feat: workspace isolation + honest success metrics (#186)
## Workspace Isolation No agent touches ~/Timmy-Time-dashboard anymore. Each agent gets a fully isolated clone under /tmp/timmy-agents/ with its own port, data directory, and TIMMY_HOME. - scripts/agent_workspace.sh: init, reset, branch, destroy per agent - Loop prompt updated: workspace paths replace worktree paths - Smoke tests run in isolated /tmp/timmy-agents/smoke/repo ## Honest Success Metrics Cycle success now requires BOTH hermes clean exit AND main green (smoke test passes). Tracks main_green_rate separately from hermes_clean_rate in summary.json. Follows from PR #162 (triage + retro system). Co-authored-by: Kimi Agent <kimi@timmy.local> Reviewed-on: http://localhost:3000/rockachopa/Timmy-time-dashboard/pulls/186 Co-authored-by: hermes <hermes@timmy.local> Co-committed-by: hermes <hermes@timmy.local>
This commit was merged in pull request #186.
This commit is contained in:
rockachopa
245
scripts/agent_workspace.sh
Normal file
245
scripts/agent_workspace.sh
Normal file
@@ -0,0 +1,245 @@
|
||||
#!/usr/bin/env bash
|
||||
# ── Agent Workspace Manager ────────────────────────────────────────────
|
||||
# Creates and maintains fully isolated environments per agent.
|
||||
# ~/Timmy-Time-dashboard is SACRED — never touched by agents.
|
||||
#
|
||||
# Each agent gets:
|
||||
# - Its own git clone (from Gitea, not the local repo)
|
||||
# - Its own port range (no collisions)
|
||||
# - Its own data/ directory (databases, files)
|
||||
# - Its own TIMMY_HOME (approvals.db, etc.)
|
||||
# - Shared Ollama backend (single GPU, shared inference)
|
||||
# - Shared Gitea (single source of truth for issues/PRs)
|
||||
#
|
||||
# Layout:
|
||||
# /tmp/timmy-agents/
|
||||
# hermes/ — Hermes loop orchestrator
|
||||
# repo/ — git clone
|
||||
# home/ — TIMMY_HOME (approvals.db, etc.)
|
||||
# env.sh — source this for agent's env vars
|
||||
# kimi-0/ — Kimi pane 0
|
||||
# repo/
|
||||
# home/
|
||||
# env.sh
|
||||
# ...
|
||||
# smoke/ — dedicated for smoke-testing main
|
||||
# repo/
|
||||
# home/
|
||||
# env.sh
|
||||
#
|
||||
# Usage:
|
||||
# agent_workspace.sh init <agent> — create or refresh
|
||||
# agent_workspace.sh reset <agent> — hard reset to origin/main
|
||||
# agent_workspace.sh branch <agent> <br> — fresh branch from main
|
||||
# agent_workspace.sh path <agent> — print repo path
|
||||
# agent_workspace.sh env <agent> — print env.sh path
|
||||
# agent_workspace.sh init-all — init all workspaces
|
||||
# agent_workspace.sh destroy <agent> — remove workspace entirely
|
||||
# ───────────────────────────────────────────────────────────────────────
|
||||
|
||||
set -o pipefail
|
||||
|
||||
CANONICAL="$HOME/Timmy-Time-dashboard"
|
||||
AGENTS_DIR="/tmp/timmy-agents"
|
||||
GITEA_REMOTE="http://localhost:3000/rockachopa/Timmy-time-dashboard.git"
|
||||
TOKEN_FILE="$HOME/.hermes/gitea_token"
|
||||
|
||||
# ── Port allocation (each agent gets a unique range) ──────────────────
|
||||
# Dashboard ports: 8100, 8101, 8102, ... (avoids real dashboard on 8000)
|
||||
# Serve ports: 8200, 8201, 8202, ...
|
||||
agent_index() {
|
||||
case "$1" in
|
||||
hermes) echo 0 ;; kimi-0) echo 1 ;; kimi-1) echo 2 ;;
|
||||
kimi-2) echo 3 ;; kimi-3) echo 4 ;; smoke) echo 9 ;;
|
||||
*) echo 0 ;;
|
||||
esac
|
||||
}
|
||||
|
||||
get_dashboard_port() { echo $(( 8100 + $(agent_index "$1") )); }
|
||||
get_serve_port() { echo $(( 8200 + $(agent_index "$1") )); }
|
||||
|
||||
log() { echo "[workspace] $*"; }
|
||||
|
||||
# ── Get authenticated remote URL ──────────────────────────────────────
|
||||
get_remote_url() {
|
||||
if [ -f "$TOKEN_FILE" ]; then
|
||||
local token=""
|
||||
token=$(cat "$TOKEN_FILE" 2>/dev/null || true)
|
||||
if [ -n "$token" ]; then
|
||||
echo "http://hermes:${token}@localhost:3000/rockachopa/Timmy-time-dashboard.git"
|
||||
return
|
||||
fi
|
||||
fi
|
||||
echo "$GITEA_REMOTE"
|
||||
}
|
||||
|
||||
# ── Create env.sh for an agent ────────────────────────────────────────
|
||||
write_env() {
|
||||
local agent="$1"
|
||||
local ws="$AGENTS_DIR/$agent"
|
||||
local repo="$ws/repo"
|
||||
local home="$ws/home"
|
||||
local dash_port=$(get_dashboard_port "$agent")
|
||||
local serve_port=$(get_serve_port "$agent")
|
||||
|
||||
cat > "$ws/env.sh" << EOF
|
||||
# Auto-generated agent environment — source this before running Timmy
|
||||
# Agent: $agent
|
||||
|
||||
export TIMMY_WORKSPACE="$repo"
|
||||
export TIMMY_HOME="$home"
|
||||
export TIMMY_AGENT_NAME="$agent"
|
||||
|
||||
# Ports (isolated per agent)
|
||||
export PORT=$dash_port
|
||||
export TIMMY_SERVE_PORT=$serve_port
|
||||
|
||||
# Ollama (shared — single GPU)
|
||||
export OLLAMA_URL="http://localhost:11434"
|
||||
|
||||
# Gitea (shared — single source of truth)
|
||||
export GITEA_URL="http://localhost:3000"
|
||||
|
||||
# Test mode defaults
|
||||
export TIMMY_TEST_MODE=1
|
||||
export TIMMY_DISABLE_CSRF=1
|
||||
export TIMMY_SKIP_EMBEDDINGS=1
|
||||
|
||||
# Override data paths to stay inside the clone
|
||||
export TIMMY_DATA_DIR="$repo/data"
|
||||
export TIMMY_BRAIN_DB="$repo/data/brain.db"
|
||||
|
||||
# Working directory
|
||||
cd "$repo"
|
||||
EOF
|
||||
|
||||
chmod +x "$ws/env.sh"
|
||||
}
|
||||
|
||||
# ── Init ──────────────────────────────────────────────────────────────
|
||||
init_workspace() {
|
||||
local agent="$1"
|
||||
local ws="$AGENTS_DIR/$agent"
|
||||
local repo="$ws/repo"
|
||||
local home="$ws/home"
|
||||
local remote
|
||||
remote=$(get_remote_url)
|
||||
|
||||
mkdir -p "$ws" "$home"
|
||||
|
||||
if [ -d "$repo/.git" ]; then
|
||||
log "$agent: refreshing existing clone..."
|
||||
cd "$repo"
|
||||
git remote set-url origin "$remote" 2>/dev/null
|
||||
git fetch origin --prune --quiet 2>/dev/null
|
||||
git checkout main --quiet 2>/dev/null
|
||||
git reset --hard origin/main --quiet 2>/dev/null
|
||||
git clean -fdx -e data/ --quiet 2>/dev/null
|
||||
else
|
||||
log "$agent: cloning from Gitea..."
|
||||
git clone "$remote" "$repo" --quiet 2>/dev/null
|
||||
cd "$repo"
|
||||
git fetch origin --prune --quiet 2>/dev/null
|
||||
fi
|
||||
|
||||
# Ensure data directory exists
|
||||
mkdir -p "$repo/data"
|
||||
|
||||
# Write env file
|
||||
write_env "$agent"
|
||||
|
||||
log "$agent: ready at $repo (port $(get_dashboard_port "$agent"))"
|
||||
}
|
||||
|
||||
# ── Reset ─────────────────────────────────────────────────────────────
|
||||
reset_workspace() {
|
||||
local agent="$1"
|
||||
local repo="$AGENTS_DIR/$agent/repo"
|
||||
|
||||
if [ ! -d "$repo/.git" ]; then
|
||||
init_workspace "$agent"
|
||||
return
|
||||
fi
|
||||
|
||||
cd "$repo"
|
||||
git merge --abort 2>/dev/null || true
|
||||
git rebase --abort 2>/dev/null || true
|
||||
git cherry-pick --abort 2>/dev/null || true
|
||||
git fetch origin --prune --quiet 2>/dev/null
|
||||
git checkout main --quiet 2>/dev/null
|
||||
git reset --hard origin/main --quiet 2>/dev/null
|
||||
git clean -fdx -e data/ --quiet 2>/dev/null
|
||||
|
||||
log "$agent: reset to origin/main"
|
||||
}
|
||||
|
||||
# ── Branch ────────────────────────────────────────────────────────────
|
||||
branch_workspace() {
|
||||
local agent="$1"
|
||||
local branch="$2"
|
||||
local repo="$AGENTS_DIR/$agent/repo"
|
||||
|
||||
if [ ! -d "$repo/.git" ]; then
|
||||
init_workspace "$agent"
|
||||
fi
|
||||
|
||||
cd "$repo"
|
||||
git fetch origin --prune --quiet 2>/dev/null
|
||||
git branch -D "$branch" 2>/dev/null || true
|
||||
git checkout -b "$branch" origin/main --quiet 2>/dev/null
|
||||
|
||||
log "$agent: on branch $branch (from origin/main)"
|
||||
}
|
||||
|
||||
# ── Path ──────────────────────────────────────────────────────────────
|
||||
print_path() {
|
||||
echo "$AGENTS_DIR/$1/repo"
|
||||
}
|
||||
|
||||
print_env() {
|
||||
echo "$AGENTS_DIR/$1/env.sh"
|
||||
}
|
||||
|
||||
# ── Init all ──────────────────────────────────────────────────────────
|
||||
init_all() {
|
||||
for agent in hermes kimi-0 kimi-1 kimi-2 kimi-3 smoke; do
|
||||
init_workspace "$agent"
|
||||
done
|
||||
log "All workspaces initialized."
|
||||
echo ""
|
||||
echo " Agent Port Path"
|
||||
echo " ────── ──── ────"
|
||||
for agent in hermes kimi-0 kimi-1 kimi-2 kimi-3 smoke; do
|
||||
printf " %-9s %d %s\n" "$agent" "$(get_dashboard_port "$agent")" "$AGENTS_DIR/$agent/repo"
|
||||
done
|
||||
}
|
||||
|
||||
# ── Destroy ───────────────────────────────────────────────────────────
|
||||
destroy_workspace() {
|
||||
local agent="$1"
|
||||
local ws="$AGENTS_DIR/$agent"
|
||||
if [ -d "$ws" ]; then
|
||||
rm -rf "$ws"
|
||||
log "$agent: destroyed"
|
||||
else
|
||||
log "$agent: nothing to destroy"
|
||||
fi
|
||||
}
|
||||
|
||||
# ── CLI dispatch ──────────────────────────────────────────────────────
|
||||
case "${1:-help}" in
|
||||
init) init_workspace "${2:?Usage: $0 init <agent>}" ;;
|
||||
reset) reset_workspace "${2:?Usage: $0 reset <agent>}" ;;
|
||||
branch) branch_workspace "${2:?Usage: $0 branch <agent> <branch>}" \
|
||||
"${3:?Usage: $0 branch <agent> <branch>}" ;;
|
||||
path) print_path "${2:?Usage: $0 path <agent>}" ;;
|
||||
env) print_env "${2:?Usage: $0 env <agent>}" ;;
|
||||
init-all) init_all ;;
|
||||
destroy) destroy_workspace "${2:?Usage: $0 destroy <agent>}" ;;
|
||||
*)
|
||||
echo "Usage: $0 {init|reset|branch|path|env|init-all|destroy} [agent] [branch]"
|
||||
echo ""
|
||||
echo "Agents: hermes, kimi-0, kimi-1, kimi-2, kimi-3, smoke"
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
@@ -4,14 +4,25 @@
|
||||
Called after each cycle completes (success or failure).
|
||||
Appends a structured entry to .loop/retro/cycles.jsonl.
|
||||
|
||||
SUCCESS DEFINITION:
|
||||
A cycle is only "success" if BOTH conditions are met:
|
||||
1. The hermes process exited cleanly (exit code 0)
|
||||
2. Main is green (smoke test passes on main after merge)
|
||||
|
||||
A cycle that merges a PR but leaves main red is a FAILURE.
|
||||
The --main-green flag records the smoke test result.
|
||||
|
||||
Usage:
|
||||
python3 scripts/cycle_retro.py --cycle 42 --success --issue 85 \
|
||||
python3 scripts/cycle_retro.py --cycle 42 --success --main-green --issue 85 \
|
||||
--type bug --duration 480 --tests-passed 1450 --tests-added 3 \
|
||||
--files-changed 2 --lines-added 45 --lines-removed 12 \
|
||||
--kimi-panes 2 --pr 155
|
||||
|
||||
python3 scripts/cycle_retro.py --cycle 43 --failure --issue 90 \
|
||||
--type feature --duration 1200 --reason "tox failed: 3 errors"
|
||||
|
||||
python3 scripts/cycle_retro.py --cycle 44 --success --no-main-green \
|
||||
--reason "PR merged but tests fail on main"
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
@@ -51,6 +62,10 @@ def parse_args() -> argparse.Namespace:
|
||||
p.add_argument("--pr", type=int, default=None, help="PR number if merged")
|
||||
p.add_argument("--reason", type=str, default="", help="Failure reason")
|
||||
p.add_argument("--notes", type=str, default="", help="Free-form observations")
|
||||
p.add_argument("--main-green", action="store_true", default=False,
|
||||
help="Smoke test passed on main after this cycle")
|
||||
p.add_argument("--no-main-green", dest="main_green", action="store_false",
|
||||
help="Smoke test failed or was not run")
|
||||
|
||||
return p.parse_args()
|
||||
|
||||
@@ -73,6 +88,8 @@ def update_summary() -> None:
|
||||
|
||||
successes = [e for e in recent if e.get("success")]
|
||||
failures = [e for e in recent if not e.get("success")]
|
||||
main_green_count = sum(1 for e in recent if e.get("main_green"))
|
||||
hermes_clean_count = sum(1 for e in recent if e.get("hermes_clean", e.get("success")))
|
||||
durations = [e["duration"] for e in recent if e.get("duration", 0) > 0]
|
||||
|
||||
# Per-type stats
|
||||
@@ -103,6 +120,8 @@ def update_summary() -> None:
|
||||
"window": len(recent),
|
||||
"total_cycles": len(entries),
|
||||
"success_rate": round(len(successes) / len(recent), 2) if recent else 0,
|
||||
"main_green_rate": round(main_green_count / len(recent), 2) if recent else 0,
|
||||
"hermes_clean_rate": round(hermes_clean_count / len(recent), 2) if recent else 0,
|
||||
"avg_duration_seconds": round(sum(durations) / len(durations)) if durations else 0,
|
||||
"total_lines_added": sum(e.get("lines_added", 0) for e in recent),
|
||||
"total_lines_removed": sum(e.get("lines_removed", 0) for e in recent),
|
||||
@@ -121,12 +140,17 @@ def update_summary() -> None:
|
||||
def main() -> None:
|
||||
args = parse_args()
|
||||
|
||||
# A cycle is only truly successful if hermes exited clean AND main is green
|
||||
truly_success = args.success and args.main_green
|
||||
|
||||
entry = {
|
||||
"timestamp": datetime.now(timezone.utc).isoformat(),
|
||||
"cycle": args.cycle,
|
||||
"issue": args.issue,
|
||||
"type": args.type,
|
||||
"success": args.success,
|
||||
"success": truly_success,
|
||||
"hermes_clean": args.success,
|
||||
"main_green": args.main_green,
|
||||
"duration": args.duration,
|
||||
"tests_passed": args.tests_passed,
|
||||
"tests_added": args.tests_added,
|
||||
@@ -135,7 +159,7 @@ def main() -> None:
|
||||
"lines_removed": args.lines_removed,
|
||||
"kimi_panes": args.kimi_panes,
|
||||
"pr": args.pr,
|
||||
"reason": args.reason if args.failure else "",
|
||||
"reason": args.reason if (args.failure or not args.main_green) else "",
|
||||
"notes": args.notes,
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user