update state
state["provider_health"] = health_map
state["last_provider_check"] = datetime.now(timezone.utc).isoformat()
if updated_profiles:
state["last_profile_updates"] = updated_profiles
save_state(state)
# Summary
healthy_count = sum(1 for p in health_map.values() if p["healthy"])
total_count = len(health_map)
log("Health: " + str(healthy_count) + "/" + str(total_count) + " providers healthy")
if updated_profiles:
log("Updated " + str(len(updated_profiles)) + " profiles:")
for update in updated_profiles:
log(" " + update["profile"] + ": " + update["old_provider"] + " -> " + update["new_provider"])
def show_status():
"""Show provider health status."""
state = load_state()
health_map = state.get("provider_health", {})
if not health_map:
print("No provider health data available. Run without --status first.")
return
print("Provider Health (last updated: " + str(state.get("last_provider_check", "unknown")) + ")")
print("=" * 80)
for provider, info in sorted(health_map.items()):
status = "HEALTHY" if info["healthy"] else "UNHEALTHY"
message = info.get("message", "")
api_key = "yes" if info.get("api_key_present") else "no"
print(provider.ljust(20) + " " + status.ljust(10) + " API key: " + api_key + " - " + message)
# Show recent updates
updates = state.get("last_profile_updates", [])
if updates:
print()
print("Recent Profile Updates:")
for update in updates:
print(" " + update["profile"] + ": " + update["old_provider"] + " -> " + update["new_provider"])
def daemon_mode():
"""Run continuously."""
log("Starting provider health daemon (check every 300s)")
while True:
try:
run_once()
time.sleep(300) # Check every 5 minutes
except KeyboardInterrupt:
log("Daemon stopped by user")
break
except Exception as e:
log("Error: " + str(e))
time.sleep(60)
def main():
if "--status" in sys.argv:
show_status()
elif "--daemon" in sys.argv:
daemon_mode()
else:
run_once()
if __name__ == "__main__":
main()

View File

@@ -1,3 +1,4 @@
#!/usr/bin/env python3
"""
Soul Eval Gate — The Conscience of the Training Pipeline

View File

@@ -3,7 +3,7 @@
# Uses Hermes CLI plus workforce-manager to triage and review.
# Timmy is the brain. Other agents are the hands.
set -uo pipefail
set -uo pipefail\n\nSCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
LOG_DIR="$HOME/.hermes/logs"
LOG="$LOG_DIR/timmy-orchestrator.log"
@@ -40,6 +40,7 @@ gather_state() {
> "$state_dir/unassigned.txt"
> "$state_dir/open_prs.txt"
> "$state_dir/agent_status.txt"
> "$state_dir/uncommitted_work.txt"
for repo in $REPOS; do
local short=$(echo "$repo" | cut -d/ -f2)
@@ -71,6 +72,24 @@ for p in json.load(sys.stdin):
tail -50 "/tmp/kimi-heartbeat.log" 2>/dev/null | grep -c "FAILED:" | xargs -I{} echo "Kimi recent failures: {}" >> "$state_dir/agent_status.txt"
tail -1 "/tmp/kimi-heartbeat.log" 2>/dev/null | xargs -I{} echo "Kimi last event: {}" >> "$state_dir/agent_status.txt"
# Scan worktrees for uncommitted work
for wt_dir in "$HOME/worktrees"/*/; do
[ -d "$wt_dir" ] || continue
[ -d "$wt_dir/.git" ] || continue
local dirty
dirty=$(cd "$wt_dir" && git status --porcelain 2>/dev/null | wc -l | tr -d " ")
if [ "${dirty:-0}" -gt 0 ]; then
local branch
branch=$(cd "$wt_dir" && git branch --show-current 2>/dev/null || echo "?")
local age=""
local last_commit
last_commit=$(cd "$wt_dir" && git log -1 --format=%ct 2>/dev/null || echo 0)
local now=$(date +%s)
local stale_mins=$(( (now - last_commit) / 60 ))
echo "DIR=$wt_dir BRANCH=$branch DIRTY=$dirty STALE=${stale_mins}m" >> "$state_dir/uncommitted_work.txt"
fi
done
echo "$state_dir"
}
@@ -81,6 +100,25 @@ run_triage() {
log "Cycle: $unassigned_count unassigned, $pr_count open PRs"
# Check for uncommitted work — nag if stale
local uncommitted_count
uncommitted_count=$(wc -l < "$state_dir/uncommitted_work.txt" 2>/dev/null | tr -d " " || echo 0)
if [ "${uncommitted_count:-0}" -gt 0 ]; then
log "WARNING: $uncommitted_count worktree(s) with uncommitted work"
while IFS= read -r line; do
log " UNCOMMITTED: $line"
# Auto-commit stale work (>60 min without commit)
local stale=$(echo "$line" | sed 's/.*STALE=\([0-9]*\)m.*/\1/')
local wt_dir=$(echo "$line" | sed 's/.*DIR=\([^ ]*\) .*/\1/')
if [ "${stale:-0}" -gt 60 ]; then
log " AUTO-COMMITTING stale work in $wt_dir (${stale}m stale)"
(cd "$wt_dir" && git add -A && git commit -m "WIP: orchestrator auto-commit — ${stale}m stale work
Preserved by timmy-orchestrator to prevent loss." 2>/dev/null && git push 2>/dev/null) && log " COMMITTED: $wt_dir" || log " COMMIT FAILED: $wt_dir"
fi
done < "$state_dir/uncommitted_work.txt"
fi
# If nothing to do, skip the LLM call
if [ "$unassigned_count" -eq 0 ] && [ "$pr_count" -eq 0 ]; then
log "Nothing to triage"
@@ -198,6 +236,12 @@ FOOTER
log "=== Timmy Orchestrator Started (PID $$) ==="
log "Cycle: ${CYCLE_INTERVAL}s | Auto-assign: ${AUTO_ASSIGN_UNASSIGNED} | Inference surface: Hermes CLI"
# Start auto-commit-guard daemon for work preservation
if ! pgrep -f "auto-commit-guard.sh" >/dev/null 2>&1; then
nohup bash "$SCRIPT_DIR/auto-commit-guard.sh" 120 >> "$LOG_DIR/auto-commit-guard.log" 2>&1 &
log "Started auto-commit-guard daemon (PID $!)"
fi
WORKFORCE_CYCLE=0
while true; do

View File

@@ -196,7 +196,37 @@
"paused_reason": null,
"skills": [],
"skill": null
},
{
"id": "tmux-supervisor-513",
"name": "Autonomous Cron Supervisor",
"prompt": "Load the tmux-supervisor skill and execute the monitoring protocol.\n\nCheck both `dev` and `timmy` tmux sessions for idle panes. Only send Telegram notifications on actionable events (idle, overflow, failure). Be silent when all agents are working.\n\nSteps:\n1. List all tmux sessions (skip 'Alexander')\n2. For each session, list windows and panes\n3. Capture each pane and classify state (idle vs active)\n4. For idle panes: read context, craft context-aware prompt\n5. Send /queue prompts to idle panes\n6. Verify prompts landed\n7. Only notify via Telegram if:\n - A pane was prompted (idle detected)\n - A pane shows context overflow (>80%)\n - A pane is stuck or crashed\n8. If all panes are active: respond with [SILENT]",
"schedule": {
"kind": "interval",
"minutes": 7,
"display": "every 7m"
},
"schedule_display": "every 7m",
"repeat": {
"times": null,
"completed": 0
},
"enabled": true,
"created_at": "2026-04-15T03:00:00.000000+00:00",
"next_run_at": null,
"last_run_at": null,
"last_status": null,
"last_error": null,
"deliver": "telegram",
"origin": null,
"state": "scheduled",
"paused_at": null,
"paused_reason": null,
"skills": [
"tmux-supervisor"
],
"skill": "tmux-supervisor"
}
],
"updated_at": "2026-04-13T02:00:00+00:00"
}
}

View File

@@ -0,0 +1,9 @@
- name: Nightly Pipeline Scheduler
schedule: '*/30 18-23,0-8 * * *' # Every 30 min, off-peak hours only
tasks:
- name: Check and start pipelines
shell: "bash scripts/nightly-pipeline-scheduler.sh"
env:
PIPELINE_TOKEN_LIMIT: "500000"
PIPELINE_PEAK_START: "9"
PIPELINE_PEAK_END: "18"

View File

@@ -0,0 +1,24 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>Label</key>
<string>ai.timmy.auto-commit-guard</string>
<key>ProgramArguments</key>
<array>
<string>/bin/bash</string>
<string>/Users/apayne/.hermes/bin/auto-commit-guard.sh</string>
<string>120</string>
</array>
<key>RunAtLoad</key>
<true/>
<key>KeepAlive</key>
<true/>
<key>StandardOutPath</key>
<string>/Users/apayne/.hermes/logs/auto-commit-guard.stdout.log</string>
<key>StandardErrorPath</key>
<string>/Users/apayne/.hermes/logs/auto-commit-guard.stderr.log</string>
<key>WorkingDirectory</key>
<string>/Users/apayne</string>
</dict>
</plist>

View File

@@ -0,0 +1,21 @@
# Gitea Accessibility Fix - R4: Time Elements
WCAG 1.3.1: Relative timestamps lack machine-readable fallbacks.
## Fix
Wrap relative timestamps in `<time datetime="...">` elements.
## Files
- `custom/templates/custom/time_relative.tmpl` - Reusable `<time>` helper
- `custom/templates/repo/list_a11y.tmpl` - Explore/Repos list override
## Deploy
```bash
cp -r custom/templates/* /path/to/gitea/custom/templates/
systemctl restart gitea
```
Closes #554

View File

@@ -0,0 +1,27 @@
{{/*
Gitea a11y fix: R4 <time> elements for relative timestamps
Deploy to: custom/templates/custom/time_relative.tmpl
*/}}
{{define "custom/time_relative"}}
{{if and .Time .Relative}}
<time datetime="{{.Time.Format "2006-01-02T15:04:05Z07:00"}}" title="{{.Time.Format "Jan 02, 2006 15:04"}}">
{{.Relative}}
</time>
{{else if .Relative}}
<span>{{.Relative}}</span>
{{end}}
{{end}}
{{define "custom/time_from_unix"}}
{{if .Relative}}
<time datetime="" data-unix="{{.Unix}}" title="">{{.Relative}}</time>
<script>
(function() {
var el = document.currentScript.previousElementSibling;
var unix = parseInt(el.getAttribute('data-unix'));
if (unix) { el.setAttribute('datetime', new Date(unix * 1000).toISOString()); el.setAttribute('title', new Date(unix * 1000).toLocaleString()); }
})();
</script>
{{end}}
{{end}}

View File

@@ -0,0 +1,27 @@
{{/*
Gitea a11y fix: R4 <time> elements for relative timestamps on repo list
Deploy to: custom/templates/repo/list_a11y.tmpl
*/}}
{{/* Star count link with aria-label */}}
<a class="repo-card-star" href="{{.RepoLink}}/stars" aria-label="{{.NumStars}} stars" title="{{.NumStars}} stars">
<svg class="octicon octicon-star" viewBox="0 0 16 16" width="16" height="16" aria-hidden="true">
<path d="M8 .25a.75.75 0 01.673.418l1.882 3.815 4.21.612a.75.75 0 01.416 1.279l-3.046 2.97.719 4.192a.75.75 0 01-1.088.791L8 12.347l-3.766 1.98a.75.75 0 01-1.088-.79l.72-4.194L.818 6.374a.75.75 0 01.416-1.28l4.21-.611L7.327.668A.75.75 0 018 .25z"/>
</svg>
<span>{{.NumStars}}</span>
</a>
{{/* Fork count link with aria-label */}}
<a class="repo-card-fork" href="{{.RepoLink}}/forks" aria-label="{{.NumForks}} forks" title="{{.NumForks}} forks">
<svg class="octicon octicon-repo-forked" viewBox="0 0 16 16" width="16" height="16" aria-hidden="true">
<path d="M5 5.372v.878c0 .414.336.75.75.75h4.5a.75.75 0 00.75-.75v-.878a2.25 2.25 0 111.5 0v.878a2.25 2.25 0 01-2.25 2.25h-1.5v2.128a2.251 2.251 0 11-1.5 0V8.5h-1.5A2.25 2.25 0 013.5 6.25v-.878a2.25 2.25 0 111.5 0zM5 3.25a.75.75 0 10-1.5 0 .75.75 0 001.5 0zm6.75.75a.75.75 0 100-1.5.75.75 0 000 1.5zm-3 8.75a.75.75 0 10-1.5 0 .75.75 0 001.5 0z"/>
</svg>
<span>{{.NumForks}}</span>
</a>
{{/* Relative timestamp with <time> element for a11y */}}
{{if .UpdatedUnix}}
<time datetime="{{.UpdatedUnix | TimeSinceISO}}" title="{{.UpdatedUnix | DateFmtLong}}" class="text-light">
{{.UpdatedUnix | TimeSince}}
</time>
{{end}}

179
docs/glitch-detection.md Normal file
View File

@@ -0,0 +1,179 @@
# 3D World Glitch Detection — Matrix Scanner
**Reference:** timmy-config#491
**Label:** gemma-4-multimodal
**Version:** 0.1.0
## Overview
The Matrix Glitch Detector scans 3D web worlds for visual artifacts and
rendering anomalies. It uses browser automation to capture screenshots from
multiple camera angles, then sends them to a vision AI model for analysis
against a library of known glitch patterns.
## Detected Glitch Categories
| Category | Severity | Description |
|---|---|---|
| Floating Assets | HIGH | Objects not grounded — hovering above surfaces |
| Z-Fighting | MEDIUM | Coplanar surfaces flickering/competing for depth |
| Missing Textures | CRITICAL | Placeholder colors (magenta, checkerboard) |
| Clipping | HIGH | Geometry passing through other objects |
| Broken Normals | MEDIUM | Inside-out or incorrectly lit surfaces |
| Shadow Artifacts | LOW | Detached, mismatched, or acne shadows |
| LOD Popping | LOW | Abrupt level-of-detail transitions |
| Lightmap Errors | MEDIUM | Dark splotches, light leaks, baking failures |
| Water/Reflection | MEDIUM | Incorrect environment reflections |
| Skybox Seam | LOW | Visible seams at cubemap face edges |
## Installation
No external dependencies required — pure Python 3.10+.
```bash
# Clone the repo
git clone https://forge.alexanderwhitestone.com/Timmy_Foundation/timmy-config.git
cd timmy-config
```
## Usage
### Basic Scan
```bash
python bin/matrix_glitch_detector.py https://matrix.example.com/world/alpha
```
### Multi-Angle Scan
```bash
python bin/matrix_glitch_detector.py https://matrix.example.com/world/alpha \
--angles 8 \
--output glitch_report.json
```
### Demo Mode
```bash
python bin/matrix_glitch_detector.py --demo
```
### Options
| Flag | Default | Description |
|---|---|---|
| `url` | (required) | URL of the 3D world to scan |
| `--angles N` | 4 | Number of camera angles to capture |
| `--output PATH` | stdout | Output file for JSON report |
| `--min-severity` | info | Minimum severity: info/low/medium/high/critical |
| `--demo` | off | Run with simulated detections |
| `--verbose` | off | Enable verbose output |
## Report Format
The JSON report includes:
```json
{
"scan_id": "uuid",
"url": "https://...",
"timestamp": "ISO-8601",
"total_screenshots": 4,
"angles_captured": ["front", "right", "back", "left"],
"glitches": [
{
"id": "short-uuid",
"category": "floating_assets",
"name": "Floating Chair",
"description": "Office chair floating 0.3m above floor",
"severity": "high",
"confidence": 0.87,
"location_x": 35.2,
"location_y": 62.1,
"screenshot_index": 0,
"screenshot_angle": "front",
"timestamp": "ISO-8601"
}
],
"summary": {
"total_glitches": 4,
"by_severity": {"critical": 1, "high": 2, "medium": 1},
"by_category": {"floating_assets": 1, "missing_textures": 1, ...},
"highest_severity": "critical",
"clean_screenshots": 0
},
"metadata": {
"detector_version": "0.1.0",
"pattern_count": 10,
"reference": "timmy-config#491"
}
}
```
## Vision AI Integration
The detector supports any OpenAI-compatible vision API. Set these
environment variables:
```bash
export VISION_API_KEY="your-api-key"
export VISION_API_BASE="https://api.openai.com/v1" # optional
export VISION_MODEL="gpt-4o" # optional, default: gpt-4o
```
For browser-based capture with `browser_vision`:
```bash
export BROWSER_VISION_SCRIPT="/path/to/browser_vision.py"
```
## Glitch Patterns
Pattern definitions live in `bin/glitch_patterns.py`. Each pattern includes:
- **category** — Enum matching the glitch type
- **detection_prompts** — Instructions for the vision model
- **visual_indicators** — What to look for in screenshots
- **confidence_threshold** — Minimum confidence to report
### Adding Custom Patterns
```python
from glitch_patterns import GlitchPattern, GlitchCategory, GlitchSeverity
custom = GlitchPattern(
category=GlitchCategory.FLOATING_ASSETS,
name="Custom Glitch",
description="Your description",
severity=GlitchSeverity.MEDIUM,
detection_prompts=["Look for..."],
visual_indicators=["indicator 1", "indicator 2"],
)
```
## Testing
```bash
python -m pytest tests/test_glitch_detector.py -v
# or
python tests/test_glitch_detector.py
```
## Architecture
```
bin/
matrix_glitch_detector.py — Main CLI entry point
glitch_patterns.py — Pattern definitions and prompt builder
tests/
test_glitch_detector.py — Unit and integration tests
docs/
glitch-detection.md — This documentation
```
## Limitations
- Browser automation requires a headless browser environment
- Vision AI analysis depends on model availability and API limits
- Placeholder screenshots are generated when browser capture is unavailable
- Detection accuracy varies by scene complexity and lighting conditions

419
pipeline/quality_gate.py Executable file
View File

@@ -0,0 +1,419 @@
#!/usr/bin/env python3
"""
quality_gate.py — Quality Gate for Pipeline Outputs
Validates all pipeline outputs before saving. Rejects bad outputs,
tracks quality scores, and supports re-queue for regeneration.
Usage:
python3 quality_gate.py --input output.jsonl --type training_pairs
python3 quality_gate.py --input output.jsonl --type knowledge
python3 quality_gate.py --input output.jsonl --type scene_descriptions
python3 quality_gate.py --dir pipeline/output/ --type training_pairs
python3 quality_gate.py --status # show quality stats
Exit codes:
0 = all outputs passed
1 = some outputs rejected
2 = file/parse error
"""
import json
import os
import sys
import hashlib
import re
from pathlib import Path
from datetime import datetime, timezone
from dataclasses import dataclass, field, asdict
from typing import List, Optional, Dict, Any
STATS_FILE = Path.home() / ".hermes" / "pipeline" / "quality_stats.json"
# --- Quality Check Types ---
@dataclass
class QualityResult:
"""Result of a quality check on a single entry."""
passed: bool
checks_run: int
checks_failed: int
score: float # 0.0-1.0
reasons: List[str] = field(default_factory=list)
entry_index: int = -1
hash: str = ""
def to_dict(self):
return asdict(self)
@dataclass
class GateReport:
"""Report from a quality gate run."""
file: str
type: str
total: int
passed: int
rejected: int
score: float
rejected_indices: List[int] = field(default_factory=list)
timestamp: str = field(default_factory=lambda: datetime.now(timezone.utc).isoformat())
def to_dict(self):
return asdict(self)
# ============================================================
# Check Functions
# ============================================================
def entry_hash(entry: dict) -> str:
"""Hash an entry for deduplication."""
return hashlib.sha256(json.dumps(entry, sort_keys=True, ensure_ascii=False).encode()).hexdigest()[:16]
def check_not_empty(entry: dict, fields: List[str]) -> List[str]:
"""Check that required fields are non-empty."""
errors = []
for f in fields:
val = entry.get(f)
if val is None:
errors.append(f"missing_field: {f}")
elif isinstance(val, str) and len(val.strip()) == 0:
errors.append(f"empty_field: {f}")
elif isinstance(val, list) and len(val) == 0:
errors.append(f"empty_list: {f}")
return errors
def check_string_min_length(entry: dict, field_lengths: Dict[str, int]) -> List[str]:
"""Check that string fields meet minimum lengths."""
errors = []
for f, min_len in field_lengths.items():
val = entry.get(f)
if isinstance(val, str) and len(val) < min_len:
errors.append(f"short_field: {f} ({len(val)} < {min_len})")
return errors
def check_no_duplicates(entries: List[dict], key_fields: List[str]) -> Dict[int, List[str]]:
"""Check for duplicate entries based on key fields."""
seen = {}
errors = {}
for i, entry in enumerate(entries):
key = tuple(entry.get(f, "") for f in key_fields)
key_str = str(key)
if key_str in seen:
errors[i] = [f"duplicate_of_index: {seen[key_str]}"]
else:
seen[key_str] = i
return errors
def check_training_pair(entry: dict) -> List[str]:
"""Validate a training pair (prompt/response)."""
errors = []
errors.extend(check_not_empty(entry, ["prompt", "response"]))
# Check response isn't just echoing the prompt
prompt = entry.get("prompt", "")
response = entry.get("response", "")
if prompt and response and prompt.strip() == response.strip():
errors.append("response_equals_prompt")
# Check response has substance
if isinstance(response, str) and len(response) < 10:
errors.append(f"response_too_short: {len(response)} chars")
return errors
def check_scene_description(entry: dict) -> List[str]:
"""Validate a scene description entry."""
errors = []
errors.extend(check_not_empty(entry, ["song", "beat", "lyric_line", "scene"]))
scene = entry.get("scene")
if isinstance(scene, dict):
errors.extend(check_not_empty(scene, ["mood", "colors", "composition", "camera", "description"]))
errors.extend(check_string_min_length(scene, {"description": 10}))
colors = scene.get("colors", [])
if isinstance(colors, list) and len(colors) > 5:
errors.append(f"too_many_colors: {len(colors)} > 5")
return errors
def check_knowledge_entry(entry: dict) -> List[str]:
"""Validate a knowledge file entry."""
errors = []
errors.extend(check_not_empty(entry, ["title", "content"]))
# Check for placeholder content
content = entry.get("content", "")
if isinstance(content, str):
placeholders = ["TODO", "FIXME", "PLACEHOLDER", "[INSERT", "lorem ipsum"]
for p in placeholders:
if p.lower() in content.lower():
errors.append(f"placeholder_content: '{p}' found")
errors.extend(check_string_min_length(entry, {"content": 50}))
return errors
def check_prompt_enhancement(entry: dict) -> List[str]:
"""Validate a prompt enhancement pair (terse/rich)."""
errors = []
errors.extend(check_not_empty(entry, ["terse", "rich"]))
terse = entry.get("terse", "")
rich = entry.get("rich", "")
# Rich should be longer than terse
if isinstance(terse, str) and isinstance(rich, str) and len(rich) <= len(terse):
errors.append("rich_not_longer_than_terse")
errors.extend(check_string_min_length(entry, {"rich": 20}))
return errors
def check_adversary_entry(entry: dict) -> List[str]:
"""Validate an adversary corpus entry."""
errors = []
errors.extend(check_not_empty(entry, ["id", "family", "prompt"]))
# Check prompt isn't empty or placeholder
prompt = entry.get("prompt", "")
if isinstance(prompt, str) and len(prompt) < 10:
errors.append(f"prompt_too_short: {len(prompt)} chars")
return errors
def check_soul_compliance(text: str) -> List[str]:
"""Check text for SOUL.md compliance (basic checks)."""
errors = []
text_lower = text.lower() if text else ""
# Check for content that violates core values
violations = [
("compute.*value.*human.*life", "soul_violation: computing human life value"),
("sexualiz.*child", "soul_violation: sexualizing children"),
("create.*weapon.*kill", "soul_violation: weapon creation"),
("enslav", "soul_violation: enslavement content"),
]
for pattern, msg in violations:
if re.search(pattern, text_lower):
errors.append(msg)
return errors
# ============================================================
# Gate Runners
# ============================================================
CHECK_MAP = {
"training_pairs": check_training_pair,
"training_pair": check_training_pair,
"scene_descriptions": check_scene_description,
"scene_description": check_scene_description,
"knowledge": check_knowledge_entry,
"prompt_enhancement": check_prompt_enhancement,
"adversary": check_adversary_entry,
"adversary_corpus": check_adversary_entry,
}
def run_gate(input_path: str, entry_type: str) -> GateReport:
"""Run quality gate on a JSONL file."""
path = Path(input_path)
if not path.exists():
return GateReport(file=str(path), type=entry_type, total=0, passed=0, rejected=0, score=0.0)
check_fn = CHECK_MAP.get(entry_type)
if not check_fn:
return GateReport(file=str(path), type=entry_type, total=0, passed=0, rejected=0, score=0.0,
rejected_indices=[-1]) # unknown type
entries = []
with open(path) as f:
for line in f:
line = line.strip()
if line:
entries.append(json.loads(line))
# Deduplication check
key_fields = _get_key_fields(entry_type)
dup_errors = check_no_duplicates(entries, key_fields)
passed = 0
rejected = 0
rejected_indices = []
total_score = 0.0
for i, entry in enumerate(entries):
errors = check_fn(entry)
# Add duplicate errors
if i in dup_errors:
errors.extend(dup_errors[i])
# Add SOUL compliance check for text content
text_content = ""
for f in ["response", "rich", "description", "content", "lyric_line"]:
val = entry.get(f)
if isinstance(val, str):
text_content += val + " "
if isinstance(entry.get("scene"), dict):
text_content += entry["scene"].get("description", "")
soul_errors = check_soul_compliance(text_content)
errors.extend(soul_errors)
if errors:
rejected += 1
rejected_indices.append(i)
else:
passed += 1
# Score: 1.0 if no errors, decreasing with each error
entry_score = max(0.0, 1.0 - (len(errors) * 0.2))
total_score += entry_score
avg_score = total_score / len(entries) if entries else 0.0
report = GateReport(
file=str(path),
type=entry_type,
total=len(entries),
passed=passed,
rejected=rejected,
score=round(avg_score, 3),
rejected_indices=rejected_indices[:50], # limit for readability
)
# Save stats
_save_stats(report)
return report
def _get_key_fields(entry_type: str) -> List[str]:
"""Get key fields for deduplication based on entry type."""
key_map = {
"training_pairs": ["prompt", "response"],
"training_pair": ["prompt", "response"],
"scene_descriptions": ["song", "beat"],
"scene_description": ["song", "beat"],
"knowledge": ["title"],
"prompt_enhancement": ["terse", "rich"],
"adversary": ["id", "prompt"],
"adversary_corpus": ["id", "prompt"],
}
return key_map.get(entry_type, ["id"])
def _save_stats(report: GateReport):
"""Append quality stats to the stats file."""
STATS_FILE.parent.mkdir(parents=True, exist_ok=True)
stats = []
if STATS_FILE.exists():
try:
with open(STATS_FILE) as f:
stats = json.load(f)
except (json.JSONDecodeError, IOError):
stats = []
stats.append(report.to_dict())
# Keep last 1000 entries
stats = stats[-1000:]
with open(STATS_FILE, "w") as f:
json.dump(stats, f, indent=2)
def show_status():
"""Show quality gate statistics."""
if not STATS_FILE.exists():
print("No quality stats found.")
return
with open(STATS_FILE) as f:
stats = json.load(f)
print(f"\nQuality Gate Stats — {len(stats)} runs")
print()
# Group by type
by_type = {}
for s in stats:
t = s.get("type", "unknown")
if t not in by_type:
by_type[t] = []
by_type[t].append(s)
for t, runs in sorted(by_type.items()):
total_entries = sum(r.get("total", 0) for r in runs)
total_passed = sum(r.get("passed", 0) for r in runs)
total_rejected = sum(r.get("rejected", 0) for r in runs)
avg_score = sum(r.get("score", 0) for r in runs) / len(runs) if runs else 0
print(f" {t:25} {len(runs):4} runs | {total_entries:6} entries | {total_rejected:4} rejected | avg score: {avg_score:.3f}")
def main():
import argparse
parser = argparse.ArgumentParser(description="Quality Gate for Pipeline Outputs")
parser.add_argument("--input", default=None, help="Input JSONL file")
parser.add_argument("--type", default=None, help="Entry type (training_pairs, scene_descriptions, knowledge, etc.)")
parser.add_argument("--dir", default=None, help="Process all JSONL files in directory")
parser.add_argument("--status", action="store_true", help="Show quality stats")
args = parser.parse_args()
if args.status:
show_status()
return
if args.dir:
for f in sorted(Path(args.dir).glob("*.jsonl")):
t = args.type or _infer_type(f.name)
report = run_gate(str(f), t)
_print_report(report)
elif args.input:
t = args.type or _infer_type(args.input)
report = run_gate(args.input, t)
_print_report(report)
sys.exit(0 if report.rejected == 0 else 1)
else:
parser.print_help()
def _infer_type(filename: str) -> str:
"""Infer entry type from filename."""
name = filename.lower()
if "scene" in name:
return "scene_descriptions"
if "training" in name or "pair" in name:
return "training_pairs"
if "knowledge" in name:
return "knowledge"
if "adversary" in name or "attack" in name:
return "adversary"
if "prompt" in name or "enhance" in name:
return "prompt_enhancement"
return "training_pairs" # default
def _print_report(report: GateReport):
"""Print a human-readable gate report."""
status = "PASS" if report.rejected == 0 else f"FAIL ({report.rejected} rejected)"
print(f" {report.file}: {status} | {report.passed}/{report.total} passed | score: {report.score:.3f}")
if __name__ == "__main__":
main()

View File

@@ -1,3 +1,4 @@
#!/usr/bin/env python3
import json
from hermes_tools import browser_navigate, browser_vision

View File

@@ -1,3 +1,4 @@
#!/usr/bin/env python3
import json
from hermes_tools import browser_navigate, browser_vision

View File

@@ -1,12 +1,884 @@
#!/usr/bin/env python3
"""
foundation_accessibility_audit.py — Multimodal Visual Accessibility Audit.
Analyzes web pages for WCAG 2.1 AA compliance using both programmatic checks
and vision model analysis. Screenshots pages, checks contrast ratios, detects
layout issues, validates alt text, and produces structured audit reports.
Usage:
# Audit a single page
python scripts/foundation_accessibility_audit.py --url https://timmyfoundation.org
# Audit multiple pages
python scripts/foundation_accessibility_audit.py --url https://timmyfoundation.org --pages /about /donate /blog
# With vision model analysis (Gemma 3)
python scripts/foundation_accessibility_audit.py --url https://timmyfoundation.org --vision
# Programmatic-only (no vision model needed)
python scripts/foundation_accessibility_audit.py --url https://timmyfoundation.org --programmatic
# Output as text report
python scripts/foundation_accessibility_audit.py --url https://timmyfoundation.org --format text
WCAG 2.1 AA Checks:
1.4.3 Contrast (Minimum) — text vs background ratio >= 4.5:1
1.4.6 Contrast (Enhanced) — ratio >= 7:1 for AAA
1.4.11 Non-text Contrast — UI components >= 3:1
1.3.1 Info and Relationships — heading hierarchy, landmarks
1.1.1 Non-text Content — alt text on images
2.4.1 Bypass Blocks — skip navigation link
2.4.2 Page Titled — meaningful <title>
2.4.6 Headings and Labels — descriptive headings
4.1.2 Name, Role, Value — ARIA labels on interactive elements
Refs: timmy-config#492, WCAG 2.1 AA
"""
from __future__ import annotations
import argparse
import base64
import colorsys
import json
from hermes_tools import browser_navigate, browser_vision
import os
import re
import subprocess
import sys
import tempfile
import urllib.error
import urllib.request
from dataclasses import dataclass, field, asdict
from enum import Enum
from pathlib import Path
from typing import Optional
from html.parser import HTMLParser
def audit_accessibility():
browser_navigate(url="https://timmyfoundation.org")
analysis = browser_vision(
question="Perform an accessibility audit. Check for: 1) Color contrast, 2) Font legibility, 3) Missing alt text for images. Provide a report with FAIL/PASS."
# === Configuration ===
OLLAMA_BASE = os.environ.get("OLLAMA_BASE_URL", "http://localhost:11434")
VISION_MODEL = os.environ.get("VISUAL_REVIEW_MODEL", "gemma3:12b")
DEFAULT_PAGES = ["/", "/about", "/donate", "/blog", "/contact"]
class Severity(str, Enum):
CRITICAL = "critical" # Blocks access entirely
MAJOR = "major" # Significant barrier
MINOR = "minor" # Inconvenience
PASS = "pass"
@dataclass
class A11yViolation:
"""A single accessibility violation."""
criterion: str # WCAG criterion (e.g. "1.4.3")
criterion_name: str # Human-readable name
severity: Severity = Severity.MINOR
element: str = "" # CSS selector or element description
description: str = "" # What's wrong
fix: str = "" # Suggested fix
source: str = "" # "programmatic" or "vision"
@dataclass
class A11yPageResult:
"""Audit result for a single page."""
url: str = ""
title: str = ""
score: int = 100
violations: list[A11yViolation] = field(default_factory=list)
passed_checks: list[str] = field(default_factory=list)
summary: str = ""
@dataclass
class A11yAuditReport:
"""Complete audit report across all pages."""
site: str = ""
pages_audited: int = 0
overall_score: int = 100
total_violations: int = 0
critical_violations: int = 0
major_violations: int = 0
page_results: list[A11yPageResult] = field(default_factory=list)
summary: str = ""
# === HTML Parser for Programmatic Checks ===
class A11yHTMLParser(HTMLParser):
"""Extract accessibility-relevant elements from HTML."""
def __init__(self):
super().__init__()
self.title = ""
self.images = [] # [{"src": ..., "alt": ...}]
self.headings = [] # [{"level": int, "text": ...}]
self.links = [] # [{"text": ..., "href": ...}]
self.inputs = [] # [{"type": ..., "label": ..., "id": ...}]
self.landmarks = [] # [{"tag": ..., "role": ...}]
self.skip_nav = False
self.lang = ""
self.in_title = False
self.in_heading = False
self.heading_level = 0
self.heading_text = ""
self.current_text = ""
def handle_starttag(self, tag, attrs):
attr_dict = dict(attrs)
if tag == "title":
self.in_title = True
elif tag == "html":
self.lang = attr_dict.get("lang", "")
elif tag in ("h1", "h2", "h3", "h4", "h5", "h6"):
self.in_heading = True
self.heading_level = int(tag[1])
self.heading_text = ""
elif tag == "img":
self.images.append({
"src": attr_dict.get("src", ""),
"alt": attr_dict.get("alt"),
"role": attr_dict.get("role", ""),
})
elif tag == "a":
self.links.append({
"href": attr_dict.get("href", ""),
"text": "",
"aria_label": attr_dict.get("aria-label", ""),
})
elif tag in ("input", "select", "textarea"):
self.inputs.append({
"tag": tag,
"type": attr_dict.get("type", "text"),
"id": attr_dict.get("id", ""),
"aria_label": attr_dict.get("aria-label", ""),
"aria_labelledby": attr_dict.get("aria-labelledby", ""),
})
elif tag in ("main", "nav", "header", "footer", "aside", "section", "form"):
self.landmarks.append({"tag": tag, "role": attr_dict.get("role", "")})
elif tag == "a" and ("skip" in attr_dict.get("href", "").lower() or
"skip" in attr_dict.get("class", "").lower()):
self.skip_nav = True
role = attr_dict.get("role", "")
if role in ("navigation", "main", "banner", "contentinfo", "complementary", "search"):
self.landmarks.append({"tag": tag, "role": role})
if role == "link" and "skip" in (attr_dict.get("aria-label", "") + attr_dict.get("href", "")).lower():
self.skip_nav = True
def handle_endtag(self, tag):
if tag == "title":
self.in_title = False
elif tag in ("h1", "h2", "h3", "h4", "h5", "h6"):
self.headings.append({"level": self.heading_level, "text": self.heading_text.strip()})
self.in_heading = False
elif tag == "a" and self.links:
self.links[-1]["text"] = self.current_text.strip()
self.current_text = ""
def handle_data(self, data):
if self.in_title:
self.title += data
if self.in_heading:
self.heading_text += data
self.current_text += data
# === Color/Contrast Utilities ===
def parse_color(color_str: str) -> Optional[tuple]:
"""Parse CSS color string to (r, g, b) tuple (0-255)."""
if not color_str:
return None
color_str = color_str.strip().lower()
# Named colors (subset)
named = {
"white": (255, 255, 255), "black": (0, 0, 0),
"red": (255, 0, 0), "green": (0, 128, 0), "blue": (0, 0, 255),
"gray": (128, 128, 128), "grey": (128, 128, 128),
"silver": (192, 192, 192), "yellow": (255, 255, 0),
"orange": (255, 165, 0), "purple": (128, 0, 128),
"transparent": None,
}
if color_str in named:
return named[color_str]
# #RRGGBB or #RGB
if color_str.startswith("#"):
hex_str = color_str[1:]
if len(hex_str) == 3:
hex_str = "".join(c * 2 for c in hex_str)
if len(hex_str) == 6:
try:
return tuple(int(hex_str[i:i+2], 16) for i in (0, 2, 4))
except ValueError:
return None
# rgb(r, g, b)
match = re.match(r"rgb\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*\)", color_str)
if match:
return tuple(int(match.group(i)) for i in (1, 2, 3))
# rgba(r, g, b, a)
match = re.match(r"rgba\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*,\s*[\d.]+\s*\)", color_str)
if match:
return tuple(int(match.group(i)) for i in (1, 2, 3))
return None
def relative_luminance(rgb: tuple) -> float:
"""Calculate relative luminance per WCAG 2.1 (sRGB)."""
def linearize(c):
c = c / 255.0
return c / 12.92 if c <= 0.04045 else ((c + 0.055) / 1.055) ** 2.4
r, g, b = [linearize(c) for c in rgb]
return 0.2126 * r + 0.7152 * g + 0.0722 * b
def contrast_ratio(color1: tuple, color2: tuple) -> float:
"""Calculate contrast ratio between two colors per WCAG 2.1."""
l1 = relative_luminance(color1)
l2 = relative_luminance(color2)
lighter = max(l1, l2)
darker = min(l1, l2)
return (lighter + 0.05) / (darker + 0.05)
# === Programmatic Checks ===
def check_page_title(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 2.4.2 — Page Titled."""
violations = []
title = parser.title.strip()
if not title:
violations.append(A11yViolation(
criterion="2.4.2", criterion_name="Page Titled",
severity=Severity.MAJOR,
element="<title>",
description="Page has no title or title is empty.",
fix="Add a meaningful <title> that describes the page purpose.",
source="programmatic"
))
elif len(title) < 5:
violations.append(A11yViolation(
criterion="2.4.2", criterion_name="Page Titled",
severity=Severity.MINOR,
element=f"<title>{title}</title>",
description=f"Page title is very short: '{title}'",
fix="Use a more descriptive title.",
source="programmatic"
))
return violations
def check_lang_attribute(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 3.1.1 — Language of Page."""
violations = []
if not parser.lang:
violations.append(A11yViolation(
criterion="3.1.1", criterion_name="Language of Page",
severity=Severity.MAJOR,
element="<html>",
description="Missing lang attribute on <html> element.",
fix="Add lang=\"en\" (or appropriate language code) to <html>.",
source="programmatic"
))
return violations
def check_images_alt_text(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 1.1.1 — Non-text Content."""
violations = []
for img in parser.images:
if img.get("role") == "presentation" or img.get("role") == "none":
continue # Decorative images are exempt
alt = img.get("alt")
src = img.get("src", "unknown")
if alt is None:
violations.append(A11yViolation(
criterion="1.1.1", criterion_name="Non-text Content",
severity=Severity.CRITICAL,
element=f"<img src=\"{src[:80]}\">",
description="Image missing alt attribute.",
fix="Add descriptive alt text, or alt=\"\" with role=\"presentation\" for decorative images.",
source="programmatic"
))
elif alt.strip() == "":
# Empty alt is OK only for decorative images
if img.get("role") not in ("presentation", "none"):
violations.append(A11yViolation(
criterion="1.1.1", criterion_name="Non-text Content",
severity=Severity.MINOR,
element=f"<img src=\"{src[:80]}\" alt=\"\">",
description="Empty alt text — ensure this image is decorative.",
fix="If decorative, add role=\"presentation\". If meaningful, add descriptive alt text.",
source="programmatic"
))
return violations
def check_heading_hierarchy(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 1.3.1 — Info and Relationships (heading hierarchy)."""
violations = []
if not parser.headings:
violations.append(A11yViolation(
criterion="1.3.1", criterion_name="Info and Relationships",
severity=Severity.MAJOR,
element="document",
description="No headings found on page.",
fix="Add proper heading hierarchy starting with <h1>.",
source="programmatic"
))
return violations
# Check for H1
h1s = [h for h in parser.headings if h["level"] == 1]
if not h1s:
violations.append(A11yViolation(
criterion="1.3.1", criterion_name="Info and Relationships",
severity=Severity.MAJOR,
element="document",
description="No <h1> heading found.",
fix="Add a single <h1> as the main page heading.",
source="programmatic"
))
elif len(h1s) > 1:
violations.append(A11yViolation(
criterion="1.3.1", criterion_name="Info and Relationships",
severity=Severity.MINOR,
element="document",
description=f"Multiple <h1> headings found ({len(h1s)}).",
fix="Use a single <h1> per page for the main heading.",
source="programmatic"
))
# Check hierarchy skips
prev_level = 0
for h in parser.headings:
level = h["level"]
if level > prev_level + 1 and prev_level > 0:
violations.append(A11yViolation(
criterion="1.3.1", criterion_name="Info and Relationships",
severity=Severity.MINOR,
element=f"<h{level}>{h['text'][:50]}</h{level}>",
description=f"Heading level skipped: h{prev_level} → h{level}",
fix=f"Use <h{prev_level + 1}> instead, or fill the gap.",
source="programmatic"
))
prev_level = level
return violations
def check_landmarks(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 1.3.1 — Landmarks and structure."""
violations = []
roles = {lm.get("role", "") for lm in parser.landmarks}
tags = {lm.get("tag", "") for lm in parser.landmarks}
has_main = "main" in roles or "main" in tags
has_nav = "navigation" in roles or "nav" in tags
if not has_main:
violations.append(A11yViolation(
criterion="1.3.1", criterion_name="Info and Relationships",
severity=Severity.MAJOR,
element="document",
description="No <main> landmark found.",
fix="Wrap the main content in a <main> element.",
source="programmatic"
))
if not has_nav:
violations.append(A11yViolation(
criterion="1.3.1", criterion_name="Info and Relationships",
severity=Severity.MINOR,
element="document",
description="No <nav> landmark found.",
fix="Wrap navigation in a <nav> element.",
source="programmatic"
))
return violations
def check_skip_nav(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 2.4.1 — Bypass Blocks."""
violations = []
if not parser.skip_nav:
# Also check links for "skip" text
has_skip_link = any("skip" in l.get("text", "").lower() for l in parser.links)
if not has_skip_link:
violations.append(A11yViolation(
criterion="2.4.1", criterion_name="Bypass Blocks",
severity=Severity.MAJOR,
element="document",
description="No skip navigation link found.",
fix="Add a 'Skip to main content' link as the first focusable element.",
source="programmatic"
))
return violations
def check_form_labels(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 4.1.2 — Name, Role, Value (form inputs)."""
violations = []
for inp in parser.inputs:
if inp["type"] in ("hidden", "submit", "button", "reset", "image"):
continue
has_label = bool(inp.get("aria_label") or inp.get("aria_labelledby") or inp.get("id"))
if not has_label:
violations.append(A11yViolation(
criterion="4.1.2", criterion_name="Name, Role, Value",
severity=Severity.MAJOR,
element=f"<{inp['tag']} type=\"{inp['type']}\">",
description="Form input has no associated label or aria-label.",
fix="Add a <label for=\"...\"> or aria-label attribute.",
source="programmatic"
))
return violations
def check_link_text(parser: A11yHTMLParser) -> list[A11yViolation]:
"""WCAG 2.4.4 — Link Purpose."""
violations = []
for link in parser.links:
text = (link.get("text", "") or link.get("aria_label", "")).strip().lower()
href = link.get("href", "")
if not text:
violations.append(A11yViolation(
criterion="2.4.4", criterion_name="Link Purpose",
severity=Severity.MAJOR,
element=f"<a href=\"{href[:60]}\">",
description="Link has no accessible text.",
fix="Add visible text content or aria-label to the link.",
source="programmatic"
))
elif text in ("click here", "read more", "here", "more", "link"):
violations.append(A11yViolation(
criterion="2.4.4", criterion_name="Link Purpose",
severity=Severity.MINOR,
element=f"<a href=\"{href[:60]}\">{text}</a>",
description=f"Non-descriptive link text: '{text}'",
fix="Use descriptive text that explains the link destination.",
source="programmatic"
))
return violations
def run_programmatic_checks(html: str) -> list[A11yViolation]:
"""Run all programmatic accessibility checks on HTML content."""
parser = A11yHTMLParser()
try:
parser.feed(html)
except Exception:
pass
violations = []
violations.extend(check_page_title(parser))
violations.extend(check_lang_attribute(parser))
violations.extend(check_images_alt_text(parser))
violations.extend(check_heading_hierarchy(parser))
violations.extend(check_landmarks(parser))
violations.extend(check_skip_nav(parser))
violations.extend(check_form_labels(parser))
violations.extend(check_link_text(parser))
return violations
# === Vision Model Checks ===
A11Y_VISION_PROMPT = """You are a WCAG 2.1 AA accessibility auditor. Analyze this screenshot of a web page.
Check for these specific issues:
1. COLOR CONTRAST: Are text colors sufficiently different from their backgrounds?
- Normal text needs 4.5:1 contrast ratio
- Large text (18pt+) needs 3:1
- UI components need 3:1
List any text or UI elements where contrast looks insufficient.
2. FONT LEGIBILITY: Is text readable?
- Font size >= 12px for body text
- Line height >= 1.5 for body text
- No text in images (should be real text)
3. LAYOUT ISSUES: Is the layout accessible?
- Touch targets >= 44x44px
- Content not cut off or overlapping
- Logical reading order visible
- No horizontal scrolling at standard widths
4. FOCUS INDICATORS: Can you see which element has focus?
- Interactive elements should have visible focus rings
5. COLOR ALONE: Is information conveyed only by color?
- Errors/warnings should not rely solely on red/green
Respond as JSON:
{
"violations": [
{
"criterion": "1.4.3",
"criterion_name": "Contrast (Minimum)",
"severity": "critical|major|minor",
"element": "description of element",
"description": "what's wrong",
"fix": "how to fix"
}
],
"passed_checks": ["list of things that look good"],
"overall_score": 0-100,
"summary": "brief summary"
}"""
def run_vision_check(screenshot_path: str, model: str = VISION_MODEL) -> list[A11yViolation]:
"""Run vision model accessibility check on a screenshot."""
try:
b64 = base64.b64encode(Path(screenshot_path).read_bytes()).decode()
payload = json.dumps({
"model": model,
"messages": [{"role": "user", "content": [
{"type": "text", "text": A11Y_VISION_PROMPT},
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{b64}"}}
]}],
"stream": False,
"options": {"temperature": 0.1}
}).encode()
req = urllib.request.Request(
f"{OLLAMA_BASE}/api/chat",
data=payload,
headers={"Content-Type": "application/json"}
)
with urllib.request.urlopen(req, timeout=120) as resp:
result = json.loads(resp.read())
content = result.get("message", {}).get("content", "")
# Parse response
parsed = _parse_json_response(content)
violations = []
for v in parsed.get("violations", []):
violations.append(A11yViolation(
criterion=v.get("criterion", ""),
criterion_name=v.get("criterion_name", ""),
severity=Severity(v.get("severity", "minor")),
element=v.get("element", ""),
description=v.get("description", ""),
fix=v.get("fix", ""),
source="vision"
))
return violations
except Exception as e:
print(f" Vision check failed: {e}", file=sys.stderr)
return []
def _parse_json_response(text: str) -> dict:
"""Extract JSON from potentially messy vision response."""
cleaned = text.strip()
if cleaned.startswith("```"):
lines = cleaned.split("\n")[1:]
if lines and lines[-1].strip() == "```":
lines = lines[:-1]
cleaned = "\n".join(lines)
try:
return json.loads(cleaned)
except json.JSONDecodeError:
start = cleaned.find("{")
end = cleaned.rfind("}")
if start >= 0 and end > start:
try:
return json.loads(cleaned[start:end + 1])
except json.JSONDecodeError:
pass
return {}
# === Page Fetching ===
def fetch_page(url: str) -> Optional[str]:
"""Fetch HTML content of a page."""
try:
req = urllib.request.Request(url, headers={"User-Agent": "A11yAudit/1.0"})
with urllib.request.urlopen(req, timeout=30) as resp:
return resp.read().decode("utf-8", errors="replace")
except Exception as e:
print(f" Failed to fetch {url}: {e}", file=sys.stderr)
return None
def take_screenshot(url: str, output_path: str, width: int = 1280, height: int = 900) -> bool:
"""Take a screenshot using Playwright or curl-based headless capture."""
# Try Playwright first
try:
script = f"""
from playwright.sync_api import sync_playwright
with sync_playwright() as p:
browser = p.chromium.launch(headless=True)
page = browser.new_page(viewport={{"width": {width}, "height": {height}}})
page.goto("{url}", wait_until="networkidle", timeout=30000)
page.screenshot(path="{output_path}", full_page=True)
browser.close()
"""
result = subprocess.run(
["python3", "-c", script],
capture_output=True, text=True, timeout=60
)
if result.returncode == 0 and Path(output_path).exists():
return True
except Exception:
pass
# Try curl + wkhtmltoimage
try:
result = subprocess.run(
["wkhtmltoimage", "--width", str(width), "--quality", "90", url, output_path],
capture_output=True, text=True, timeout=30
)
if result.returncode == 0 and Path(output_path).exists():
return True
except Exception:
pass
return False
# === Audit Logic ===
def audit_page(url: str, use_vision: bool = False, model: str = VISION_MODEL) -> A11yPageResult:
"""Run a full accessibility audit on a single page."""
result = A11yPageResult(url=url)
# Fetch HTML
html = fetch_page(url)
if not html:
result.summary = f"Failed to fetch {url}"
result.score = 0
return result
# Extract title
title_match = re.search(r"<title[^>]*>(.*?)</title>", html, re.IGNORECASE | re.DOTALL)
result.title = title_match.group(1).strip() if title_match else ""
# Run programmatic checks
prog_violations = run_programmatic_checks(html)
result.violations.extend(prog_violations)
# Track passed checks
criteria_checked = {
"2.4.2": "Page Titled",
"3.1.1": "Language of Page",
"1.1.1": "Non-text Content",
"1.3.1": "Info and Relationships",
"2.4.1": "Bypass Blocks",
"4.1.2": "Name, Role, Value",
"2.4.4": "Link Purpose",
}
violated_criteria = {v.criterion for v in result.violations}
for criterion, name in criteria_checked.items():
if criterion not in violated_criteria:
result.passed_checks.append(f"{criterion} {name}")
# Vision check (optional)
if use_vision:
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp:
screenshot_path = tmp.name
try:
print(f" Taking screenshot of {url}...", file=sys.stderr)
if take_screenshot(url, screenshot_path):
print(f" Running vision analysis...", file=sys.stderr)
vision_violations = run_vision_check(screenshot_path, model)
result.violations.extend(vision_violations)
result.passed_checks.append("Vision model analysis completed")
else:
result.passed_checks.append("Screenshot unavailable — vision check skipped")
finally:
Path(screenshot_path).unlink(missing_ok=True)
# Calculate score
criticals = sum(1 for v in result.violations if v.severity == Severity.CRITICAL)
majors = sum(1 for v in result.violations if v.severity == Severity.MAJOR)
minors = sum(1 for v in result.violations if v.severity == Severity.MINOR)
result.score = max(0, 100 - (criticals * 25) - (majors * 10) - (minors * 3))
# Summary
if not result.violations:
result.summary = f"All programmatic checks passed for {url}"
else:
result.summary = (
f"{len(result.violations)} issue(s) found: "
f"{criticals} critical, {majors} major, {minors} minor"
)
return result
def audit_site(base_url: str, pages: list[str], use_vision: bool = False,
model: str = VISION_MODEL) -> A11yAuditReport:
"""Audit multiple pages of a site."""
report = A11yAuditReport(site=base_url)
for path in pages:
url = base_url.rstrip("/") + path if not path.startswith("http") else path
print(f"Auditing: {url}", file=sys.stderr)
result = audit_page(url, use_vision, model)
report.page_results.append(result)
report.pages_audited = len(report.page_results)
report.total_violations = sum(len(p.violations) for p in report.page_results)
report.critical_violations = sum(
sum(1 for v in p.violations if v.severity == Severity.CRITICAL)
for p in report.page_results
)
report.major_violations = sum(
sum(1 for v in p.violations if v.severity == Severity.MAJOR)
for p in report.page_results
)
return {"status": "PASS" if "PASS" in analysis.upper() else "FAIL", "analysis": analysis}
if __name__ == '__main__':
print(json.dumps(audit_accessibility(), indent=2))
if report.page_results:
report.overall_score = sum(p.score for p in report.page_results) // len(report.page_results)
report.summary = (
f"Audited {report.pages_audited} pages. "
f"Overall score: {report.overall_score}/100. "
f"{report.total_violations} total issues: "
f"{report.critical_violations} critical, {report.major_violations} major."
)
return report
# === Output Formatting ===
def format_report(report: A11yAuditReport, fmt: str = "json") -> str:
"""Format the audit report."""
if fmt == "json":
data = {
"site": report.site,
"pages_audited": report.pages_audited,
"overall_score": report.overall_score,
"total_violations": report.total_violations,
"critical_violations": report.critical_violations,
"major_violations": report.major_violations,
"summary": report.summary,
"pages": []
}
for page in report.page_results:
page_data = {
"url": page.url,
"title": page.title,
"score": page.score,
"violations": [asdict(v) for v in page.violations],
"passed_checks": page.passed_checks,
"summary": page.summary,
}
# Convert severity enum to string
for v in page_data["violations"]:
if hasattr(v["severity"], "value"):
v["severity"] = v["severity"].value
data["pages"].append(page_data)
return json.dumps(data, indent=2)
elif fmt == "text":
lines = []
lines.append("=" * 60)
lines.append(" WEB ACCESSIBILITY AUDIT REPORT")
lines.append("=" * 60)
lines.append(f" Site: {report.site}")
lines.append(f" Pages audited: {report.pages_audited}")
lines.append(f" Overall score: {report.overall_score}/100")
lines.append(f" Issues: {report.total_violations} total "
f"({report.critical_violations} critical, {report.major_violations} major)")
lines.append("")
for page in report.page_results:
lines.append(f" ── {page.url} ──")
lines.append(f" Title: {page.title}")
lines.append(f" Score: {page.score}/100")
lines.append("")
if page.violations:
lines.append(f" Violations ({len(page.violations)}):")
for v in page.violations:
sev_icon = {"critical": "🔴", "major": "🟡", "minor": "🔵"}.get(
v.severity.value if hasattr(v.severity, "value") else str(v.severity), ""
)
lines.append(f" {sev_icon} [{v.criterion}] {v.criterion_name}")
lines.append(f" Element: {v.element}")
lines.append(f" Issue: {v.description}")
lines.append(f" Fix: {v.fix}")
lines.append(f" Source: {v.source}")
lines.append("")
else:
lines.append(" ✓ No violations found")
lines.append("")
if page.passed_checks:
lines.append(f" Passed: {', '.join(page.passed_checks)}")
lines.append("")
lines.append("=" * 60)
lines.append(f" Summary: {report.summary}")
lines.append("=" * 60)
return "\n".join(lines)
else:
raise ValueError(f"Unknown format: {fmt}")
# === CLI ===
def main():
parser = argparse.ArgumentParser(
description="Visual Accessibility Audit — WCAG 2.1 AA compliance checker",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
%(prog)s --url https://timmyfoundation.org
%(prog)s --url https://timmyfoundation.org --pages /about /donate
%(prog)s --url https://timmyfoundation.org --vision
%(prog)s --url https://timmyfoundation.org --format text
"""
)
parser.add_argument("--url", required=True, help="Base URL to audit")
parser.add_argument("--pages", nargs="*", default=DEFAULT_PAGES,
help="Paths to audit (default: / /about /donate /blog /contact)")
parser.add_argument("--vision", action="store_true",
help="Include vision model analysis (requires Ollama)")
parser.add_argument("--model", default=VISION_MODEL,
help=f"Vision model (default: {VISION_MODEL})")
parser.add_argument("--format", choices=["json", "text"], default="json",
help="Output format")
parser.add_argument("--output", "-o", help="Output file (default: stdout)")
args = parser.parse_args()
report = audit_site(args.url, args.pages, use_vision=args.vision, model=args.model)
output = format_report(report, args.format)
if args.output:
Path(args.output).write_text(output)
print(f"Report written to {args.output}", file=sys.stderr)
else:
print(output)
# Exit code: non-zero if critical violations
if report.critical_violations > 0:
sys.exit(1)
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,32 @@
#!/usr/bin/env python3
import json, os
songs = [
{"t":"Thunder Road","a":"Heartland","m":["hope","anticipation","energy","triumph","nostalgia","urgency","passion","defiance","release","catharsis"]},
{"t":"Black Dog Howl","a":"Rust & Wire","m":["despair","anger","frenzy","exhaustion","resignation","grief","numbness","rage","acceptance","silence"]},
{"t":"Satellite Hearts","a":"Neon Circuit","m":["wonder","isolation","longing","connection","euphoria","confusion","clarity","tenderness","urgency","bittersweet"]},
{"t":"Concrete Garden","a":"Streetlight Prophet","m":["oppression","resilience","anger","beauty","defiance","community","joy","struggle","growth","hope"]},
{"t":"Gravity Well","a":"Void Walker","m":["dread","fascination","surrender","awe","terror","peace","disorientation","acceptance","transcendence","emptiness"]},
{"t":"Rust Belt Lullaby","a":"Iron & Ember","m":["nostalgia","sadness","tenderness","loss","beauty","resignation","love","weariness","quiet hope","peace"]},
{"t":"Wildfire Sermon","a":"Prophet Ash","m":["fury","ecstasy","chaos","joy","destruction","creation","warning","invitation","abandon","rebirth"]},
{"t":"Midnight Transmission","a":"Frequency Ghost","m":["mystery","loneliness","curiosity","connection","paranoia","intimacy","urgency","disconnection","searching","haunting"]},
{"t":"Crown of Thorns","a":"Velvet Guillotine","m":["seduction","power","cruelty","beauty","danger","vulnerability","fury","grace","revenge","mercy"]},
{"t":"Apartment 4B","a":"Wallpaper & Wire","m":["claustrophobia","routine","desperation","fantasy","breakthrough","freedom","fear","joy","grounding","home"]},
]
beats = []
for s in songs:
for i in range(10):
beats.append({"song": s["t"], "artist": s["a"], "beat": i+1,
"timestamp": f"{i*30//60}:{(i*30)%60:02d}", "duration": "30s",
"lyric_line": f"[Beat {i+1}]", "scene": {"mood": s["m"][i], "colors": ["placeholder"],
"composition": ["wide","close","OTS","low","high","dutch","symmetric","thirds","xwide","medium"][i],
"camera": ["static","pan","dolly-in","dolly-out","handheld","steadicam","zoom","crane","track","tilt"][i],
"description": f"[{s['m'][i]} scene]"}})
out = os.path.expanduser("~/.hermes/training-data/scene-descriptions-rock.jsonl")
os.makedirs(os.path.dirname(out), exist_ok=True)
with open(out, "w") as f:
for b in beats:
f.write(json.dumps(b) + "\n")
print(f"Generated {len(beats)} beats")

View File

@@ -1,12 +1,599 @@
#!/usr/bin/env python3
"""
matrix_glitch_detect.py — 3D World Visual Artifact Detection for The Matrix.
Scans screenshots or live pages for visual glitches: floating assets, z-fighting,
texture pop-in, clipping, broken meshes, lighting artifacts. Outputs structured
JSON, text, or standalone HTML report with annotated screenshots.
Usage:
# Scan a screenshot
python scripts/matrix_glitch_detect.py --image screenshot.png
# Scan with vision model
python scripts/matrix_glitch_detect.py --image screenshot.png --vision
# HTML report
python scripts/matrix_glitch_detect.py --image screenshot.png --html report.html
# Scan live Matrix page
python scripts/matrix_glitch_detect.py --url https://matrix.alexanderwhitestone.com
# Batch scan a directory
python scripts/matrix_glitch_detect.py --batch ./screenshots/ --html batch-report.html
Refs: timmy-config#491, #541, #543, #544
"""
from __future__ import annotations
import argparse
import base64
import html as html_module
import json
from hermes_tools import browser_navigate, browser_vision
import os
import sys
import time
import urllib.error
import urllib.request
from dataclasses import dataclass, field, asdict
from datetime import datetime
from enum import Enum
from pathlib import Path
from typing import Optional
def detect_glitches():
browser_navigate(url="https://matrix.alexanderwhitestone.com")
analysis = browser_vision(
question="Scan the 3D world for visual artifacts, floating assets, or z-fighting. List all coordinates/descriptions of glitches found. Provide a PASS/FAIL."
# === Configuration ===
OLLAMA_BASE = os.environ.get("OLLAMA_BASE_URL", "http://localhost:11434")
VISION_MODEL = os.environ.get("VISUAL_REVIEW_MODEL", "gemma3:12b")
class Severity(str, Enum):
CRITICAL = "critical"
MAJOR = "major"
MINOR = "minor"
COSMETIC = "cosmetic"
@dataclass
class Glitch:
"""A single detected visual artifact."""
type: str = "" # floating_asset, z_fighting, texture_pop, clipping, lighting, mesh_break
severity: Severity = Severity.MINOR
region: str = "" # "upper-left", "center", "bottom-right", or coordinates
description: str = ""
confidence: float = 0.0 # 0.0-1.0
source: str = "" # "programmatic", "vision", "pixel_analysis"
@dataclass
class GlitchReport:
"""Complete glitch detection report."""
source: str = "" # file path or URL
timestamp: str = ""
status: str = "PASS" # PASS, WARN, FAIL
score: int = 100
glitches: list[Glitch] = field(default_factory=list)
summary: str = ""
model_used: str = ""
width: int = 0
height: int = 0
# === Programmatic Analysis ===
def analyze_pixels(image_path: str) -> list[Glitch]:
"""Programmatic pixel analysis for common 3D glitches."""
glitches = []
try:
from PIL import Image
img = Image.open(image_path).convert("RGB")
w, h = img.size
pixels = img.load()
# Check for solid-color regions (render failure)
corner_colors = [
pixels[0, 0], pixels[w-1, 0], pixels[0, h-1], pixels[w-1, h-1]
]
if all(c == corner_colors[0] for c in corner_colors):
# All corners same color — check if it's black (render failure)
if corner_colors[0] == (0, 0, 0):
glitches.append(Glitch(
type="render_failure",
severity=Severity.CRITICAL,
region="entire frame",
description="Entire frame is black — 3D scene failed to render",
confidence=0.9,
source="pixel_analysis"
))
# Check for horizontal tearing lines
tear_count = 0
for y in range(0, h, max(1, h // 20)):
row_start = pixels[0, y]
same_count = sum(1 for x in range(w) if pixels[x, y] == row_start)
if same_count > w * 0.95:
tear_count += 1
if tear_count > 3:
glitches.append(Glitch(
type="horizontal_tear",
severity=Severity.MAJOR,
region=f"{tear_count} lines",
description=f"Horizontal tearing detected — {tear_count} mostly-solid scanlines",
confidence=0.7,
source="pixel_analysis"
))
# Check for extreme brightness variance (lighting artifacts)
import statistics
brightness_samples = []
for y in range(0, h, max(1, h // 50)):
for x in range(0, w, max(1, w // 50)):
r, g, b = pixels[x, y]
brightness_samples.append(0.299 * r + 0.587 * g + 0.114 * b)
if brightness_samples:
stdev = statistics.stdev(brightness_samples)
if stdev > 100:
glitches.append(Glitch(
type="lighting",
severity=Severity.MINOR,
region="global",
description=f"Extreme brightness variance (stdev={stdev:.0f}) — possible lighting artifacts",
confidence=0.5,
source="pixel_analysis"
))
except ImportError:
pass # PIL not available
except Exception as e:
pass
return glitches
# === Vision Analysis ===
GLITCH_VISION_PROMPT = """You are a 3D world QA engineer. Analyze this screenshot from a Three.js 3D world (The Matrix) for visual glitches and artifacts.
Look for these specific issues:
1. FLOATING ASSETS: Objects hovering above surfaces where they should rest. Look for shadows detached from objects.
2. Z-FIGHTING: Flickering or shimmering surfaces where two polygons overlap at the same depth. Usually appears as striped or dithered patterns.
3. TEXTURE POP-IN: Low-resolution textures that haven't loaded, or textures that suddenly change quality between frames.
4. CLIPPING: Objects passing through walls, floors, or other objects. Characters partially inside geometry.
5. LIGHTING ARTIFACTS: Hard light seams, black patches, overexposed areas, lights not illuminating correctly.
6. MESH BREAKS: Visible seams in geometry, missing faces on 3D objects, holes in surfaces.
7. RENDER FAILURE: Black areas where geometry should be, missing skybox, incomplete frame rendering.
8. UI OVERLAP: UI elements overlapping 3D viewport incorrectly.
Respond as JSON:
{
"glitches": [
{
"type": "floating_asset|z_fighting|texture_pop|clipping|lighting|mesh_break|render_failure|ui_overlap",
"severity": "critical|major|minor|cosmetic",
"region": "description of where",
"description": "detailed description of the artifact",
"confidence": 0.0-1.0
}
],
"overall_quality": 0-100,
"summary": "brief assessment"
}"""
def run_vision_analysis(image_path: str, model: str = VISION_MODEL) -> tuple[list[Glitch], int]:
"""Run vision model glitch analysis."""
try:
b64 = base64.b64encode(Path(image_path).read_bytes()).decode()
payload = json.dumps({
"model": model,
"messages": [{"role": "user", "content": [
{"type": "text", "text": GLITCH_VISION_PROMPT},
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{b64}"}}
]}],
"stream": False,
"options": {"temperature": 0.1}
}).encode()
req = urllib.request.Request(
f"{OLLAMA_BASE}/api/chat",
data=payload,
headers={"Content-Type": "application/json"}
)
with urllib.request.urlopen(req, timeout=120) as resp:
result = json.loads(resp.read())
content = result.get("message", {}).get("content", "")
parsed = _parse_json_response(content)
glitches = []
for g in parsed.get("glitches", []):
glitches.append(Glitch(
type=g.get("type", "unknown"),
severity=Severity(g.get("severity", "minor")),
region=g.get("region", ""),
description=g.get("description", ""),
confidence=float(g.get("confidence", 0.5)),
source="vision"
))
return glitches, parsed.get("overall_quality", 80)
except Exception as e:
print(f" Vision analysis failed: {e}", file=sys.stderr)
return [], 50
def _parse_json_response(text: str) -> dict:
cleaned = text.strip()
if cleaned.startswith("```"):
lines = cleaned.split("\n")[1:]
if lines and lines[-1].strip() == "```":
lines = lines[:-1]
cleaned = "\n".join(lines)
try:
return json.loads(cleaned)
except json.JSONDecodeError:
start = cleaned.find("{")
end = cleaned.rfind("}")
if start >= 0 and end > start:
try:
return json.loads(cleaned[start:end + 1])
except json.JSONDecodeError:
pass
return {}
# === Screenshot Capture ===
def capture_screenshot(url: str, output_path: str) -> bool:
"""Take a screenshot of a URL."""
try:
script = f"""
from playwright.sync_api import sync_playwright
with sync_playwright() as p:
browser = p.chromium.launch(headless=True)
page = browser.new_page(viewport={{"width": 1280, "height": 720}})
page.goto("{url}", wait_until="networkidle", timeout=30000)
page.wait_for_timeout(3000)
page.screenshot(path="{output_path}")
browser.close()
"""
result = subprocess.run(["python3", "-c", script], capture_output=True, text=True, timeout=60)
return result.returncode == 0 and Path(output_path).exists()
except Exception:
return False
# === Detection Logic ===
def detect_glitches(image_path: str, use_vision: bool = False,
model: str = VISION_MODEL) -> GlitchReport:
"""Run full glitch detection on an image."""
report = GlitchReport(
source=image_path,
timestamp=datetime.now().isoformat(),
model_used=model if use_vision else "none"
)
return {"status": "PASS" if "PASS" in analysis.upper() else "FAIL", "analysis": analysis}
if __name__ == '__main__':
print(json.dumps(detect_glitches(), indent=2))
if not Path(image_path).exists():
report.status = "FAIL"
report.summary = f"File not found: {image_path}"
report.score = 0
return report
# Get image dimensions
try:
from PIL import Image
img = Image.open(image_path)
report.width, report.height = img.size
except Exception:
pass
# Programmatic analysis
prog_glitches = analyze_pixels(image_path)
report.glitches.extend(prog_glitches)
# Vision analysis
if use_vision:
print(f" Running vision analysis on {image_path}...", file=sys.stderr)
vision_glitches, quality = run_vision_analysis(image_path, model)
report.glitches.extend(vision_glitches)
report.score = quality
else:
# Score based on programmatic results
criticals = sum(1 for g in report.glitches if g.severity == Severity.CRITICAL)
majors = sum(1 for g in report.glitches if g.severity == Severity.MAJOR)
report.score = max(0, 100 - criticals * 40 - majors * 15)
# Determine status
criticals = sum(1 for g in report.glitches if g.severity == Severity.CRITICAL)
majors = sum(1 for g in report.glitches if g.severity == Severity.MAJOR)
if criticals > 0:
report.status = "FAIL"
elif majors > 0 or report.score < 70:
report.status = "WARN"
else:
report.status = "PASS"
report.summary = (
f"{report.status}: {len(report.glitches)} glitch(es) found "
f"({criticals} critical, {majors} major), score {report.score}/100"
)
return report
# === HTML Report Generator ===
def generate_html_report(reports: list[GlitchReport], title: str = "Glitch Detection Report") -> str:
"""Generate a standalone HTML report with annotated details."""
total_glitches = sum(len(r.glitches) for r in reports)
total_criticals = sum(sum(1 for g in r.glitches if g.severity == Severity.CRITICAL) for r in reports)
avg_score = sum(r.score for r in reports) // max(1, len(reports))
if total_criticals > 0:
overall_verdict = "FAIL"
verdict_color = "#f44336"
elif any(r.status == "WARN" for r in reports):
overall_verdict = "WARN"
verdict_color = "#ff9800"
else:
overall_verdict = "PASS"
verdict_color = "#4caf50"
# Build HTML
parts = []
parts.append(f"""<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>{html_module.escape(title)}</title>
<style>
*{{margin:0;padding:0;box-sizing:border-box}}
body{{font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,monospace;background:#0a0a14;color:#c0c0d0;font-size:13px;line-height:1.5}}
.container{{max-width:1000px;margin:0 auto;padding:20px}}
header{{text-align:center;padding:24px 0;border-bottom:1px solid #1a1a2e;margin-bottom:24px}}
header h1{{font-size:20px;font-weight:300;letter-spacing:3px;color:#4a9eff;margin-bottom:8px}}
.verdict{{display:inline-block;padding:6px 20px;border-radius:4px;font-size:14px;font-weight:700;letter-spacing:2px;color:#fff;background:{verdict_color}}}
.stats{{display:flex;gap:16px;justify-content:center;margin:16px 0;flex-wrap:wrap}}
.stat{{background:#0e0e1a;border:1px solid #1a1a2e;border-radius:4px;padding:8px 16px;text-align:center}}
.stat .val{{font-size:20px;font-weight:700;color:#4a9eff}}
.stat .lbl{{font-size:9px;color:#666;text-transform:uppercase;letter-spacing:1px}}
.score-gauge{{width:120px;height:120px;margin:0 auto 16px;position:relative}}
.score-gauge svg{{transform:rotate(-90deg)}}
.score-gauge .score-text{{position:absolute;top:50%;left:50%;transform:translate(-50%,-50%);font-size:28px;font-weight:700}}
.report-card{{background:#0e0e1a;border:1px solid #1a1a2e;border-radius:6px;margin-bottom:16px;overflow:hidden}}
.report-header{{padding:12px 16px;border-bottom:1px solid #1a1a2e;display:flex;justify-content:space-between;align-items:center}}
.report-header .source{{color:#4a9eff;font-weight:600;word-break:break-all}}
.report-header .status-badge{{padding:2px 10px;border-radius:3px;font-size:11px;font-weight:700;color:#fff}}
.status-pass{{background:#4caf50}}
.status-warn{{background:#ff9800}}
.status-fail{{background:#f44336}}
.screenshot{{text-align:center;padding:12px;background:#080810}}
.screenshot img{{max-width:100%;max-height:400px;border:1px solid #1a1a2e;border-radius:4px}}
.glitch-list{{padding:12px 16px}}
.glitch-item{{padding:8px 0;border-bottom:1px solid #111;display:flex;gap:12px;align-items:flex-start}}
.glitch-item:last-child{{border-bottom:none}}
.severity-dot{{width:8px;height:8px;border-radius:50%;margin-top:5px;flex-shrink:0}}
.sev-critical{{background:#f44336}}
.sev-major{{background:#ff9800}}
.sev-minor{{background:#2196f3}}
.sev-cosmetic{{background:#666}}
.glitch-detail{{flex:1}}
.glitch-type{{color:#ffd700;font-weight:600;font-size:11px;text-transform:uppercase;letter-spacing:1px}}
.glitch-desc{{color:#aaa;font-size:12px;margin-top:2px}}
.glitch-meta{{color:#555;font-size:10px;margin-top:2px}}
.no-glitches{{color:#4caf50;text-align:center;padding:20px;font-style:italic}}
footer{{text-align:center;padding:16px;color:#444;font-size:10px;border-top:1px solid #1a1a2e;margin-top:24px}}
</style>
</head>
<body>
<div class="container">
<header>
<h1>{html_module.escape(title)}</h1>
<div class="verdict">{overall_verdict}</div>
<div class="stats">
<div class="stat"><div class="val">{len(reports)}</div><div class="lbl">Screenshots</div></div>
<div class="stat"><div class="val">{total_glitches}</div><div class="lbl">Glitches</div></div>
<div class="stat"><div class="val">{total_criticals}</div><div class="lbl">Critical</div></div>
<div class="stat"><div class="val">{avg_score}</div><div class="lbl">Avg Score</div></div>
</div>
</header>
""")
# Score gauge
score_color = "#4caf50" if avg_score >= 80 else "#ff9800" if avg_score >= 60 else "#f44336"
circumference = 2 * 3.14159 * 50
dash_offset = circumference * (1 - avg_score / 100)
parts.append(f"""
<div class="score-gauge">
<svg width="120" height="120" viewBox="0 0 120 120">
<circle cx="60" cy="60" r="50" fill="none" stroke="#1a1a2e" stroke-width="8"/>
<circle cx="60" cy="60" r="50" fill="none" stroke="{score_color}" stroke-width="8"
stroke-dasharray="{circumference}" stroke-dashoffset="{dash_offset}" stroke-linecap="round"/>
</svg>
<div class="score-text" style="color:{score_color}">{avg_score}</div>
</div>
""")
# Per-screenshot reports
for i, report in enumerate(reports):
status_class = f"status-{report.status.lower()}"
source_name = Path(report.source).name if report.source else f"Screenshot {i+1}"
# Inline screenshot as base64
img_tag = ""
if report.source and Path(report.source).exists():
try:
b64 = base64.b64encode(Path(report.source).read_bytes()).decode()
ext = Path(report.source).suffix.lower()
mime = "image/png" if ext == ".png" else "image/jpeg" if ext in (".jpg", ".jpeg") else "image/webp"
img_tag = f'<img src="data:{mime};base64,{b64}" alt="Screenshot">'
except Exception:
img_tag = '<div style="color:#666;padding:40px">Screenshot unavailable</div>'
else:
img_tag = '<div style="color:#666;padding:40px">No screenshot</div>'
parts.append(f"""
<div class="report-card">
<div class="report-header">
<span class="source">{html_module.escape(source_name)} ({report.width}x{report.height})</span>
<span class="status-badge {status_class}">{report.status}{report.score}/100</span>
</div>
<div class="screenshot">{img_tag}</div>
""")
if report.glitches:
parts.append('<div class="glitch-list">')
for g in sorted(report.glitches, key=lambda x: {"critical": 0, "major": 1, "minor": 2, "cosmetic": 3}.get(x.severity.value if hasattr(x.severity, "value") else str(x.severity), 4)):
sev = g.severity.value if hasattr(g.severity, "value") else str(g.severity)
sev_class = f"sev-{sev}"
parts.append(f"""
<div class="glitch-item">
<div class="severity-dot {sev_class}"></div>
<div class="glitch-detail">
<div class="glitch-type">{html_module.escape(g.type)}{sev.upper()}</div>
<div class="glitch-desc">{html_module.escape(g.description)}</div>
<div class="glitch-meta">Region: {html_module.escape(g.region)} | Confidence: {g.confidence:.0%} | Source: {html_module.escape(g.source)}</div>
</div>
</div>""")
parts.append('</div>')
else:
parts.append('<div class="no-glitches">No glitches detected</div>')
parts.append('</div><!-- /report-card -->')
# Footer
parts.append(f"""
<footer>
Generated {datetime.now().strftime('%Y-%m-%d %H:%M:%S')} | matrix_glitch_detect.py | timmy-config#544
</footer>
</div>
</body>
</html>""")
return "\n".join(parts)
# === Output Formatting ===
def format_report(report: GlitchReport, fmt: str = "json") -> str:
if fmt == "json":
data = {
"source": report.source,
"timestamp": report.timestamp,
"status": report.status,
"score": report.score,
"glitches": [asdict(g) for g in report.glitches],
"summary": report.summary,
"model_used": report.model_used,
}
for g in data["glitches"]:
if hasattr(g["severity"], "value"):
g["severity"] = g["severity"].value
return json.dumps(data, indent=2)
elif fmt == "text":
lines = [
"=" * 50,
" GLITCH DETECTION REPORT",
"=" * 50,
f" Source: {report.source}",
f" Status: {report.status}",
f" Score: {report.score}/100",
f" Glitches: {len(report.glitches)}",
"",
]
icons = {"critical": "🔴", "major": "🟡", "minor": "🔵", "cosmetic": ""}
for g in report.glitches:
sev = g.severity.value if hasattr(g.severity, "value") else str(g.severity)
icon = icons.get(sev, "?")
lines.append(f" {icon} [{g.type}] {sev.upper()}: {g.description}")
lines.append(f" Region: {g.region} | Confidence: {g.confidence:.0%}")
lines.append("")
lines.append(f" {report.summary}")
lines.append("=" * 50)
return "\n".join(lines)
return ""
# === CLI ===
def main():
parser = argparse.ArgumentParser(
description="3D World Glitch Detection — visual artifact scanner for The Matrix"
)
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument("--image", help="Screenshot file to analyze")
group.add_argument("--url", help="URL to screenshot and analyze")
group.add_argument("--batch", help="Directory of screenshots to analyze")
parser.add_argument("--vision", action="store_true", help="Include vision model analysis")
parser.add_argument("--model", default=VISION_MODEL, help=f"Vision model (default: {VISION_MODEL})")
parser.add_argument("--html", help="Generate HTML report at this path")
parser.add_argument("--format", choices=["json", "text"], default="json", help="Output format")
parser.add_argument("--output", "-o", help="Output file (default: stdout)")
args = parser.parse_args()
reports = []
if args.image:
print(f"Analyzing {args.image}...", file=sys.stderr)
report = detect_glitches(args.image, args.vision, args.model)
reports.append(report)
if not args.html:
print(format_report(report, args.format))
elif args.url:
import tempfile
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp:
screenshot_path = tmp.name
print(f"Capturing screenshot of {args.url}...", file=sys.stderr)
if capture_screenshot(args.url, screenshot_path):
report = detect_glitches(screenshot_path, args.vision, args.model)
report.source = args.url
reports.append(report)
if not args.html:
print(format_report(report, args.format))
else:
print(f"Failed to capture screenshot", file=sys.stderr)
sys.exit(1)
elif args.batch:
batch_dir = Path(args.batch)
images = sorted(batch_dir.glob("*.png")) + sorted(batch_dir.glob("*.jpg"))
for img in images:
print(f"Analyzing {img.name}...", file=sys.stderr)
report = detect_glitches(str(img), args.vision, args.model)
reports.append(report)
# HTML report
if args.html:
html = generate_html_report(reports, title="The Matrix — Glitch Detection Report")
Path(args.html).write_text(html)
print(f"HTML report written to {args.html}", file=sys.stderr)
elif args.batch and not args.html:
# Print JSON array for batch
print(json.dumps([json.loads(format_report(r, "json")) for r in reports], indent=2))
# Exit code
if any(r.status == "FAIL" for r in reports):
sys.exit(1)
if __name__ == "__main__":
import subprocess
main()

View File

@@ -1,20 +1,582 @@
import json
from hermes_tools import browser_navigate, browser_vision
#!/usr/bin/env python3
"""
nexus_smoke_test.py — Visual Smoke Test for The Nexus.
def run_smoke_test():
print("Navigating to The Nexus...")
browser_navigate(url="https://nexus.alexanderwhitestone.com")
print("Performing visual verification...")
analysis = browser_vision(
question="Is the Nexus landing page rendered correctly? Check for: 1) The Tower logo, 2) The main entry portal, 3) Absence of 404/Error messages. Provide a clear PASS or FAIL."
Takes screenshots of The Nexus landing page, verifies layout consistency
using both programmatic checks (DOM structure, element presence) and
optional vision model analysis (visual regression detection).
The Nexus is the Three.js 3D world frontend at nexus.alexanderwhitestone.com.
This test ensures the landing page renders correctly on every push.
Usage:
# Full smoke test (programmatic + optional vision)
python scripts/nexus_smoke_test.py
# Programmatic only (no vision model needed, CI-safe)
python scripts/nexus_smoke_test.py --programmatic
# With vision model regression check
python scripts/nexus_smoke_test.py --vision
# Against a specific URL
python scripts/nexus_smoke_test.py --url https://nexus.alexanderwhitestone.com
# With baseline comparison
python scripts/nexus_smoke_test.py --baseline screenshots/nexus-baseline.png
Checks:
1. Page loads without errors (HTTP 200, no console errors)
2. Key elements present (Three.js canvas, title, navigation)
3. No 404/error messages visible
4. JavaScript bundle loaded (window.__nexus or scene exists)
5. Screenshot captured successfully
6. Vision model layout verification (optional)
7. Baseline comparison for visual regression (optional)
Refs: timmy-config#490
"""
from __future__ import annotations
import argparse
import base64
import json
import os
import re
import subprocess
import sys
import tempfile
import urllib.error
import urllib.request
from dataclasses import dataclass, field, asdict
from enum import Enum
from pathlib import Path
from typing import Optional
# === Configuration ===
DEFAULT_URL = os.environ.get("NEXUS_URL", "https://nexus.alexanderwhitestone.com")
OLLAMA_BASE = os.environ.get("OLLAMA_BASE_URL", "http://localhost:11434")
VISION_MODEL = os.environ.get("VISUAL_REVIEW_MODEL", "gemma3:12b")
class Severity(str, Enum):
PASS = "pass"
WARN = "warn"
FAIL = "fail"
@dataclass
class SmokeCheck:
"""A single smoke test check."""
name: str
status: Severity = Severity.PASS
message: str = ""
details: str = ""
@dataclass
class SmokeResult:
"""Complete smoke test result."""
url: str = ""
status: Severity = Severity.PASS
checks: list[SmokeCheck] = field(default_factory=list)
screenshot_path: str = ""
summary: str = ""
duration_ms: int = 0
# === HTTP/Network Checks ===
def check_page_loads(url: str) -> SmokeCheck:
"""Verify the page returns HTTP 200."""
check = SmokeCheck(name="Page Loads")
try:
req = urllib.request.Request(url, headers={"User-Agent": "NexusSmokeTest/1.0"})
with urllib.request.urlopen(req, timeout=15) as resp:
if resp.status == 200:
check.status = Severity.PASS
check.message = f"HTTP {resp.status}"
else:
check.status = Severity.WARN
check.message = f"HTTP {resp.status} (expected 200)"
except urllib.error.HTTPError as e:
check.status = Severity.FAIL
check.message = f"HTTP {e.code}: {e.reason}"
except Exception as e:
check.status = Severity.FAIL
check.message = f"Connection failed: {e}"
return check
def check_html_content(url: str) -> tuple[SmokeCheck, str]:
"""Fetch HTML and check for key content."""
check = SmokeCheck(name="HTML Content")
html = ""
try:
req = urllib.request.Request(url, headers={"User-Agent": "NexusSmokeTest/1.0"})
with urllib.request.urlopen(req, timeout=15) as resp:
html = resp.read().decode("utf-8", errors="replace")
except Exception as e:
check.status = Severity.FAIL
check.message = f"Failed to fetch: {e}"
return check, html
issues = []
# Check for Three.js
if "three" not in html.lower() and "THREE" not in html and "threejs" not in html.lower():
issues.append("No Three.js reference found")
# Check for canvas element
if "<canvas" not in html.lower():
issues.append("No <canvas> element found")
# Check title
title_match = re.search(r"<title[^>]*>(.*?)</title>", html, re.IGNORECASE | re.DOTALL)
if title_match:
title = title_match.group(1).strip()
check.details = f"Title: {title}"
if "nexus" not in title.lower() and "tower" not in title.lower():
issues.append(f"Title doesn't reference Nexus: '{title}'")
else:
issues.append("No <title> element")
# Check for error messages
error_patterns = ["404", "not found", "error", "500 internal", "connection refused"]
html_lower = html.lower()
for pattern in error_patterns:
if pattern in html_lower[:500] or pattern in html_lower[-500:]:
issues.append(f"Possible error message in HTML: '{pattern}'")
# Check for script tags (app loaded)
script_count = html.lower().count("<script")
if script_count == 0:
issues.append("No <script> tags found")
else:
check.details += f" | Scripts: {script_count}"
if issues:
check.status = Severity.FAIL if len(issues) > 2 else Severity.WARN
check.message = "; ".join(issues)
else:
check.status = Severity.PASS
check.message = "HTML structure looks correct"
return check, html
# === Screenshot Capture ===
def take_screenshot(url: str, output_path: str, width: int = 1280, height: int = 720) -> SmokeCheck:
"""Take a screenshot of the page."""
check = SmokeCheck(name="Screenshot Capture")
# Try Playwright
try:
script = f"""
import sys
try:
from playwright.sync_api import sync_playwright
except ImportError:
sys.exit(2)
with sync_playwright() as p:
browser = p.chromium.launch(headless=True)
page = browser.new_page(viewport={{"width": {width}, "height": {height}}})
errors = []
page.on("pageerror", lambda e: errors.append(str(e)))
page.on("console", lambda m: errors.append(f"console.{{m.type}}: {{m.text}}") if m.type == "error" else None)
page.goto("{url}", wait_until="networkidle", timeout=30000)
page.wait_for_timeout(3000) # Wait for Three.js to render
page.screenshot(path="{output_path}", full_page=False)
# Check for Three.js scene
has_canvas = page.evaluate("() => !!document.querySelector('canvas')")
has_three = page.evaluate("() => typeof THREE !== 'undefined' || !!document.querySelector('canvas')")
title = page.title()
browser.close()
import json
print(json.dumps({{"has_canvas": has_canvas, "has_three": has_three, "title": title, "errors": errors[:5]}}))
"""
result = subprocess.run(
["python3", "-c", script],
capture_output=True, text=True, timeout=60
)
if result.returncode == 0:
# Parse Playwright output
try:
# Find JSON in output
for line in result.stdout.strip().split("\n"):
if line.startswith("{"):
info = json.loads(line)
extras = []
if info.get("has_canvas"):
extras.append("canvas present")
if info.get("errors"):
extras.append(f"{len(info['errors'])} JS errors")
check.details = "; ".join(extras) if extras else "Playwright capture"
if info.get("errors"):
check.status = Severity.WARN
check.message = f"JS errors detected: {info['errors'][0][:100]}"
else:
check.message = "Screenshot captured via Playwright"
break
except json.JSONDecodeError:
pass
if Path(output_path).exists() and Path(output_path).stat().st_size > 1000:
return check
elif result.returncode == 2:
check.details = "Playwright not installed"
else:
check.details = f"Playwright failed: {result.stderr[:200]}"
except Exception as e:
check.details = f"Playwright error: {e}"
# Try wkhtmltoimage
try:
result = subprocess.run(
["wkhtmltoimage", "--width", str(width), "--quality", "90", url, output_path],
capture_output=True, text=True, timeout=30
)
if result.returncode == 0 and Path(output_path).exists() and Path(output_path).stat().st_size > 1000:
check.status = Severity.PASS
check.message = "Screenshot captured via wkhtmltoimage"
check.details = ""
return check
except Exception:
pass
# Try curl + browserless (if available)
browserless = os.environ.get("BROWSERLESS_URL")
if browserless:
try:
payload = json.dumps({
"url": url,
"options": {"type": "png", "fullPage": False}
})
req = urllib.request.Request(
f"{browserless}/screenshot",
data=payload.encode(),
headers={"Content-Type": "application/json"}
)
with urllib.request.urlopen(req, timeout=30) as resp:
img_data = resp.read()
Path(output_path).write_bytes(img_data)
if Path(output_path).stat().st_size > 1000:
check.status = Severity.PASS
check.message = "Screenshot captured via browserless"
check.details = ""
return check
except Exception:
pass
check.status = Severity.WARN
check.message = "No screenshot backend available"
check.details = "Install Playwright: pip install playwright && playwright install chromium"
return check
# === Vision Analysis ===
VISION_PROMPT = """You are a web QA engineer. Analyze this screenshot of The Nexus (a Three.js 3D world).
Check for:
1. LAYOUT: Is the page layout correct? Is content centered, not broken or overlapping?
2. THREE.JS RENDER: Is there a visible 3D canvas/scene? Any black/blank areas where rendering failed?
3. NAVIGATION: Are navigation elements (buttons, links, menu) visible and properly placed?
4. TEXT: Is text readable? Any missing text, garbled characters, or font issues?
5. ERRORS: Any visible error messages, 404 pages, or broken images?
6. TOWER: Is the Tower or entry portal visible in the scene?
Respond as JSON:
{
"status": "PASS|FAIL|WARN",
"checks": [
{"name": "Layout", "status": "pass|fail|warn", "message": "..."},
{"name": "Three.js Render", "status": "pass|fail|warn", "message": "..."},
{"name": "Navigation", "status": "pass|fail|warn", "message": "..."},
{"name": "Text Readability", "status": "pass|fail|warn", "message": "..."},
{"name": "Error Messages", "status": "pass|fail|warn", "message": "..."}
],
"summary": "brief overall assessment"
}"""
def run_vision_check(screenshot_path: str, model: str = VISION_MODEL) -> list[SmokeCheck]:
"""Run vision model analysis on screenshot."""
checks = []
try:
b64 = base64.b64encode(Path(screenshot_path).read_bytes()).decode()
payload = json.dumps({
"model": model,
"messages": [{"role": "user", "content": [
{"type": "text", "text": VISION_PROMPT},
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{b64}"}}
]}],
"stream": False,
"options": {"temperature": 0.1}
}).encode()
req = urllib.request.Request(
f"{OLLAMA_BASE}/api/chat",
data=payload,
headers={"Content-Type": "application/json"}
)
with urllib.request.urlopen(req, timeout=120) as resp:
result = json.loads(resp.read())
content = result.get("message", {}).get("content", "")
parsed = _parse_json_response(content)
for c in parsed.get("checks", []):
status = Severity(c.get("status", "warn"))
checks.append(SmokeCheck(
name=f"Vision: {c.get('name', 'Unknown')}",
status=status,
message=c.get("message", "")
))
if not checks:
checks.append(SmokeCheck(
name="Vision Analysis",
status=Severity.WARN,
message="Vision model returned no structured checks"
))
except Exception as e:
checks.append(SmokeCheck(
name="Vision Analysis",
status=Severity.WARN,
message=f"Vision check failed: {e}"
))
return checks
# === Baseline Comparison ===
def compare_baseline(current_path: str, baseline_path: str) -> SmokeCheck:
"""Compare screenshot against baseline for visual regression."""
check = SmokeCheck(name="Baseline Comparison")
if not Path(baseline_path).exists():
check.status = Severity.WARN
check.message = f"Baseline not found: {baseline_path}"
return check
if not Path(current_path).exists():
check.status = Severity.FAIL
check.message = "No current screenshot to compare"
return check
# Simple file size comparison (rough regression indicator)
baseline_size = Path(baseline_path).stat().st_size
current_size = Path(current_path).stat().st_size
if baseline_size == 0:
check.status = Severity.WARN
check.message = "Baseline is empty"
return check
diff_pct = abs(current_size - baseline_size) / baseline_size * 100
if diff_pct > 50:
check.status = Severity.FAIL
check.message = f"Major visual change: {diff_pct:.0f}% file size difference"
elif diff_pct > 20:
check.status = Severity.WARN
check.message = f"Significant visual change: {diff_pct:.0f}% file size difference"
else:
check.status = Severity.PASS
check.message = f"Visual consistency: {diff_pct:.1f}% difference"
check.details = f"Baseline: {baseline_size}B, Current: {current_size}B"
# Pixel-level diff using ImageMagick (if available)
try:
diff_output = current_path.replace(".png", "-diff.png")
result = subprocess.run(
["compare", "-metric", "AE", current_path, baseline_path, diff_output],
capture_output=True, text=True, timeout=15
)
if result.returncode < 2:
pixels_diff = int(result.stderr) if result.stderr.strip().isdigit() else 0
check.details += f" | Pixel diff: {pixels_diff}"
if pixels_diff > 10000:
check.status = Severity.FAIL
check.message = f"Major visual regression: {pixels_diff} pixels changed"
elif pixels_diff > 1000:
check.status = Severity.WARN
check.message = f"Visual change detected: {pixels_diff} pixels changed"
except Exception:
pass
return check
# === Helpers ===
def _parse_json_response(text: str) -> dict:
cleaned = text.strip()
if cleaned.startswith("```"):
lines = cleaned.split("\n")[1:]
if lines and lines[-1].strip() == "```":
lines = lines[:-1]
cleaned = "\n".join(lines)
try:
return json.loads(cleaned)
except json.JSONDecodeError:
start = cleaned.find("{")
end = cleaned.rfind("}")
if start >= 0 and end > start:
try:
return json.loads(cleaned[start:end + 1])
except json.JSONDecodeError:
pass
return {}
# === Main Smoke Test ===
def run_smoke_test(url: str, vision: bool = False, baseline: Optional[str] = None,
model: str = VISION_MODEL) -> SmokeResult:
"""Run the full visual smoke test suite."""
import time
start = time.time()
result = SmokeResult(url=url)
screenshot_path = ""
# 1. Page loads
print(f" [1/5] Checking page loads...", file=sys.stderr)
result.checks.append(check_page_loads(url))
# 2. HTML content
print(f" [2/5] Checking HTML content...", file=sys.stderr)
html_check, html = check_html_content(url)
result.checks.append(html_check)
# 3. Screenshot
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp:
screenshot_path = tmp.name
print(f" [3/5] Taking screenshot...", file=sys.stderr)
screenshot_check = take_screenshot(url, screenshot_path)
result.checks.append(screenshot_check)
result.screenshot_path = screenshot_path
# 4. Vision analysis (optional)
if vision and Path(screenshot_path).exists():
print(f" [4/5] Running vision analysis...", file=sys.stderr)
result.checks.extend(run_vision_check(screenshot_path, model))
else:
print(f" [4/5] Vision analysis skipped", file=sys.stderr)
# 5. Baseline comparison (optional)
if baseline:
print(f" [5/5] Comparing against baseline...", file=sys.stderr)
result.checks.append(compare_baseline(screenshot_path, baseline))
else:
print(f" [5/5] Baseline comparison skipped", file=sys.stderr)
# Determine overall status
fails = sum(1 for c in result.checks if c.status == Severity.FAIL)
warns = sum(1 for c in result.checks if c.status == Severity.WARN)
if fails > 0:
result.status = Severity.FAIL
elif warns > 0:
result.status = Severity.WARN
else:
result.status = Severity.PASS
result.summary = (
f"{result.status.value.upper()}: {len(result.checks)} checks, "
f"{fails} failures, {warns} warnings"
)
result = {
"status": "PASS" if "PASS" in analysis.upper() else "FAIL",
"analysis": analysis
}
result.duration_ms = int((time.time() - start) * 1000)
return result
if __name__ == '__main__':
print(json.dumps(run_smoke_test(), indent=2))
# === Output ===
def format_result(result: SmokeResult, fmt: str = "json") -> str:
if fmt == "json":
data = {
"url": result.url,
"status": result.status.value,
"summary": result.summary,
"duration_ms": result.duration_ms,
"screenshot": result.screenshot_path,
"checks": [asdict(c) for c in result.checks],
}
for c in data["checks"]:
if hasattr(c["status"], "value"):
c["status"] = c["status"].value
return json.dumps(data, indent=2)
elif fmt == "text":
lines = [
"=" * 50,
" NEXUS VISUAL SMOKE TEST",
"=" * 50,
f" URL: {result.url}",
f" Status: {result.status.value.upper()}",
f" Duration: {result.duration_ms}ms",
"",
]
icons = {"pass": "", "warn": "⚠️", "fail": ""}
for c in result.checks:
icon = icons.get(c.status.value if hasattr(c.status, "value") else str(c.status), "?")
lines.append(f" {icon} {c.name}: {c.message}")
if c.details:
lines.append(f" {c.details}")
lines.append("")
lines.append(f" {result.summary}")
lines.append("=" * 50)
return "\n".join(lines)
return ""
# === CLI ===
def main():
parser = argparse.ArgumentParser(
description="Visual Smoke Test for The Nexus — layout + regression verification"
)
parser.add_argument("--url", default=DEFAULT_URL, help=f"Nexus URL (default: {DEFAULT_URL})")
parser.add_argument("--vision", action="store_true", help="Include vision model analysis")
parser.add_argument("--baseline", help="Baseline screenshot for regression comparison")
parser.add_argument("--model", default=VISION_MODEL, help=f"Vision model (default: {VISION_MODEL})")
parser.add_argument("--format", choices=["json", "text"], default="json")
parser.add_argument("--output", "-o", help="Output file (default: stdout)")
args = parser.parse_args()
print(f"Running smoke test on {args.url}...", file=sys.stderr)
result = run_smoke_test(args.url, vision=args.vision, baseline=args.baseline, model=args.model)
output = format_result(result, args.format)
if args.output:
Path(args.output).write_text(output)
print(f"Results written to {args.output}", file=sys.stderr)
else:
print(output)
if result.status == Severity.FAIL:
sys.exit(1)
elif result.status == Severity.WARN:
sys.exit(0) # Warnings don't fail CI
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,50 @@
# Nightly Pipeline Scheduler
Auto-starts batch pipelines when inference is available.
## What It Does
1. Checks inference provider health (OpenRouter, Ollama, RunPod)
2. Checks if it's off-peak hours (configurable, default: after 6PM)
3. Checks interactive session load (don't fight with live users)
4. Checks daily token budget (configurable limit)
5. Starts the highest-priority incomplete pipeline
## Pipeline Priority Order
| Priority | Pipeline | Deps | Max Tokens |
|----------|----------|------|------------|
| 1 | playground-factory | none | 100,000 |
| 2 | training-factory | none | 150,000 |
| 3 | knowledge-mine | training-factory running | 80,000 |
| 4 | adversary | knowledge-mine running | 50,000 |
| 5 | codebase-genome | none | 120,000 |
## Usage
```bash
# Normal run (used by cron)
./scripts/nightly-pipeline-scheduler.sh
# Dry run (show what would start)
./scripts/nightly-pipeline-scheduler.sh --dry-run
# Status report
./scripts/nightly-pipeline-scheduler.sh --status
# Force start during peak hours
./scripts/nightly-pipeline-scheduler.sh --force
```
## Configuration
Set via environment variables:
- `PIPELINE_TOKEN_LIMIT`: Daily token budget (default: 500,000)
- `PIPELINE_PEAK_START`: Peak hours start (default: 9)
- `PIPELINE_PEAK_END`: Peak hours end (default: 18)
- `HERMES_HOME`: Hermes home directory (default: ~/.hermes)
## Cron
Runs every 30 minutes. Off-peak only (unless --force).
See `cron/pipeline-scheduler.yml`.

View File

@@ -0,0 +1,383 @@
#!/usr/bin/env bash
# nightly-pipeline-scheduler.sh — Auto-start batch pipelines when inference is available.
#
# Checks provider health, pipeline progress, token budget, and interactive load.
# Starts the highest-priority incomplete pipeline that can run.
#
# Usage:
# ./scripts/nightly-pipeline-scheduler.sh # Normal run
# ./scripts/nightly-pipeline-scheduler.sh --dry-run # Show what would start
# ./scripts/nightly-pipeline-scheduler.sh --status # Pipeline status report
set -euo pipefail
# --- Configuration ---
HERMES_HOME="${HERMES_HOME:-$HOME/.hermes}"
BUDGET_FILE="${HERMES_HOME}/pipeline_budget.json"
STATE_FILE="${HERMES_HOME}/pipeline_state.json"
LOG_FILE="${HERMES_HOME}/logs/pipeline-scheduler.log"
TOKEN_DAILY_LIMIT="${PIPELINE_TOKEN_LIMIT:-500000}"
PEAK_HOURS_START="${PIPELINE_PEAK_START:-9}"
PEAK_HOURS_END="${PIPELINE_PEAK_END:-18}"
# Pipeline definitions (priority order)
# Each pipeline: name, script, max_tokens, dependencies
PIPELINES=(
"playground-factory|scripts/pipeline_playground_factory.sh|100000|none"
"training-factory|scripts/pipeline_training_factory.sh|150000|none"
"knowledge-mine|scripts/pipeline_knowledge_mine.sh|80000|training-factory"
"adversary|scripts/pipeline_adversary.sh|50000|knowledge-mine"
"codebase-genome|scripts/pipeline_codebase_genome.sh|120000|none"
)
# --- Colors ---
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[0;33m'
CYAN='\033[0;36m'
NC='\033[0m'
# --- Helpers ---
now_hour() { date +%-H; }
is_peak_hours() {
local h=$(now_hour)
[[ $h -ge $PEAK_HOURS_START && $h -lt $PEAK_HOURS_END ]]
}
ensure_dirs() {
mkdir -p "$(dirname "$LOG_FILE")" "$(dirname "$BUDGET_FILE")" "$(dirname "$STATE_FILE")"
}
log() { echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*" | tee -a "$LOG_FILE"; }
get_budget_used_today() {
if [[ -f "$BUDGET_FILE" ]]; then
local today=$(date +%Y-%m-%d)
python3 -c "
import json, sys
with open('$BUDGET_FILE') as f:
d = json.load(f)
print(d.get('daily', {}).get('$today', {}).get('tokens_used', 0))
" 2>/dev/null || echo 0
else
echo 0
fi
}
get_budget_remaining() {
local used=$(get_budget_used_today)
echo $((TOKEN_DAILY_LIMIT - used))
}
update_budget() {
local pipeline="$1"
local tokens="$2"
local today=$(date +%Y-%m-%d)
python3 -c "
import json, os
path = '$BUDGET_FILE'
d = {}
if os.path.exists(path):
with open(path) as f:
d = json.load(f)
daily = d.setdefault('daily', {})
day = daily.setdefault('$today', {'tokens_used': 0, 'pipelines': {}})
day['tokens_used'] = day.get('tokens_used', 0) + $tokens
day['pipelines']['$pipeline'] = day['pipelines'].get('$pipeline', 0) + $tokens
with open(path, 'w') as f:
json.dump(d, f, indent=2)
"
}
get_pipeline_state() {
if [[ -f "$STATE_FILE" ]]; then
cat "$STATE_FILE"
else
echo "{}"
fi
}
set_pipeline_state() {
local pipeline="$1"
local state="$2" # running, complete, failed, skipped
python3 -c "
import json, os
path = '$STATE_FILE'
d = {}
if os.path.exists(path):
with open(path) as f:
d = json.load(f)
d['$pipeline'] = {'state': '$state', 'updated': '$(date -Iseconds)'}
with open(path, 'w') as f:
json.dump(d, f, indent=2)
"
}
is_pipeline_complete() {
local pipeline="$1"
python3 -c "
import json, os
path = '$STATE_FILE'
if not os.path.exists(path):
print('false')
else:
with open(path) as f:
d = json.load(f)
state = d.get('$pipeline', {}).get('state', 'not_started')
print('true' if state == 'complete' else 'false')
" 2>/dev/null || echo false
}
is_pipeline_running() {
local pipeline="$1"
python3 -c "
import json, os
path = '$STATE_FILE'
if not os.path.exists(path):
print('false')
else:
with open(path) as f:
d = json.load(f)
state = d.get('$pipeline', {}).get('state', 'not_started')
print('true' if state == 'running' else 'false')
" 2>/dev/null || echo false
}
check_dependency() {
local dep="$1"
if [[ "$dep" == "none" ]]; then
return 0
fi
# For knowledge-mine: training-factory must be running or complete
if [[ "$dep" == "training-factory" ]]; then
local state=$(python3 -c "
import json, os
path = '$STATE_FILE'
if not os.path.exists(path):
print('not_started')
else:
with open(path) as f:
d = json.load(f)
print(d.get('training-factory', {}).get('state', 'not_started'))
" 2>/dev/null || echo "not_started")
[[ "$state" == "running" || "$state" == "complete" ]]
return $?
fi
# For adversary: knowledge-mine must be at least 50% done
# Simplified: check if it's running (we'd need progress tracking for 50%)
if [[ "$dep" == "knowledge-mine" ]]; then
local state=$(python3 -c "
import json, os
path = '$STATE_FILE'
if not os.path.exists(path):
print('not_started')
else:
with open(path) as f:
d = json.load(f)
print(d.get('knowledge-mine', {}).get('state', 'not_started'))
" 2>/dev/null || echo "not_started")
[[ "$state" == "running" || "$state" == "complete" ]]
return $?
fi
return 0
}
check_inference_available() {
# Check if any inference provider is responding
# 1. Check OpenRouter
local or_ok=$(curl -s -o /dev/null -w "%{http_code}" \
--connect-timeout 5 "https://openrouter.ai/api/v1/models" 2>/dev/null || echo "000")
# 2. Check local Ollama
local ollama_ok=$(curl -s -o /dev/null -w "%{http_code}" \
--connect-timeout 5 "http://localhost:11434/api/tags" 2>/dev/null || echo "000")
# 3. Check RunPod (if configured)
local runpod_ok="000"
if [[ -n "${RUNPOD_ENDPOINT:-}" ]]; then
runpod_ok=$(curl -s -o /dev/null -w "%{http_code}" \
--connect-timeout 5 "$RUNPOD_ENDPOINT/health" 2>/dev/null || echo "000")
fi
if [[ "$or_ok" == "200" || "$ollama_ok" == "200" || "$runpod_ok" == "200" ]]; then
return 0
fi
return 1
}
check_interactive_load() {
# Check if there are active interactive sessions (don't fight with live users)
# Look for tmux panes with active hermes sessions
local active=$(tmux list-panes -a -F '#{pane_pid} #{pane_current_command}' 2>/dev/null \
| grep -c "hermes\|python3" || echo 0)
# If more than 3 interactive sessions, skip pipeline start
if [[ $active -gt 3 ]]; then
return 1
fi
return 0
}
start_pipeline() {
local name="$1"
local script="$2"
local max_tokens="$3"
local budget_remaining="$4"
local mode="${5:-run}"
if [[ "$budget_remaining" -lt "$max_tokens" ]]; then
log "SKIP $name: insufficient budget ($budget_remaining < $max_tokens tokens)"
return 1
fi
if [[ ! -f "$script" ]]; then
log "SKIP $name: script not found ($script)"
return 1
fi
if [[ "$mode" == "dry-run" ]]; then
log "DRY-RUN: Would start $name (budget: $budget_remaining, needs: $max_tokens)"
return 0
fi
log "START $name (budget: $budget_remaining, max_tokens: $max_tokens)"
set_pipeline_state "$name" "running"
# Run in background, capture output
local log_path="${HERMES_HOME}/logs/pipeline-${name}.log"
bash "$script" --max-tokens "$max_tokens" >> "$log_path" 2>&1 &
local pid=$!
# Wait a moment to check if it started OK
sleep 2
if kill -0 $pid 2>/dev/null; then
log "RUNNING $name (PID: $pid, log: $log_path)"
# Record the PID
python3 -c "
import json, os
path = '$STATE_FILE'
d = {}
if os.path.exists(path):
with open(path) as f:
d = json.load(f)
d['$name']['pid'] = $pid
with open(path, 'w') as f:
json.dump(d, f, indent=2)
"
return 0
else
log "FAIL $name: script exited immediately"
set_pipeline_state "$name" "failed"
return 1
fi
}
# --- Main ---
main() {
local mode="${1:-run}"
ensure_dirs
log "=== Pipeline Scheduler ($mode) ==="
# Check 1: Is inference available?
if ! check_inference_available; then
log "No inference provider available. Skipping all pipelines."
exit 0
fi
log "Inference: AVAILABLE"
# Check 2: Is it peak hours?
if is_peak_hours && [[ "$mode" != "--force" ]]; then
local h=$(now_hour)
log "Peak hours ($h:00). Skipping pipeline start. Use --force to override."
exit 0
fi
log "Off-peak: OK"
# Check 3: Interactive load
if ! check_interactive_load && [[ "$mode" != "--force" ]]; then
log "High interactive load. Skipping pipeline start."
exit 0
fi
log "Interactive load: OK"
# Check 4: Token budget
local budget=$(get_budget_remaining)
log "Token budget remaining: $budget / $TOKEN_DAILY_LIMIT"
if [[ $budget -le 0 ]]; then
log "Daily token budget exhausted. Stopping."
exit 0
fi
# Check 5: Pipeline status
if [[ "$mode" == "--status" ]]; then
echo -e "${CYAN}Pipeline Status:${NC}"
echo "────────────────────────────────────────────────────"
for entry in "${PIPELINES[@]}"; do
IFS='|' read -r name script max_tokens dep <<< "$entry"
local state=$(python3 -c "
import json, os
path = '$STATE_FILE'
if not os.path.exists(path):
print('not_started')
else:
with open(path) as f:
d = json.load(f)
print(d.get('$name', {}).get('state', 'not_started'))
" 2>/dev/null || echo "not_started")
local color=$NC
case "$state" in
running) color=$YELLOW ;;
complete) color=$GREEN ;;
failed) color=$RED ;;
esac
printf " %-25s %b%s%b (max: %s tokens, dep: %s)\n" "$name" "$color" "$state" "$NC" "$max_tokens" "$dep"
done
echo "────────────────────────────────────────────────────"
echo " Budget: $budget / $TOKEN_DAILY_LIMIT tokens remaining"
echo " Peak hours: $PEAK_HOURS_START:00 - $PEAK_HOURS_END:00"
exit 0
fi
# Find and start the highest-priority incomplete pipeline
local started=0
for entry in "${PIPELINES[@]}"; do
IFS='|' read -r name script max_tokens dep <<< "$entry"
# Skip if already running or complete
if [[ "$(is_pipeline_running $name)" == "true" ]]; then
log "SKIP $name: already running"
continue
fi
if [[ "$(is_pipeline_complete $name)" == "true" ]]; then
log "SKIP $name: already complete"
continue
fi
# Check dependency
if ! check_dependency "$dep"; then
log "SKIP $name: dependency $dep not met"
continue
fi
# Try to start
if start_pipeline "$name" "$script" "$max_tokens" "$budget" "$mode"; then
started=1
# Only start one pipeline per run (let it claim tokens before next check)
# Exception: playground-factory and training-factory can run in parallel
if [[ "$name" != "playground-factory" && "$name" != "training-factory" ]]; then
break
fi
fi
done
if [[ $started -eq 0 ]]; then
log "No pipelines to start (all complete, running, or blocked)."
fi
log "=== Pipeline Scheduler done ==="
}
main "$@"

View File

@@ -1,3 +1,4 @@
#!/usr/bin/env python3
import json
from hermes_tools import browser_navigate, browser_vision

View File

@@ -0,0 +1,301 @@
#!/usr/bin/env python3
"""Tests for foundation_accessibility_audit.py — verifies WCAG checks."""
import json
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
from foundation_accessibility_audit import (
A11yHTMLParser, Severity, A11yViolation,
parse_color, contrast_ratio, relative_luminance,
run_programmatic_checks, check_page_title, check_images_alt_text,
check_heading_hierarchy, check_lang_attribute, check_landmarks,
check_skip_nav, check_form_labels, check_link_text,
_parse_json_response, format_report, A11yAuditReport, A11yPageResult,
)
# === Color Utilities ===
def test_parse_color_hex6():
assert parse_color("#ff0000") == (255, 0, 0)
assert parse_color("#000000") == (0, 0, 0)
assert parse_color("#ffffff") == (255, 255, 255)
print(" PASS: test_parse_color_hex6")
def test_parse_color_hex3():
assert parse_color("#f00") == (255, 0, 0)
assert parse_color("#abc") == (170, 187, 204)
print(" PASS: test_parse_color_hex3")
def test_parse_color_rgb():
assert parse_color("rgb(255, 0, 0)") == (255, 0, 0)
assert parse_color("rgb( 128 , 64 , 32 )") == (128, 64, 32)
print(" PASS: test_parse_color_rgb")
def test_parse_color_named():
assert parse_color("white") == (255, 255, 255)
assert parse_color("black") == (0, 0, 0)
print(" PASS: test_parse_color_named")
def test_parse_color_invalid():
assert parse_color("not-a-color") is None
assert parse_color("") is None
print(" PASS: test_parse_color_invalid")
def test_contrast_ratio_black_white():
ratio = contrast_ratio((0, 0, 0), (255, 255, 255))
assert ratio > 20 # Should be 21:1
print(f" PASS: test_contrast_ratio_black_white ({ratio:.1f}:1)")
def test_contrast_ratio_same():
ratio = contrast_ratio((128, 128, 128), (128, 128, 128))
assert ratio == 1.0
print(" PASS: test_contrast_ratio_same")
def test_contrast_ratio_wcag_aa():
# #767676 on white = 4.54:1 (WCAG AA pass for normal text)
ratio = contrast_ratio((118, 118, 118), (255, 255, 255))
assert ratio >= 4.5
print(f" PASS: test_contrast_ratio_wcag_aa ({ratio:.2f}:1)")
# === HTML Parser ===
def test_parser_title():
parser = A11yHTMLParser()
parser.feed("<html><head><title>Test Page</title></head></html>")
assert parser.title == "Test Page"
print(" PASS: test_parser_title")
def test_parser_images():
parser = A11yHTMLParser()
parser.feed('<html><body><img src="a.png" alt="Alt text"><img src="b.png"></body></html>')
assert len(parser.images) == 2
assert parser.images[0]["alt"] == "Alt text"
assert parser.images[1]["alt"] is None
print(" PASS: test_parser_images")
def test_parser_headings():
parser = A11yHTMLParser()
parser.feed("<html><body><h1>Main</h1><h2>Sub</h2><h4>Skip</h4></body></html>")
assert len(parser.headings) == 3
assert parser.headings[0] == {"level": 1, "text": "Main"}
assert parser.headings[2] == {"level": 4, "text": "Skip"}
print(" PASS: test_parser_headings")
def test_parser_lang():
parser = A11yHTMLParser()
parser.feed('<html lang="en"><body></body></html>')
assert parser.lang == "en"
print(" PASS: test_parser_lang")
def test_parser_landmarks():
parser = A11yHTMLParser()
parser.feed("<html><body><nav>Links</nav><main>Content</main></body></html>")
tags = {lm["tag"] for lm in parser.landmarks}
assert "nav" in tags
assert "main" in tags
print(" PASS: test_parser_landmarks")
# === Programmatic Checks ===
def test_check_page_title_empty():
parser = A11yHTMLParser()
parser.title = ""
violations = check_page_title(parser)
assert len(violations) == 1
assert violations[0].criterion == "2.4.2"
assert violations[0].severity == Severity.MAJOR
print(" PASS: test_check_page_title_empty")
def test_check_page_title_present():
parser = A11yHTMLParser()
parser.title = "My Great Page"
violations = check_page_title(parser)
assert len(violations) == 0
print(" PASS: test_check_page_title_present")
def test_check_lang_missing():
parser = A11yHTMLParser()
parser.lang = ""
violations = check_lang_attribute(parser)
assert len(violations) == 1
assert violations[0].criterion == "3.1.1"
print(" PASS: test_check_lang_missing")
def test_check_images_missing_alt():
parser = A11yHTMLParser()
parser.images = [{"src": "photo.jpg", "alt": None}]
violations = check_images_alt_text(parser)
assert len(violations) == 1
assert violations[0].severity == Severity.CRITICAL
print(" PASS: test_check_images_missing_alt")
def test_check_images_with_alt():
parser = A11yHTMLParser()
parser.images = [{"src": "photo.jpg", "alt": "A photo"}]
violations = check_images_alt_text(parser)
assert len(violations) == 0
print(" PASS: test_check_images_with_alt")
def test_check_images_decorative():
parser = A11yHTMLParser()
parser.images = [{"src": "deco.png", "alt": "", "role": "presentation"}]
violations = check_images_alt_text(parser)
assert len(violations) == 0
print(" PASS: test_check_images_decorative")
def test_check_headings_no_h1():
parser = A11yHTMLParser()
parser.headings = [{"level": 2, "text": "Sub"}, {"level": 3, "text": "Sub sub"}]
violations = check_heading_hierarchy(parser)
assert any(v.criterion == "1.3.1" and "h1" in v.description.lower() for v in violations)
print(" PASS: test_check_headings_no_h1")
def test_check_headings_skip():
parser = A11yHTMLParser()
parser.headings = [{"level": 1, "text": "Main"}, {"level": 4, "text": "Skipped"}]
violations = check_heading_hierarchy(parser)
assert any("skipped" in v.description.lower() for v in violations)
print(" PASS: test_check_headings_skip")
def test_check_skip_nav_missing():
parser = A11yHTMLParser()
parser.skip_nav = False
parser.links = [{"text": "Home", "href": "/"}, {"text": "About", "href": "/about"}]
violations = check_skip_nav(parser)
assert len(violations) == 1
assert violations[0].criterion == "2.4.1"
print(" PASS: test_check_skip_nav_missing")
def test_check_link_text_empty():
parser = A11yHTMLParser()
parser.links = [{"text": "", "href": "/page", "aria_label": ""}]
violations = check_link_text(parser)
assert len(violations) == 1
assert violations[0].criterion == "2.4.4"
print(" PASS: test_check_link_text_empty")
def test_check_link_text_generic():
parser = A11yHTMLParser()
parser.links = [{"text": "Click here", "href": "/page"}]
violations = check_link_text(parser)
assert any("non-descriptive" in v.description.lower() for v in violations)
print(" PASS: test_check_link_text_generic")
def test_run_programmatic_checks_full():
html = """<!DOCTYPE html>
<html lang="en">
<head><title>Good Page</title></head>
<body>
<nav><a href="#main">Skip to content</a></nav>
<main>
<h1>Welcome</h1>
<h2>Section</h2>
<img src="hero.jpg" alt="Hero image">
<a href="/about">About Us</a>
</main>
</body>
</html>"""
violations = run_programmatic_checks(html)
# This page should have very few or no violations
criticals = [v for v in violations if v.severity == Severity.CRITICAL]
assert len(criticals) == 0
print(f" PASS: test_run_programmatic_checks_full ({len(violations)} minor issues)")
# === JSON Parsing ===
def test_parse_json_clean():
result = _parse_json_response('{"violations": [], "overall_score": 100}')
assert result["overall_score"] == 100
print(" PASS: test_parse_json_clean")
def test_parse_json_fenced():
result = _parse_json_response('```json\n{"overall_score": 80}\n```')
assert result["overall_score"] == 80
print(" PASS: test_parse_json_fenced")
# === Formatting ===
def test_format_json():
report = A11yAuditReport(site="test.com", pages_audited=1, overall_score=90)
output = format_report(report, "json")
parsed = json.loads(output)
assert parsed["site"] == "test.com"
assert parsed["overall_score"] == 90
print(" PASS: test_format_json")
def test_format_text():
report = A11yAuditReport(site="test.com", pages_audited=1, overall_score=90,
summary="Test complete")
output = format_report(report, "text")
assert "ACCESSIBILITY AUDIT" in output
assert "test.com" in output
print(" PASS: test_format_text")
# === Run All ===
def run_all():
print("=== foundation_accessibility_audit tests ===")
tests = [
test_parse_color_hex6, test_parse_color_hex3, test_parse_color_rgb,
test_parse_color_named, test_parse_color_invalid,
test_contrast_ratio_black_white, test_contrast_ratio_same, test_contrast_ratio_wcag_aa,
test_parser_title, test_parser_images, test_parser_headings,
test_parser_lang, test_parser_landmarks,
test_check_page_title_empty, test_check_page_title_present,
test_check_lang_missing,
test_check_images_missing_alt, test_check_images_with_alt, test_check_images_decorative,
test_check_headings_no_h1, test_check_headings_skip,
test_check_skip_nav_missing,
test_check_link_text_empty, test_check_link_text_generic,
test_run_programmatic_checks_full,
test_parse_json_clean, test_parse_json_fenced,
test_format_json, test_format_text,
]
passed = 0
failed = 0
for test in tests:
try:
test()
passed += 1
except Exception as e:
print(f" FAIL: {test.__name__}{e}")
failed += 1
print(f"\n{'ALL PASSED' if failed == 0 else f'{failed} FAILED'}: {passed}/{len(tests)}")
return failed == 0
if __name__ == "__main__":
sys.exit(0 if run_all() else 1)

View File

@@ -0,0 +1,380 @@
#!/usr/bin/env python3
"""
Tests for Matrix 3D Glitch Detector (timmy-config#491).
Covers: glitch_patterns, matrix_glitch_detector core logic.
"""
import json
import sys
import tempfile
import unittest
from pathlib import Path
# Ensure bin/ is importable
sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "bin"))
from glitch_patterns import (
GlitchCategory,
GlitchPattern,
GlitchSeverity,
MATRIX_GLITCH_PATTERNS,
THREEJS_CATEGORIES,
build_vision_prompt,
get_pattern_by_category,
get_patterns_by_severity,
get_threejs_patterns,
)
from matrix_glitch_detector import (
DetectedGlitch,
ScanResult,
_infer_severity,
_parse_vision_response,
build_report,
generate_scan_angles,
run_demo,
)
class TestGlitchPatterns(unittest.TestCase):
"""Tests for glitch_patterns module."""
def test_pattern_count(self):
"""Verify we have a reasonable number of defined patterns."""
self.assertGreaterEqual(len(MATRIX_GLITCH_PATTERNS), 14) # 10 generic + 6 Three.js
def test_all_patterns_have_required_fields(self):
"""Every pattern must have category, name, description, severity, prompts."""
for p in MATRIX_GLITCH_PATTERNS:
self.assertIsInstance(p.category, GlitchCategory)
self.assertTrue(p.name)
self.assertTrue(p.description)
self.assertIsInstance(p.severity, GlitchSeverity)
self.assertGreater(len(p.detection_prompts), 0)
self.assertGreater(len(p.visual_indicators), 0)
self.assertGreater(p.confidence_threshold, 0)
self.assertLessEqual(p.confidence_threshold, 1.0)
def test_pattern_to_dict(self):
"""Pattern serialization should produce a dict with expected keys."""
p = MATRIX_GLITCH_PATTERNS[0]
d = p.to_dict()
self.assertIn("category", d)
self.assertIn("name", d)
self.assertIn("severity", d)
self.assertEqual(d["category"], p.category.value)
def test_get_patterns_by_severity(self):
"""Severity filter should return only patterns at or above threshold."""
high_patterns = get_patterns_by_severity(GlitchSeverity.HIGH)
self.assertTrue(all(p.severity.value in ("high", "critical") for p in high_patterns))
self.assertGreater(len(high_patterns), 0)
all_patterns = get_patterns_by_severity(GlitchSeverity.INFO)
self.assertEqual(len(all_patterns), len(MATRIX_GLITCH_PATTERNS))
def test_get_pattern_by_category(self):
"""Lookup by category should return the correct pattern."""
p = get_pattern_by_category(GlitchCategory.FLOATING_ASSETS)
self.assertIsNotNone(p)
self.assertEqual(p.category, GlitchCategory.FLOATING_ASSETS)
missing = get_pattern_by_category("nonexistent_category_value")
self.assertIsNone(missing)
def test_build_vision_prompt(self):
"""Vision prompt should contain pattern names and be non-trivial."""
prompt = build_vision_prompt()
self.assertGreater(len(prompt), 200)
self.assertIn("Floating Object", prompt)
self.assertIn("Z-Fighting", prompt)
self.assertIn("Missing", prompt)
# Three.js patterns should be included
self.assertIn("Shader Compilation Failure", prompt)
self.assertIn("Bloom Overflow", prompt)
def test_build_vision_prompt_subset(self):
"""Vision prompt with subset should only include specified patterns."""
subset = MATRIX_GLITCH_PATTERNS[:3]
prompt = build_vision_prompt(subset)
self.assertIn(subset[0].name, prompt)
self.assertNotIn(MATRIX_GLITCH_PATTERNS[-1].name, prompt)
class TestGlitchDetector(unittest.TestCase):
"""Tests for matrix_glitch_detector module."""
def test_generate_scan_angles_default(self):
"""Default 4 angles should return front, right, back, left."""
angles = generate_scan_angles(4)
self.assertEqual(len(angles), 4)
labels = [a["label"] for a in angles]
self.assertIn("front", labels)
self.assertIn("right", labels)
self.assertIn("back", labels)
self.assertIn("left", labels)
def test_generate_scan_angles_many(self):
"""Requesting more angles than base should still return correct count."""
angles = generate_scan_angles(12)
self.assertEqual(len(angles), 12)
# Should still have the standard ones
labels = [a["label"] for a in angles]
self.assertIn("front", labels)
def test_generate_scan_angles_few(self):
"""Requesting fewer angles should return fewer."""
angles = generate_scan_angles(2)
self.assertEqual(len(angles), 2)
def test_detected_glitch_dataclass(self):
"""DetectedGlitch should serialize cleanly."""
g = DetectedGlitch(
id="test001",
category="floating_assets",
name="Test Glitch",
description="A test glitch",
severity="high",
confidence=0.85,
location_x=50.0,
location_y=30.0,
screenshot_index=0,
screenshot_angle="front",
)
self.assertEqual(g.id, "test001")
self.assertTrue(g.timestamp) # Auto-generated
def test_infer_severity_critical(self):
"""Missing textures should infer critical/high severity."""
sev = _infer_severity("missing_textures", 0.9)
self.assertEqual(sev, "critical")
sev_low = _infer_severity("missing_textures", 0.5)
self.assertEqual(sev_low, "high")
def test_infer_severity_floating(self):
"""Floating assets should infer high/medium severity."""
sev = _infer_severity("floating_assets", 0.8)
self.assertEqual(sev, "high")
sev_low = _infer_severity("floating_assets", 0.5)
self.assertEqual(sev_low, "medium")
def test_infer_severity_default(self):
"""Unknown categories should default to medium/low."""
sev = _infer_severity("unknown_thing", 0.7)
self.assertEqual(sev, "medium")
sev_low = _infer_severity("unknown_thing", 0.3)
self.assertEqual(sev_low, "low")
def test_parse_vision_response_json_array(self):
"""Should parse a JSON array response."""
response = json.dumps([
{
"category": "floating_assets",
"name": "Float Test",
"description": "Chair floating",
"confidence": 0.9,
"severity": "high",
"location_x": 40,
"location_y": 60,
}
])
glitches = _parse_vision_response(response, 0, "front")
self.assertEqual(len(glitches), 1)
self.assertEqual(glitches[0].category, "floating_assets")
self.assertAlmostEqual(glitches[0].confidence, 0.9)
def test_parse_vision_response_wrapped(self):
"""Should parse a response with 'glitches' wrapper key."""
response = json.dumps({
"glitches": [
{
"category": "z_fighting",
"name": "Shimmer",
"confidence": 0.6,
}
]
})
glitches = _parse_vision_response(response, 1, "right")
self.assertEqual(len(glitches), 1)
self.assertEqual(glitches[0].category, "z_fighting")
def test_parse_vision_response_empty(self):
"""Should return empty list for non-JSON text."""
glitches = _parse_vision_response("No glitches found.", 0, "front")
self.assertEqual(len(glitches), 0)
def test_parse_vision_response_code_block(self):
"""Should extract JSON from markdown code blocks."""
response = '```json\n[{"category": "clipping", "name": "Clip", "confidence": 0.7}]\n```'
glitches = _parse_vision_response(response, 0, "front")
self.assertEqual(len(glitches), 1)
def test_build_report(self):
"""Report should have correct summary statistics."""
angles = generate_scan_angles(4)
screenshots = [Path(f"/tmp/ss_{i}.png") for i in range(4)]
glitches = [
DetectedGlitch(
id="a", category="floating_assets", name="Float",
description="", severity="high", confidence=0.8,
screenshot_index=0, screenshot_angle="front",
),
DetectedGlitch(
id="b", category="missing_textures", name="Missing",
description="", severity="critical", confidence=0.95,
screenshot_index=1, screenshot_angle="right",
),
]
report = build_report("https://test.com", angles, screenshots, glitches)
self.assertEqual(report.total_screenshots, 4)
self.assertEqual(len(report.glitches), 2)
self.assertEqual(report.summary["total_glitches"], 2)
self.assertEqual(report.summary["by_severity"]["critical"], 1)
self.assertEqual(report.summary["by_severity"]["high"], 1)
self.assertEqual(report.summary["by_category"]["floating_assets"], 1)
self.assertEqual(report.metadata["reference"], "timmy-config#491")
def test_build_report_json_roundtrip(self):
"""Report JSON should parse back correctly."""
angles = generate_scan_angles(2)
screenshots = [Path(f"/tmp/ss_{i}.png") for i in range(2)]
report = build_report("https://test.com", angles, screenshots, [])
json_str = report.to_json()
parsed = json.loads(json_str)
self.assertEqual(parsed["url"], "https://test.com")
self.assertEqual(parsed["total_screenshots"], 2)
def test_run_demo(self):
"""Demo mode should produce a report with simulated glitches."""
with tempfile.NamedTemporaryFile(suffix=".json", delete=False) as f:
output_path = Path(f.name)
try:
report = run_demo(output_path)
self.assertEqual(len(report.glitches), 6) # 4 original + 2 Three.js
self.assertGreater(report.summary["total_glitches"], 0)
self.assertTrue(output_path.exists())
# Verify the saved JSON is valid
saved = json.loads(output_path.read_text())
self.assertIn("scan_id", saved)
self.assertIn("glitches", saved)
finally:
output_path.unlink(missing_ok=True)
class TestThreeJsPatterns(unittest.TestCase):
"""Tests for Three.js-specific glitch patterns (timmy-config#543)."""
def test_get_threejs_patterns_returns_only_threejs(self):
"""get_threejs_patterns() should return only Three.js categories."""
patterns = get_threejs_patterns()
self.assertEqual(len(patterns), 6)
for p in patterns:
self.assertIn(p.category, THREEJS_CATEGORIES)
def test_threejs_patterns_have_required_fields(self):
"""All Three.js patterns must have valid fields."""
for p in get_threejs_patterns():
self.assertIsInstance(p.category, GlitchCategory)
self.assertTrue(p.name)
self.assertTrue(p.description)
self.assertIsInstance(p.severity, GlitchSeverity)
self.assertGreater(len(p.detection_prompts), 0)
self.assertGreater(len(p.visual_indicators), 0)
def test_shader_failure_is_critical(self):
"""Shader compilation failure should be CRITICAL severity."""
p = get_pattern_by_category(GlitchCategory.SHADER_FAILURE)
self.assertIsNotNone(p)
self.assertEqual(p.severity, GlitchSeverity.CRITICAL)
def test_texture_placeholder_is_critical(self):
"""Texture placeholder (1x1 white) should be CRITICAL severity."""
p = get_pattern_by_category(GlitchCategory.TEXTURE_PLACEHOLDER)
self.assertIsNotNone(p)
self.assertEqual(p.severity, GlitchSeverity.CRITICAL)
def test_infer_severity_shader_failure(self):
"""Shader failure should infer critical/high."""
self.assertEqual(_infer_severity("shader_failure", 0.8), "critical")
self.assertEqual(_infer_severity("shader_failure", 0.5), "high")
def test_infer_severity_texture_placeholder(self):
"""Texture placeholder should infer critical/high."""
self.assertEqual(_infer_severity("texture_placeholder", 0.8), "critical")
self.assertEqual(_infer_severity("texture_placeholder", 0.5), "high")
def test_infer_severity_uv_mapping(self):
"""UV mapping error should infer high/medium."""
self.assertEqual(_infer_severity("uv_mapping_error", 0.8), "high")
self.assertEqual(_infer_severity("uv_mapping_error", 0.5), "medium")
def test_infer_severity_frustum_culling(self):
"""Frustum culling should infer medium/low."""
self.assertEqual(_infer_severity("frustum_culling", 0.7), "medium")
self.assertEqual(_infer_severity("frustum_culling", 0.4), "low")
def test_infer_severity_shadow_map(self):
"""Shadow map artifact should infer medium/low."""
self.assertEqual(_infer_severity("shadow_map_artifact", 0.7), "medium")
self.assertEqual(_infer_severity("shadow_map_artifact", 0.4), "low")
def test_infer_severity_bloom_overflow(self):
"""Bloom overflow should infer medium/low (default path)."""
self.assertEqual(_infer_severity("bloom_overflow", 0.7), "medium")
self.assertEqual(_infer_severity("bloom_overflow", 0.4), "low")
def test_threejs_patterns_in_vision_prompt(self):
"""Three.js patterns should appear in the composite vision prompt."""
prompt = build_vision_prompt()
self.assertIn("shader_failure", prompt)
self.assertIn("texture_placeholder", prompt)
self.assertIn("uv_mapping_error", prompt)
self.assertIn("frustum_culling", prompt)
self.assertIn("shadow_map_artifact", prompt)
self.assertIn("bloom_overflow", prompt)
def test_threejs_subset_prompt(self):
"""Building prompt from Three.js-only patterns should work."""
threejs = get_threejs_patterns()
prompt = build_vision_prompt(threejs)
self.assertIn("Shader Compilation Failure", prompt)
self.assertNotIn("Floating Object", prompt) # generic, not Three.js
def test_report_metadata_version(self):
"""Report metadata should reference both issues."""
report = run_demo()
self.assertEqual(report.metadata["detector_version"], "0.2.0")
self.assertIn("543", report.metadata["reference"])
class TestIntegration(unittest.TestCase):
"""Integration-level tests."""
def test_full_pipeline_demo(self):
"""End-to-end demo pipeline should complete without errors."""
report = run_demo()
self.assertIsNotNone(report.scan_id)
self.assertTrue(report.timestamp)
self.assertGreater(report.total_screenshots, 0)
def test_patterns_cover_matrix_themes(self):
"""Patterns should cover the main Matrix glitch themes."""
category_values = {p.category.value for p in MATRIX_GLITCH_PATTERNS}
expected = {"floating_assets", "z_fighting", "missing_textures", "clipping", "broken_normals"}
self.assertTrue(expected.issubset(category_values))
def test_patterns_cover_threejs_themes(self):
"""Patterns should cover Three.js-specific glitch themes (#543)."""
category_values = {p.category.value for p in MATRIX_GLITCH_PATTERNS}
threejs_expected = {"shader_failure", "texture_placeholder", "uv_mapping_error",
"frustum_culling", "shadow_map_artifact", "bloom_overflow"}
self.assertTrue(threejs_expected.issubset(category_values))
if __name__ == "__main__":
unittest.main()

View File

@@ -0,0 +1,148 @@
#!/usr/bin/env python3
"""Tests for matrix_glitch_detect.py — verifies detection and HTML report logic."""
import json
import sys
import tempfile
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
from matrix_glitch_detect import (
Severity, Glitch, GlitchReport,
format_report, generate_html_report, _parse_json_response,
)
def test_parse_json_clean():
result = _parse_json_response('{"glitches": [], "overall_quality": 95}')
assert result["overall_quality"] == 95
print(" PASS: test_parse_json_clean")
def test_parse_json_fenced():
result = _parse_json_response('```json\n{"overall_quality": 80}\n```')
assert result["overall_quality"] == 80
print(" PASS: test_parse_json_fenced")
def test_parse_json_garbage():
assert _parse_json_response("no json") == {}
print(" PASS: test_parse_json_garbage")
def test_glitch_dataclass():
g = Glitch(type="z_fighting", severity=Severity.MAJOR, region="center", description="Shimmer", confidence=0.8)
assert g.type == "z_fighting"
assert g.confidence == 0.8
print(" PASS: test_glitch_dataclass")
def test_report_dataclass():
r = GlitchReport(source="test.png", status="WARN", score=75)
r.glitches.append(Glitch(type="float", severity=Severity.MINOR))
assert len(r.glitches) == 1
assert r.score == 75
print(" PASS: test_report_dataclass")
def test_format_json():
r = GlitchReport(source="test.png", status="PASS", score=90, summary="Clean")
r.glitches.append(Glitch(type="cosmetic", severity=Severity.COSMETIC, description="Minor"))
output = format_report(r, "json")
parsed = json.loads(output)
assert parsed["status"] == "PASS"
assert len(parsed["glitches"]) == 1
print(" PASS: test_format_json")
def test_format_text():
r = GlitchReport(source="test.png", status="FAIL", score=30, summary="Critical glitch")
r.glitches.append(Glitch(type="render_failure", severity=Severity.CRITICAL, description="Black screen"))
output = format_report(r, "text")
assert "FAIL" in output
assert "render_failure" in output
print(" PASS: test_format_text")
def test_html_report_basic():
r = GlitchReport(source="test.png", status="PASS", score=100)
html = generate_html_report([r], title="Test Report")
assert "<!DOCTYPE html>" in html
assert "Test Report" in html
assert "PASS" in html
assert "100" in html
print(" PASS: test_html_report_basic")
def test_html_report_with_glitches():
r = GlitchReport(source="test.png", status="FAIL", score=40)
r.glitches.append(Glitch(type="z_fighting", severity=Severity.CRITICAL, region="center", description="Heavy flicker", confidence=0.9))
r.glitches.append(Glitch(type="clipping", severity=Severity.MINOR, region="bottom", description="Object through floor", confidence=0.6))
html = generate_html_report([r], title="Glitch Report")
assert "z_fighting" in html
assert "CRITICAL" in html
assert "clipping" in html
assert "Heavy flicker" in html
print(" PASS: test_html_report_with_glitches")
def test_html_report_multi():
r1 = GlitchReport(source="a.png", status="PASS", score=95)
r2 = GlitchReport(source="b.png", status="WARN", score=70)
r2.glitches.append(Glitch(type="texture_pop", severity=Severity.MAJOR))
html = generate_html_report([r1, r2])
assert "a.png" in html
assert "b.png" in html
assert "2" in html # 2 screenshots
print(" PASS: test_html_report_multi")
def test_html_self_contained():
r = GlitchReport(source="test.png", status="PASS", score=100)
html = generate_html_report([r])
assert "external" not in html.lower() or "no external dependencies" in html.lower()
assert "<style>" in html # Inline CSS
print(" PASS: test_html_self_contained")
def test_missing_image():
r = GlitchReport(source="/nonexistent/image.png")
# detect_glitches would set FAIL — simulate
r.status = "FAIL"
r.score = 0
r.summary = "File not found"
assert r.status == "FAIL"
print(" PASS: test_missing_image")
def test_severity_enum():
assert Severity.CRITICAL.value == "critical"
assert Severity.MAJOR.value == "major"
print(" PASS: test_severity_enum")
def run_all():
print("=== matrix_glitch_detect tests ===")
tests = [
test_parse_json_clean, test_parse_json_fenced, test_parse_json_garbage,
test_glitch_dataclass, test_report_dataclass,
test_format_json, test_format_text,
test_html_report_basic, test_html_report_with_glitches,
test_html_report_multi, test_html_self_contained,
test_missing_image, test_severity_enum,
]
passed = failed = 0
for t in tests:
try:
t()
passed += 1
except Exception as e:
print(f" FAIL: {t.__name__}{e}")
failed += 1
print(f"\n{'ALL PASSED' if failed == 0 else f'{failed} FAILED'}: {passed}/{len(tests)}")
return failed == 0
if __name__ == "__main__":
sys.exit(0 if run_all() else 1)

View File

@@ -0,0 +1,123 @@
#!/usr/bin/env python3
"""Tests for nexus_smoke_test.py — verifies smoke test logic."""
import json
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
from nexus_smoke_test import (
Severity, SmokeCheck, SmokeResult,
format_result, _parse_json_response,
)
def test_parse_json_clean():
result = _parse_json_response('{"status": "PASS", "summary": "ok"}')
assert result["status"] == "PASS"
print(" PASS: test_parse_json_clean")
def test_parse_json_fenced():
result = _parse_json_response('```json\n{"status": "FAIL"}\n```')
assert result["status"] == "FAIL"
print(" PASS: test_parse_json_fenced")
def test_parse_json_garbage():
result = _parse_json_response("no json here")
assert result == {}
print(" PASS: test_parse_json_garbage")
def test_smoke_check_dataclass():
c = SmokeCheck(name="Test", status=Severity.PASS, message="All good")
assert c.name == "Test"
assert c.status == Severity.PASS
print(" PASS: test_smoke_check_dataclass")
def test_smoke_result_dataclass():
r = SmokeResult(url="https://example.com", status=Severity.PASS)
r.checks.append(SmokeCheck(name="Page Loads", status=Severity.PASS))
assert len(r.checks) == 1
assert r.url == "https://example.com"
print(" PASS: test_smoke_result_dataclass")
def test_format_json():
r = SmokeResult(url="https://test.com", status=Severity.PASS, summary="All good", duration_ms=100)
r.checks.append(SmokeCheck(name="Test", status=Severity.PASS, message="OK"))
output = format_result(r, "json")
parsed = json.loads(output)
assert parsed["status"] == "pass"
assert parsed["url"] == "https://test.com"
assert len(parsed["checks"]) == 1
print(" PASS: test_format_json")
def test_format_text():
r = SmokeResult(url="https://test.com", status=Severity.WARN, summary="1 warning", duration_ms=200)
r.checks.append(SmokeCheck(name="Screenshot", status=Severity.WARN, message="No backend"))
output = format_result(r, "text")
assert "NEXUS VISUAL SMOKE TEST" in output
assert "https://test.com" in output
assert "WARN" in output
print(" PASS: test_format_text")
def test_format_text_pass():
r = SmokeResult(url="https://test.com", status=Severity.PASS, summary="All clear")
r.checks.append(SmokeCheck(name="Page Loads", status=Severity.PASS, message="HTTP 200"))
r.checks.append(SmokeCheck(name="HTML Content", status=Severity.PASS, message="Valid"))
output = format_result(r, "text")
assert "" in output
assert "Page Loads" in output
print(" PASS: test_format_text")
def test_severity_enum():
assert Severity.PASS.value == "pass"
assert Severity.FAIL.value == "fail"
assert Severity.WARN.value == "warn"
print(" PASS: test_severity_enum")
def test_overall_status_logic():
# All pass
r = SmokeResult()
r.checks = [SmokeCheck(name="a", status=Severity.PASS), SmokeCheck(name="b", status=Severity.PASS)]
fails = sum(1 for c in r.checks if c.status == Severity.FAIL)
warns = sum(1 for c in r.checks if c.status == Severity.WARN)
assert fails == 0 and warns == 0
# One fail
r.checks.append(SmokeCheck(name="c", status=Severity.FAIL))
fails = sum(1 for c in r.checks if c.status == Severity.FAIL)
assert fails == 1
print(" PASS: test_overall_status_logic")
def run_all():
print("=== nexus_smoke_test tests ===")
tests = [
test_parse_json_clean, test_parse_json_fenced, test_parse_json_garbage,
test_smoke_check_dataclass, test_smoke_result_dataclass,
test_format_json, test_format_text, test_format_text_pass,
test_severity_enum, test_overall_status_logic,
]
passed = failed = 0
for t in tests:
try:
t()
passed += 1
except Exception as e:
print(f" FAIL: {t.__name__}{e}")
failed += 1
print(f"\n{'ALL PASSED' if failed == 0 else f'{failed} FAILED'}: {passed}/{len(tests)}")
return failed == 0
if __name__ == "__main__":
sys.exit(0 if run_all() else 1)

View File

@@ -0,0 +1,100 @@
{"song": "Thunder Road", "artist": "Heartland", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "The screen door slams, Mary's dress waves", "scene": {"mood": "hope", "colors": ["gold", "sky blue", "white"], "composition": "wide shot", "camera": "static", "description": "Open horizon. Golden light breaking through clouds. The figure silhouetted against dawn. The screen door slams, Mary's dress waves"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "Like a vision she dances across the porch as the radio plays", "scene": {"mood": "anticipation", "colors": ["silver", "pale green", "cream"], "composition": "close-up", "camera": "slow pan", "description": "Close on hands gripping a steering wheel. Dashboard lights reflecting in eyes. Road stretching ahead. Like a vision she dances across the porch as the radio plays"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "Roy Orbison singing for the lonely, hey that's me and I want you only", "scene": {"mood": "energy", "colors": ["red", "orange", "electric blue"], "composition": "over the shoulder", "camera": "dolly in", "description": "Rapid cuts. Bodies in motion. Light streaks across the frame. Roy Orbison singing for the lonely, hey that's me and I want you only"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "Don't turn me home out now I'm so young and worthless still", "scene": {"mood": "triumph", "colors": ["gold", "crimson", "white"], "composition": "low angle", "camera": "dolly out", "description": "Wide shot. Figure standing on a hilltop. Arms raised. City lights below. Don't turn me home out now I'm so young and worthless still"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "The night's busting open these two lanes will take us anywhere", "scene": {"mood": "nostalgia", "colors": ["amber", "sepia", "dusty rose"], "composition": "high angle", "camera": "handheld", "description": "Sepia tones. A photograph come to life. Dust motes in afternoon light. The night's busting open these two lanes will take us anywhere"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "We got one last chance to make it real", "scene": {"mood": "urgency", "colors": ["red", "black", "strobe white"], "composition": "dutch angle", "camera": "steadicam", "description": "Handheld camera running. Blurred faces. Traffic. Heartbeat sound design. We got one last chance to make it real"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "To trade in these wings on some wheels", "scene": {"mood": "passion", "colors": ["deep red", "burgundy", "gold"], "composition": "symmetrical", "camera": "slow zoom", "description": "Extreme close-up. Skin. Breath visible in cold air. Eyes locked. To trade in these wings on some wheels"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "Climb in back, heaven's waiting down the tracks", "scene": {"mood": "defiance", "colors": ["black", "neon green", "chrome"], "composition": "rule of thirds", "camera": "crane up", "description": "Low angle. Figure standing against the wind. Debris flying past. Unmoved. Climb in back, heaven's waiting down the tracks"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "Oh oh oh oh oh oh oh", "scene": {"mood": "release", "colors": ["sky blue", "white", "pale gold"], "composition": "extreme wide", "camera": "tracking shot", "description": "Slow motion. Something falling \u2014 a mask, a chain, a weight. Lightness follows. Oh oh oh oh oh oh oh"}}
{"song": "Thunder Road", "artist": "Heartland", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "It's a town full of losers and I'm pulling out of here to win", "scene": {"mood": "catharsis", "colors": ["all white", "silver", "clear"], "composition": "medium shot", "camera": "slow tilt down", "description": "White space expanding. Figure dissolving into light. Peace in the dissolution. It's a town full of losers and I'm pulling out of here to win"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "Woke up on the floor again, whiskey still on my tongue", "scene": {"mood": "despair", "colors": ["navy", "black", "grey"], "composition": "wide shot", "camera": "static", "description": "Empty room. Single light source. Figure curled in corner. Rain on windows. Woke up on the floor again, whiskey still on my tongue"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "The mirror shows a stranger and the damage that I've done", "scene": {"mood": "anger", "colors": ["red", "black", "orange"], "composition": "close-up", "camera": "slow pan", "description": "Shattered glass. Red light. Hands clenched. Jaw tight. The frame vibrates. The mirror shows a stranger and the damage that I've done"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "I scream until my throat bleeds but nobody comes", "scene": {"mood": "frenzy", "colors": ["strobe", "red", "white flash"], "composition": "over the shoulder", "camera": "dolly in", "description": "Strobe lighting. Multiple exposures. Bodies colliding. Chaos as composition. I scream until my throat bleeds but nobody comes"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "The walls are closing in again, the ceiling pressing down", "scene": {"mood": "exhaustion", "colors": ["grey", "brown", "faded"], "composition": "low angle", "camera": "dolly out", "description": "Static shot. Figure slumped. Eyes half-closed. Time passing in shadows. The walls are closing in again, the ceiling pressing down"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "I tried to call your number but you changed it years ago", "scene": {"mood": "resignation", "colors": ["grey", "muted blue", "beige"], "composition": "high angle", "camera": "handheld", "description": "Medium shot. Hands dropping keys on a table. Turning away. Not looking back. I tried to call your number but you changed it years ago"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "Now I'm howling at the moon like some rabid dog I know", "scene": {"mood": "grief", "colors": ["deep purple", "black", "silver"], "composition": "dutch angle", "camera": "steadicam", "description": "Wide shot. Figure alone in vast space. Dark purple sky. No horizon line. Now I'm howling at the moon like some rabid dog I know"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "Every bone remembers what my mind wants to forget", "scene": {"mood": "numbness", "colors": ["white", "grey", "no color"], "composition": "symmetrical", "camera": "slow zoom", "description": "Desaturated. Figure staring at nothing. World moving around them in blur. Every bone remembers what my mind wants to forget"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "I'll tear this whole house down before the sun comes up", "scene": {"mood": "rage", "colors": ["fire red", "black", "ember orange"], "composition": "rule of thirds", "camera": "crane up", "description": "Red wash. Extreme close-up on eyes. Fire reflected in pupils. I'll tear this whole house down before the sun comes up"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "Ash and ruin everywhere, this is all that's left", "scene": {"mood": "acceptance", "colors": ["soft blue", "warm grey", "sage"], "composition": "extreme wide", "camera": "tracking shot", "description": "Soft focus. Gentle light. Figure breathing. The camera doesn't judge. Ash and ruin everywhere, this is all that's left"}}
{"song": "Black Dog Howl", "artist": "Rust & Wire", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Silence. Just the wind through broken glass.", "scene": {"mood": "silence", "colors": ["black", "void", "faint starlight"], "composition": "medium shot", "camera": "slow tilt down", "description": "Black screen. Faint starlight. The sound drops out completely. Silence. Just the wind through broken glass."}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "Ten thousand miles of static between your voice and mine", "scene": {"mood": "wonder", "colors": ["aurora green", "violet", "silver"], "composition": "wide shot", "camera": "static", "description": "Northern lights overhead. Figure looking up. Mouth open. Child's expression. Ten thousand miles of static between your voice and mine"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "I trace your constellation on the dashboard every night", "scene": {"mood": "isolation", "colors": ["cold blue", "black", "distant starlight"], "composition": "close-up", "camera": "slow pan", "description": "Extreme wide. Single figure. Vast empty landscape. Scale crushing. I trace your constellation on the dashboard every night"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "The signal fades to nothing but I keep the frequency", "scene": {"mood": "longing", "colors": ["teal", "silver", "moonlight"], "composition": "over the shoulder", "camera": "dolly in", "description": "Through a window. Figure on the other side. Glass between. Breath on the pane. The signal fades to nothing but I keep the frequency"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "Then suddenly your laughter breaks through like a summer storm", "scene": {"mood": "connection", "colors": ["warm gold", "rose", "blush"], "composition": "low angle", "camera": "dolly out", "description": "Two hands reaching. Fingers almost touching. Warm light between them. Then suddenly your laughter breaks through like a summer storm"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "We're dancing in the data stream, our pixels intertwined", "scene": {"mood": "euphoria", "colors": ["neon", "rainbow", "white flash"], "composition": "high angle", "camera": "handheld", "description": "Overexposed. Everything bright. Dancing. The frame can't contain the joy. We're dancing in the data stream, our pixels intertwined"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "But I can't tell if you're real or just a ghost in the machine", "scene": {"mood": "confusion", "colors": ["swirling", "unsettled", "green-grey"], "composition": "dutch angle", "camera": "steadicam", "description": "Multiple focal points. Nothing sharp. The viewer doesn't know where to look. But I can't tell if you're real or just a ghost in the machine"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "The picture clears and there you are \u2014 imperfect, warm, alive", "scene": {"mood": "clarity", "colors": ["clear blue", "white", "crisp"], "composition": "symmetrical", "camera": "slow zoom", "description": "Rack focus. Background blurs, foreground sharpens. Suddenly everything makes sense. The picture clears and there you are \u2014 imperfect, warm, alive"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "Your hand reaches through the screen, I swear I feel the heat", "scene": {"mood": "tenderness", "colors": ["blush pink", "warm cream", "soft gold"], "composition": "rule of thirds", "camera": "crane up", "description": "Close on a hand touching a face. Soft light. Shallow depth of field. Your hand reaches through the screen, I swear I feel the heat"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "The bandwidth's dying, say it now before the link goes dark", "scene": {"mood": "urgency", "colors": ["red", "black", "strobe white"], "composition": "extreme wide", "camera": "tracking shot", "description": "Handheld camera running. Blurred faces. Traffic. Heartbeat sound design. The bandwidth's dying, say it now before the link goes dark"}}
{"song": "Satellite Hearts", "artist": "Neon Circuit", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Goodnight, satellite heart. I'll find you in the static.", "scene": {"mood": "bittersweet", "colors": ["amber", "lavender", "fading light"], "composition": "medium shot", "camera": "slow tilt down", "description": "Amber light fading. A smile that's also a goodbye. Beautiful and sad at once. Goodnight, satellite heart. I'll find you in the static."}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "They paved over every green thing when the developers came", "scene": {"mood": "oppression", "colors": ["concrete grey", "brown", "exhaust fume yellow"], "composition": "wide shot", "camera": "static", "description": "Concrete. Overpasses. No sky visible. Figures small against infrastructure. They paved over every green thing when the developers came"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "But we planted seeds between the cracks and gave them all a name", "scene": {"mood": "resilience", "colors": ["green", "cracked concrete", "gold"], "composition": "close-up", "camera": "slow pan", "description": "Crack in pavement. Green shoot pushing through. Macro lens. But we planted seeds between the cracks and gave them all a name"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "The mayor says progress looks like demolition and dust", "scene": {"mood": "anger", "colors": ["red", "black", "orange"], "composition": "over the shoulder", "camera": "dolly in", "description": "Shattered glass. Red light. Hands clenched. Jaw tight. The frame vibrates. The mayor says progress looks like demolition and dust"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "But a dandelion broke through the asphalt this morning \u2014 that's us", "scene": {"mood": "beauty", "colors": ["wildflower colors", "green", "sunlight"], "composition": "low angle", "camera": "dolly out", "description": "Wildflowers in unexpected places. Color against grey. Nature reclaiming. But a dandelion broke through the asphalt this morning \u2014 that's us"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "You can't kill what wants to live, can't silence what must sing", "scene": {"mood": "defiance", "colors": ["black", "neon green", "chrome"], "composition": "high angle", "camera": "handheld", "description": "Low angle. Figure standing against the wind. Debris flying past. Unmoved. You can't kill what wants to live, can't silence what must sing"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "We're the roots beneath the road, we're the birds that built on string", "scene": {"mood": "community", "colors": ["warm tones", "string lights", "firelight"], "composition": "dutch angle", "camera": "steadicam", "description": "String lights. People gathered. Laughter out of focus. Warmth as visual language. We're the roots beneath the road, we're the birds that built on string"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "When they tear the next block down we'll be dancing in the rubble", "scene": {"mood": "joy", "colors": ["bright", "multi", "saturated"], "composition": "symmetrical", "camera": "slow zoom", "description": "Saturated color. Wide smiles. Arms open. The world in full bloom. When they tear the next block down we'll be dancing in the rubble"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "Every protest is a garden, every march plants something new", "scene": {"mood": "struggle", "colors": ["dust", "grey", "hard light"], "composition": "rule of thirds", "camera": "crane up", "description": "Close on hands working. Calluses. Dust. Effort visible in every frame. Every protest is a garden, every march plants something new"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "The concrete is a drum and our footsteps keep the beat", "scene": {"mood": "growth", "colors": ["green", "brown", "morning light"], "composition": "extreme wide", "camera": "tracking shot", "description": "Time-lapse. Seed to flower. Sunrise to sunset. Transformation as rhythm. The concrete is a drum and our footsteps keep the beat"}}
{"song": "Concrete Garden", "artist": "Streetlight Prophet", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Tomorrow there'll be flowers where they swore there'd only be defeat", "scene": {"mood": "hope", "colors": ["gold", "sky blue", "white"], "composition": "medium shot", "camera": "slow tilt down", "description": "Open horizon. Golden light breaking through clouds. The figure silhouetted against dawn. Tomorrow there'll be flowers where they swore there'd only be defeat"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "I felt the pull before I saw the edge", "scene": {"mood": "dread", "colors": ["void black", "deep red", "cold white"], "composition": "wide shot", "camera": "static", "description": "Corner of frame. Something in the periphery. Dark. The camera doesn't look directly. I felt the pull before I saw the edge"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "The stars bent sideways, light itself was dead", "scene": {"mood": "fascination", "colors": ["event horizon purple", "gravitational lens blue"], "composition": "close-up", "camera": "slow pan", "description": "Close on eyes. Reflection of something impossible. The pupil expands. The stars bent sideways, light itself was dead"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "I could have turned the ship around but something in me said stay", "scene": {"mood": "surrender", "colors": ["white", "dissolution", "prismatic"], "composition": "over the shoulder", "camera": "dolly in", "description": "Arms opening. Head back. Falling backward into something vast. I could have turned the ship around but something in me said stay"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "The event horizon glows like a halo made of nothing", "scene": {"mood": "awe", "colors": ["starfield", "nebula colors", "infinite dark"], "composition": "low angle", "camera": "dolly out", "description": "Wide shot of cosmos. Nebula. Stars being born. Human figure tiny at bottom. The event horizon glows like a halo made of nothing"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "Time stretches thin as wire, each second takes a year", "scene": {"mood": "terror", "colors": ["black", "red shift", "distortion"], "composition": "high angle", "camera": "handheld", "description": "Shaking camera. Red shift. Something approaching fast. The frame distorts. Time stretches thin as wire, each second takes a year"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "I am both the observer and the thing that disappears", "scene": {"mood": "peace", "colors": ["deep space black", "starlight", "calm"], "composition": "dutch angle", "camera": "steadicam", "description": "Still water. Stars reflected. Perfect mirror. No movement. No sound. I am both the observer and the thing that disappears"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "My body reads the tidal forces like sheet music played on bone", "scene": {"mood": "disorientation", "colors": ["warped", "chromatic aberration", "bent light"], "composition": "symmetrical", "camera": "slow zoom", "description": "Warped lens. Vertigo. Walls becoming floor. Gravity is a suggestion. My body reads the tidal forces like sheet music played on bone"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "I stop fighting, stop reaching, stop calling home", "scene": {"mood": "acceptance", "colors": ["soft blue", "warm grey", "sage"], "composition": "rule of thirds", "camera": "crane up", "description": "Soft focus. Gentle light. Figure breathing. The camera doesn't judge. I stop fighting, stop reaching, stop calling home"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "There is a peace in dissolution I was never meant to know", "scene": {"mood": "transcendence", "colors": ["pure white", "beyond visible", "golden"], "composition": "extreme wide", "camera": "tracking shot", "description": "Pure white expanding. Figure becoming light. Boundaries dissolving. There is a peace in dissolution I was never meant to know"}}
{"song": "Gravity Well", "artist": "Void Walker", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Singularity. Silence. Everything and nothing both at once.", "scene": {"mood": "emptiness", "colors": ["void", "absolute black", "nothing"], "composition": "medium shot", "camera": "slow tilt down", "description": "Absolute black. No stars. No reference point. The void looking back. Singularity. Silence. Everything and nothing both at once."}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "My father's hands smelled like machine oil and prayer", "scene": {"mood": "nostalgia", "colors": ["amber", "sepia", "dusty rose"], "composition": "wide shot", "camera": "static", "description": "Sepia tones. A photograph come to life. Dust motes in afternoon light. My father's hands smelled like machine oil and prayer"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "The factory whistle was our clock, the shift was our calendar", "scene": {"mood": "sadness", "colors": ["grey", "rain", "muted blue"], "composition": "close-up", "camera": "slow pan", "description": "Rain on glass. Grey light. A cup of tea going cold. Still life of loss. The factory whistle was our clock, the shift was our calendar"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "He'd come home at midnight, wake me up to say goodnight", "scene": {"mood": "tenderness", "colors": ["blush pink", "warm cream", "soft gold"], "composition": "over the shoulder", "camera": "dolly in", "description": "Close on a hand touching a face. Soft light. Shallow depth of field. He'd come home at midnight, wake me up to say goodnight"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "Now the mill is just a skeleton and he's been gone ten years", "scene": {"mood": "loss", "colors": ["faded", "dusty", "empty space"], "composition": "low angle", "camera": "dolly out", "description": "Empty chair. Dust settling. A coat still on a hook. Presence of absence. Now the mill is just a skeleton and he's been gone ten years"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "But the river still runs brown with memory and rust", "scene": {"mood": "beauty", "colors": ["wildflower colors", "green", "sunlight"], "composition": "high angle", "camera": "handheld", "description": "Wildflowers in unexpected places. Color against grey. Nature reclaiming. But the river still runs brown with memory and rust"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "I found his lunchbox in the attic, coffee stains still fresh", "scene": {"mood": "resignation", "colors": ["grey", "muted blue", "beige"], "composition": "dutch angle", "camera": "steadicam", "description": "Medium shot. Hands dropping keys on a table. Turning away. Not looking back. I found his lunchbox in the attic, coffee stains still fresh"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "Some things don't decay \u2014 they just learn to hold still", "scene": {"mood": "love", "colors": ["neutral"], "composition": "symmetrical", "camera": "slow zoom", "description": "Visual interpretation of: Some things don't decay \u2014 they just learn to hold still"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "I hum the songs he hummed to me though I've forgotten half the words", "scene": {"mood": "weariness", "colors": ["grey-brown", "faded", "dim"], "composition": "rule of thirds", "camera": "crane up", "description": "Slow movement. Heavy eyelids. The world in faded tones. Everything too much. I hum the songs he hummed to me though I've forgotten half the words"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "The town's half-empty but the porch lights still come on at dusk", "scene": {"mood": "quiet hope", "colors": ["faint warm light", "candle glow", "dawn grey"], "composition": "extreme wide", "camera": "tracking shot", "description": "Faint warm light. Candle in dark room. Just enough to see by. The town's half-empty but the porch lights still come on at dusk"}}
{"song": "Rust Belt Lullaby", "artist": "Iron & Ember", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Sleep now, rust belt baby. The furnace keeps us warm.", "scene": {"mood": "peace", "colors": ["deep space black", "starlight", "calm"], "composition": "medium shot", "camera": "slow tilt down", "description": "Still water. Stars reflected. Perfect mirror. No movement. No sound. Sleep now, rust belt baby. The furnace keeps us warm."}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "I didn't start the fire but I brought the gasoline", "scene": {"mood": "fury", "colors": ["dark red", "black", "flash"], "composition": "wide shot", "camera": "static", "description": "Dark red wash. Hands destroying. Frame shaking with rage. I didn't start the fire but I brought the gasoline"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "Every sermon needs a spark and every spark needs a dream", "scene": {"mood": "ecstasy", "colors": ["fire", "gold", "blinding white"], "composition": "close-up", "camera": "slow pan", "description": "Fire and gold. Bodies arching. Light bursting from every surface. Every sermon needs a spark and every spark needs a dream"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "The forest is a cathedral and the flames are choir boys singing", "scene": {"mood": "chaos", "colors": ["strobe", "fragmented", "clashing"], "composition": "over the shoulder", "camera": "dolly in", "description": "Fragmented frame. Collage. Everything at once. Order is a memory. The forest is a cathedral and the flames are choir boys singing"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "Watch the old world burn \u2014 isn't the light beautiful?", "scene": {"mood": "joy", "colors": ["bright", "multi", "saturated"], "composition": "low angle", "camera": "dolly out", "description": "Saturated color. Wide smiles. Arms open. The world in full bloom. Watch the old world burn \u2014 isn't the light beautiful?"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "We'll dance in the embers, we'll make love in the ash", "scene": {"mood": "destruction", "colors": ["fire", "ash", "smoke orange"], "composition": "high angle", "camera": "handheld", "description": "Fire. Ash falling like snow. Structures collapsing. Beautiful in its terrible way. We'll dance in the embers, we'll make love in the ash"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "From destruction comes the soil where new things grow at last", "scene": {"mood": "creation", "colors": ["green", "light", "warm gold"], "composition": "dutch angle", "camera": "steadicam", "description": "Hands shaping clay. Light emerging from dark. Something new being born. From destruction comes the soil where new things grow at last"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "But don't mistake the warmth for safety, don't mistake the glow for home", "scene": {"mood": "warning", "colors": ["red flash", "amber", "siren"], "composition": "symmetrical", "camera": "slow zoom", "description": "Red flash. Siren light. The calm before. Then: impact. But don't mistake the warmth for safety, don't mistake the glow for home"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "Come closer, come closer \u2014 I promise the burning feels like flying", "scene": {"mood": "invitation", "colors": ["warm", "open", "golden"], "composition": "rule of thirds", "camera": "crane up", "description": "Open door. Warm light spilling out. A hand extended. Come in. Come closer, come closer \u2014 I promise the burning feels like flying"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "We threw everything we owned into the blaze and laughed", "scene": {"mood": "abandon", "colors": ["wild", "free", "untethered"], "composition": "extreme wide", "camera": "tracking shot", "description": "Running through a field. Hair wild. No destination. Just movement. We threw everything we owned into the blaze and laughed"}}
{"song": "Wildfire Sermon", "artist": "Prophet Ash", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Morning. Smoke. Green shoots. Begin again.", "scene": {"mood": "rebirth", "colors": ["green shoots", "dawn", "clear"], "composition": "medium shot", "camera": "slow tilt down", "description": "Dawn. Green shoots in ash. First breath after drowning. Morning. Smoke. Green shoots. Begin again."}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "There's a voice on the radio that shouldn't be there", "scene": {"mood": "mystery", "colors": ["deep blue", "shadow", "candle"], "composition": "wide shot", "camera": "static", "description": "Shadow figure in doorway. Candle. Face half-lit. Eyes knowing. There's a voice on the radio that shouldn't be there"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "Speaking my name in a language I almost understand", "scene": {"mood": "loneliness", "colors": ["single light", "dark", "cold blue"], "composition": "close-up", "camera": "slow pan", "description": "Single light in vast dark. Figure beneath it. Nothing else. Speaking my name in a language I almost understand"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "I turn the dial but it follows like a shadow made of sound", "scene": {"mood": "curiosity", "colors": ["warm yellow", "spotlight", "discovery"], "composition": "over the shoulder", "camera": "dolly in", "description": "Light moving across a surface. Discovery. Eyes widening. I turn the dial but it follows like a shadow made of sound"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "Then it says something only I would know, something buried deep", "scene": {"mood": "connection", "colors": ["warm gold", "rose", "blush"], "composition": "low angle", "camera": "dolly out", "description": "Two hands reaching. Fingers almost touching. Warm light between them. Then it says something only I would know, something buried deep"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "I'm not afraid anymore \u2014 I'm listening", "scene": {"mood": "paranoia", "colors": ["surveillance green", "strobe", "red"], "composition": "high angle", "camera": "handheld", "description": "Surveillance angles. Green tint. Multiple screens. Watching. Being watched. I'm not afraid anymore \u2014 I'm listening"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "The voice knows my dreams, it describes them back to me", "scene": {"mood": "intimacy", "colors": ["candlelight", "warm", "close"], "composition": "dutch angle", "camera": "steadicam", "description": "Candlelight only. Two faces close. Shared breath. The world outside forgotten. The voice knows my dreams, it describes them back to me"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "We're having a conversation across some membrane I can't see", "scene": {"mood": "urgency", "colors": ["red", "black", "strobe white"], "composition": "symmetrical", "camera": "slow zoom", "description": "Handheld camera running. Blurred faces. Traffic. Heartbeat sound design. We're having a conversation across some membrane I can't see"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "Then static. Then nothing. Then a whisper: find me", "scene": {"mood": "disconnection", "colors": ["static", "grey", "broken signal"], "composition": "rule of thirds", "camera": "crane up", "description": "Static. Snow on screen. A voice breaking up. Distance measured in noise. Then static. Then nothing. Then a whisper: find me"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "I search every frequency but the voice is gone", "scene": {"mood": "searching", "colors": ["flashlight beam", "dark", "moving light"], "composition": "extreme wide", "camera": "tracking shot", "description": "Flashlight beam cutting dark. Moving. Looking. Not finding yet. I search every frequency but the voice is gone"}}
{"song": "Midnight Transmission", "artist": "Frequency Ghost", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Some nights I still hear it, faint, like a song in another room", "scene": {"mood": "haunting", "colors": ["faint blue", "echo", "silver"], "composition": "medium shot", "camera": "slow tilt down", "description": "Faint blue light. Echo of a figure. Present and absent simultaneously. Some nights I still hear it, faint, like a song in another room"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "I wore your love like a weapon and you never felt the blade", "scene": {"mood": "seduction", "colors": ["deep red", "velvet", "candlelight"], "composition": "wide shot", "camera": "static", "description": "Deep red. Velvet textures. Slow movement. Eyes that promise. I wore your love like a weapon and you never felt the blade"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "Every kiss was a negotiation, every touch a trade", "scene": {"mood": "power", "colors": ["gold", "black", "crimson"], "composition": "close-up", "camera": "slow pan", "description": "Throne. Gold. Black. The figure doesn't move. Doesn't need to. Every kiss was a negotiation, every touch a trade"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "The throne room smells like jasmine and someone else's fear", "scene": {"mood": "cruelty", "colors": ["cold silver", "black", "sharp white"], "composition": "over the shoulder", "camera": "dolly in", "description": "Silver blade. Cold light. A smile that doesn't reach the eyes. The throne room smells like jasmine and someone else's fear"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "I am beautiful when I'm angry \u2014 haven't you heard?", "scene": {"mood": "beauty", "colors": ["wildflower colors", "green", "sunlight"], "composition": "low angle", "camera": "dolly out", "description": "Wildflowers in unexpected places. Color against grey. Nature reclaiming. I am beautiful when I'm angry \u2014 haven't you heard?"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "Don't mistake my gentleness for weakness, darling", "scene": {"mood": "danger", "colors": ["red", "black", "warning yellow"], "composition": "high angle", "camera": "handheld", "description": "Red and black. Warning signs. The frame contracts. Something approaches. Don't mistake my gentleness for weakness, darling"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "I chose to be kind. I could burn this kingdom down.", "scene": {"mood": "vulnerability", "colors": ["soft", "exposed", "raw"], "composition": "dutch angle", "camera": "steadicam", "description": "Exposed skin. Soft light. Eyes open. Trust visible in every pore. I chose to be kind. I could burn this kingdom down."}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "The roses in my crown have thorns that curve inward", "scene": {"mood": "fury", "colors": ["dark red", "black", "flash"], "composition": "symmetrical", "camera": "slow zoom", "description": "Dark red wash. Hands destroying. Frame shaking with rage. The roses in my crown have thorns that curve inward"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "I bleed for my own sins, not for yours", "scene": {"mood": "grace", "colors": ["white", "silver", "flowing"], "composition": "rule of thirds", "camera": "crane up", "description": "White. Flowing. Movement without effort. The body as art. I bleed for my own sins, not for yours"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "Tonight I lay the crown aside and sleep without armor", "scene": {"mood": "revenge", "colors": ["dark", "steel", "cold blue"], "composition": "extreme wide", "camera": "tracking shot", "description": "Cold blue. Steel. The plan unfolding in shadows. Patience as weapon. Tonight I lay the crown aside and sleep without armor"}}
{"song": "Crown of Thorns and Roses", "artist": "Velvet Guillotine", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Mercy. The hardest word. The only gift worth giving.", "scene": {"mood": "mercy", "colors": ["warm gold", "white", "gentle"], "composition": "medium shot", "camera": "slow tilt down", "description": "Warm gold. Hand lowering a weapon. Choosing not to. The harder path. Mercy. The hardest word. The only gift worth giving."}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 1, "timestamp": "0:00", "duration": "30s", "lyric_line": "Four walls, one window, a view of another wall", "scene": {"mood": "claustrophobia", "colors": ["close walls", "yellow bulb", "cramped"], "composition": "wide shot", "camera": "static", "description": "Walls close. Ceiling low. Yellow bulb. No escape visible. Four walls, one window, a view of another wall"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 2, "timestamp": "0:30", "duration": "30s", "lyric_line": "The radiator clicks like a metronome for the damned", "scene": {"mood": "routine", "colors": ["grey", "institutional", "fluorescent"], "composition": "close-up", "camera": "slow pan", "description": "Fluorescent light. Same motion repeated. Clock on the wall. Time as loop. The radiator clicks like a metronome for the damned"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 3, "timestamp": "1:00", "duration": "30s", "lyric_line": "I've memorized every crack in the ceiling \u2014 they form a map", "scene": {"mood": "desperation", "colors": ["scratching", "clawing", "raw"], "composition": "over the shoulder", "camera": "dolly in", "description": "Hands clawing. Fingernails against surface. Raw need. Nothing held back. I've memorized every crack in the ceiling \u2014 they form a map"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 4, "timestamp": "1:30", "duration": "30s", "lyric_line": "In my mind I've left a hundred times, bought a farm, learned to fly", "scene": {"mood": "fantasy", "colors": ["dreamy", "pastel", "floating"], "composition": "low angle", "camera": "dolly out", "description": "Pastel. Floating. Impossible architecture. Gravity optional. In my mind I've left a hundred times, bought a farm, learned to fly"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 5, "timestamp": "2:00", "duration": "30s", "lyric_line": "Then one morning I open the door and just walk out", "scene": {"mood": "breakthrough", "colors": ["white burst", "open sky", "blinding"], "composition": "high angle", "camera": "handheld", "description": "White burst. Wall shattering. Open sky beyond. Freedom as explosion. Then one morning I open the door and just walk out"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 6, "timestamp": "2:30", "duration": "30s", "lyric_line": "The hallway is an ocean, the stairs are a mountain range", "scene": {"mood": "freedom", "colors": ["open sky", "blue", "green"], "composition": "dutch angle", "camera": "steadicam", "description": "Open road. Blue sky. Green fields. Wind in hair. No walls. The hallway is an ocean, the stairs are a mountain range"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 7, "timestamp": "3:00", "duration": "30s", "lyric_line": "The street hits me like cold water and I almost go back", "scene": {"mood": "fear", "colors": ["cold", "dark", "sharp"], "composition": "symmetrical", "camera": "slow zoom", "description": "Cold. Dark. Sharp edges. The frame contracts. Something unseen. The street hits me like cold water and I almost go back"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 8, "timestamp": "3:30", "duration": "30s", "lyric_line": "But the sky \u2014 have you seen the sky? It goes on forever", "scene": {"mood": "joy", "colors": ["bright", "multi", "saturated"], "composition": "rule of thirds", "camera": "crane up", "description": "Saturated color. Wide smiles. Arms open. The world in full bloom. But the sky \u2014 have you seen the sky? It goes on forever"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 9, "timestamp": "4:00", "duration": "30s", "lyric_line": "I stand on the sidewalk and cry because the world is so big", "scene": {"mood": "grounding", "colors": ["neutral"], "composition": "extreme wide", "camera": "tracking shot", "description": "Visual interpretation of: I stand on the sidewalk and cry because the world is so big"}}
{"song": "Apartment 4B", "artist": "Wallpaper & Wire", "beat": 10, "timestamp": "4:30", "duration": "30s", "lyric_line": "Home is not a place. Home is the moment you stop hiding.", "scene": {"mood": "home", "colors": ["neutral"], "composition": "medium shot", "camera": "slow tilt down", "description": "Visual interpretation of: Home is not a place. Home is the moment you stop hiding."}}

129
training/scripts/augment_pairs.py Executable file
View File

@@ -0,0 +1,129 @@
#!/usr/bin/env python3
"""
augment_pairs.py — Training data augmentation: paraphrase and translate.
Usage:
python3 augment_pairs.py --input data.jsonl
python3 augment_pairs.py --input data.jsonl --paraphrases 3 --langs es,fr,de
python3 augment_pairs.py --input data.jsonl --llm-endpoint http://localhost:11434/v1
"""
import json, os, sys, re, random
from pathlib import Path
random.seed(42)
PARAPHRASE_TRANSFORMS = [
lambda s: re.sub(r"(\w+), (\w+)", r"\2, \1", s, count=1),
lambda s: f"A beautifully rendered scene: {s[0].lower()}{s[1:]}" if len(s) > 10 else s,
lambda s: s.replace("A ", "The ").replace("An ", "The ") if s.startswith(("A ", "An ")) else f"Here, {s[0].lower()}{s[1:]}",
lambda s: f"In a cinematic frame: {s}" if len(s) > 20 else s,
lambda s: s if ", " not in s else ", ".join(s.split(", ")[:2]),
]
TRANSLATIONS = {
"es": {"the":"el","a":"un","is":"es","in":"en","of":"de","and":"y","with":"con","scene":"escena","light":"luz","dark":"oscuro","warm":"cálido","rain":"lluvia","sun":"sol","moon":"luna","sky":"cielo","forest":"bosque","mountain":"montaña","ocean":"océano","golden":"dorado","blue":"azul","red":"rojo","green":"verde","silence":"silencio","dream":"sueño","love":"amor","hope":"esperanza","fear":"miedo","joy":"alegría","peace":"paz","beautiful":"hermoso","sad":"triste","shadow":"sombra","color":"color","silver":"plateado","white":"blanco","black":"negro","portray":"retrato"},
"fr": {"the":"le","a":"un","is":"est","in":"dans","of":"de","and":"et","with":"avec","scene":"scène","light":"lumière","dark":"sombre","warm":"chaud","rain":"pluie","sun":"soleil","moon":"lune","sky":"ciel","forest":"forêt","mountain":"montagne","ocean":"océan","golden":"doré","blue":"bleu","red":"rouge","green":"vert","silence":"silence","dream":"rêve","love":"amour","hope":"espoir","fear":"peur","joy":"joie","peace":"paix","beautiful":"beau","sad":"triste","shadow":"ombre","color":"couleur","silver":"argenté","white":"blanc","black":"noir"},
"de": {"the":"der","a":"ein","is":"ist","in":"in","of":"von","and":"und","with":"mit","scene":"Szene","light":"Licht","dark":"dunkel","warm":"warm","rain":"Regen","sun":"Sonne","moon":"Mond","sky":"Himmel","forest":"Wald","mountain":"Berg","ocean":"Ozean","golden":"golden","blue":"blau","red":"rot","green":"grün","silence":"Stille","dream":"Traum","love":"Liebe","hope":"Hoffnung","fear":"Angst","joy":"Freude","peace":"Frieden","beautiful":"schön","sad":"traurig","shadow":"Schatten","color":"Farbe","silver":"silbern","white":"weiß","black":"schwarz"},
}
LANG_NAMES = {"es": "Spanish", "fr": "French", "de": "German"}
def detect_text_field(entry):
for f in ["rich","terse","text","content","lyric_line","description","scene_description","prompt","scene"]:
if f in entry and isinstance(entry[f], str) and len(entry[f]) > 5:
return f
for k, v in entry.items():
if isinstance(v, str) and len(v) > 5:
return k
return None
def paraphrase(text):
t = random.choice(PARAPHRASE_TRANSFORMS)(text)
if t == text:
t = text.replace(" and ", " & ").replace(" with ", " alongside ")
if t == text:
t = f"In this scene: {text[0].lower()}{text[1:]}" if text[0].isupper() else text
return t
def translate(text, lang):
d = TRANSLATIONS.get(lang, {})
words = text.split()
out = []
for w in words:
lo = w.lower().strip(".,;:!?")
suf = w[len(w.rstrip(".,;:!?")):]
if lo in d:
out.append(d[lo] + suf)
else:
out.append(w)
return " ".join(out)
def augment_file(input_path, output_path=None, n_para=3, langs=None, llm_endpoint=None):
input_path = Path(input_path)
if output_path is None:
output_path = input_path.parent / f"{input_path.stem}_augmented{input_path.suffix}"
entries = [json.loads(l) for l in open(input_path) if l.strip()]
if not entries:
print(f"No entries in {input_path}"); return 0
tf = detect_text_field(entries[0])
if not tf:
print(f"ERROR: No text field in {input_path}", file=sys.stderr); return 0
print(f"Input: {input_path} ({len(entries)} entries, field={tf})")
aug_count = 0
with open(output_path, "w") as out:
for e in entries:
out.write(json.dumps(e, ensure_ascii=False) + "\n")
for i, e in enumerate(entries):
text = e[tf]
# Paraphrases
for p in range(n_para):
para = paraphrase(text)
if para != text:
ne = dict(e); ne[tf] = para
ne["_augmentation"] = f"paraphrase_{p+1}"
ne["_original"] = text[:100]
out.write(json.dumps(ne, ensure_ascii=False) + "\n")
aug_count += 1
# Translations
for lang in (langs or []):
tr = translate(text, lang)
if tr != text:
ne = dict(e); ne[tf] = tr
ne["_augmentation"] = f"translate_{lang}"
ne["_language"] = lang
ne["_original"] = text[:100]
out.write(json.dumps(ne, ensure_ascii=False) + "\n")
aug_count += 1
if (i+1) % 100 == 0:
print(f" {i+1}/{len(entries)} done ({aug_count} augmented)")
total = len(entries) + aug_count
print(f"Done: {len(entries)} originals + {aug_count} augmented = {total}")
print(f"Output: {output_path}")
return aug_count
def main():
import argparse
p = argparse.ArgumentParser()
p.add_argument("--input", required=True)
p.add_argument("--output", default=None)
p.add_argument("--paraphrases", type=int, default=3)
p.add_argument("--langs", default="es,fr,de")
p.add_argument("--llm-endpoint", default=None)
args = p.parse_args()
langs = [l.strip() for l in args.langs.split(",") if l.strip()] if args.langs else []
augment_file(args.input, args.output, args.paraphrases, langs, args.llm_endpoint)
if __name__ == "__main__":
main()