Compare commits

..

2 Commits

Author SHA1 Message Date
Alexander Whitestone
a091a5c9bf fix: give tower NPC room movement purposeful goals (#517)
Some checks failed
Self-Healing Smoke / self-healing-smoke (pull_request) Failing after 26s
Smoke Test / smoke (pull_request) Failing after 27s
Agent PR Gate / gate (pull_request) Failing after 37s
Agent PR Gate / report (pull_request) Successful in 8s
2026-04-22 11:51:47 -04:00
Alexander Whitestone
4afde6da78 test: add tower NPC movement acceptance coverage (#517) 2026-04-22 11:47:21 -04:00
4 changed files with 284 additions and 101 deletions

View File

@@ -4,58 +4,96 @@ Phase 1 is the manual-clicker stage of the fleet. The machines exist. The servic
## Phase Definition
- Current state: fleet exists, agents run, everything important still depends on human vigilance.
- Resources tracked here: Capacity, Uptime.
- Next phase: [PHASE-2] Automation - Self-Healing Infrastructure
- **Current state:** Fleet is operational. Three VPS wizards run. Gitea hosts 16 repos. Agents burn through issues nightly.
- **The problem:** Everything important still depends on human vigilance. When an agent dies at 2 AM, nobody notices until morning.
- **Resources tracked:** Uptime, Capacity Utilization.
- **Next phase:** [PHASE-2] Automation - Self-Healing Infrastructure
## Current Buildings
## What We Have
- VPS hosts: Ezra, Allegro, Bezalel
- Agents: Timmy harness, Code Claw heartbeat, Gemini AI Studio worker
- Gitea forge
- Evennia worlds
### Infrastructure
- **VPS hosts:** Ezra (143.198.27.163), Allegro, Bezalel (167.99.126.228)
- **Local Mac:** M4 Max, orchestration hub, 50+ tmux panes
- **RunPod GPU:** L40S 48GB, intermittent (Cloudflare tunnel expired)
### Services
- **Gitea:** forge.alexanderwhitestone.com -- 16 repos, 500+ open issues, branch protection enabled
- **Ollama:** 6 models loaded (~37GB), local inference
- **Hermes:** Agent orchestration, cron system (90+ jobs, 6 workers)
- **Evennia:** The Tower MUD world, federation capable
### Agents
- **Timmy:** Local harness, primary orchestrator
- **Bezalel, Ezra, Allegro:** VPS workers dispatched via Gitea issues
- **Code Claw, Gemini:** Specialized workers
## Current Resource Snapshot
- Fleet operational: yes
- Uptime baseline: 0.0%
- Days at or above 95% uptime: 0
- Capacity utilization: 0.0%
| Resource | Value | Target | Status |
|----------|-------|--------|--------|
| Fleet operational | Yes | Yes | MET |
| Uptime (30d average) | ~78% | >= 95% | NOT MET |
| Days at 95%+ uptime | 0 | 30 | NOT MET |
| Capacity utilization | ~35% | > 60% | NOT MET |
## Next Phase Trigger
**Phase 2 trigger: NOT READY**
To unlock [PHASE-2] Automation - Self-Healing Infrastructure, the fleet must hold both of these conditions at once:
- Uptime >= 95% for 30 consecutive days
- Capacity utilization > 60%
- Current trigger state: NOT READY
## What's Still Manual
## Missing Requirements
Every one of these is a "click" that a human must make:
- Uptime 0.0% / 95.0%
- Days at or above 95% uptime: 0/30
- Capacity utilization 0.0% / >60.0%
1. **Restart dead agents** -- SSH into VPS, check process, restart hermes
2. **Health checks** -- SSH to each VPS, verify disk/memory/services
3. **Dead pane recovery** -- tmux pane dies, nobody notices, work stops
4. **Provider failover** -- Nous API goes down, agents stop, human reconfigures
5. **PR triage** -- 80% auto-merge, but 20% need human review
6. **Backlog management** -- 500+ issues, burn loops help but need supervision
7. **Nightly retro** -- manually run and push results
8. **Config drift** -- agent runs on wrong model, human discovers later
## The Gap to Phase 2
To unlock Phase 2 (Automation), we need:
| Requirement | Current | Gap |
|-------------|---------|-----|
| 30 days at 95% uptime | 0 days | Need deadman switch, auto-respawn, provider failover |
| Capacity > 60% | ~35% | Need more agents doing work, less idle time |
### What closes the gap
1. **Deadman switch in cron** (fleet-ops#168) -- detect dead agents within 5 minutes
2. **Auto-respawn** (fleet-ops#173) -- restart dead tmux panes automatically
3. **Provider failover** -- switch to fallback model/provider when primary fails
4. **Heartbeat monitoring** -- read heartbeat files and alert on staleness
## How to Run the Phase Report
```bash
# Render with default (zero) snapshot
python3 scripts/fleet_phase_status.py
# Render with real snapshot
python3 scripts/fleet_phase_status.py --snapshot configs/phase-1-snapshot.json
# Output as JSON
python3 scripts/fleet_phase_status.py --snapshot configs/phase-1-snapshot.json --json
# Write to file
python3 scripts/fleet_phase_status.py --snapshot configs/phase-1-snapshot.json --output docs/FLEET_PHASE_1_SURVIVAL.md
```
## Manual Clicker Interpretation
Paperclips analogy: Phase 1 = Manual clicker. You ARE the automation.
Every restart, every SSH, every check is a manual click.
## Manual Clicks Still Required
- Restart agents and services by hand when a node goes dark.
- SSH into machines to verify health, disk, and memory.
- Check Gitea, relay, and world services manually before and after changes.
- Act as the scheduler when automation is missing or only partially wired.
## Repo Signals Already Present
- `scripts/fleet_health_probe.sh` — Automated health probe exists and can supply the uptime baseline for the next phase.
- `scripts/fleet_milestones.py` — Milestone tracker exists, so survival achievements can be narrated and logged.
- `scripts/auto_restart_agent.sh` — Auto-restart tooling already exists as phase-2 groundwork.
- `scripts/backup_pipeline.sh` — Backup pipeline scaffold exists for post-survival automation work.
- `infrastructure/timmy-bridge/reports/generate_report.py` — Bridge reporting exists and can summarize heartbeat-driven uptime.
The goal of Phase 1 is not to automate. It's to **name what needs automating**. Every manual click documented here is a Phase 2 ticket.
## Notes
- The fleet is alive, but the human is still the control loop.
- Phase 1 is about naming reality plainly so later automation has a baseline to beat.
- Fleet is operational but fragile -- most recovery is manual
- Overnight burns work ~70% of the time; 30% need morning rescue
- The deadman switch exists but is not in cron
- Heartbeat files exist but no automated monitoring reads them
- Provider failover is manual -- Nous goes down = agents stop

View File

@@ -454,23 +454,112 @@ class TimmyAI:
class NPCAI:
"""AI for non-player characters. They make choices based on goals."""
GOAL_ROOM_TARGETS = {
"Marcus": {
"sit": "Garden",
"speak_truth": "Threshold",
"remember": "Bridge",
},
"Bezalel": {
"forge": "Forge",
"tend_fire": "Forge",
"create_key": "Forge",
},
"Allegro": {
"oversee": "Threshold",
"keep_time": "Tower",
"check_tunnel": "Bridge",
},
"Ezra": {
"study": "Tower",
"read_whiteboard": "Tower",
"find_pattern": "Tower",
},
"Gemini": {
"observe": "Threshold",
"tend_garden": "Garden",
"listen": "Garden",
},
"Claude": {
"inspect": "Threshold",
"organize": "Tower",
"enforce_order": "Bridge",
},
"ClawCode": {
"forge": "Forge",
"test_edge": "Bridge",
"build_weapon": "Forge",
},
"Kimi": {
"contemplate": "Garden",
"read": "Tower",
"remember": "Bridge",
},
}
GOAL_CYCLES = {
"Marcus": ("sit", "speak_truth", "remember"),
"Allegro": ("oversee", "keep_time", "check_tunnel"),
"Claude": ("inspect", "organize", "enforce_order"),
"ClawCode": ("test_edge", "forge", "build_weapon"),
"Kimi": ("contemplate", "read", "remember"),
}
def __init__(self, world):
self.world = world
def _available_targets(self, available, prefix):
return [a.split(":", 1)[1] for a in available if a.startswith(f"{prefix}:")]
def _target_room_for(self, char_name, goal):
return self.GOAL_ROOM_TARGETS.get(char_name, {}).get(goal)
def _next_direction_toward(self, current_room, target_room):
if current_room == target_room:
return None
frontier = [(current_room, [])]
seen = {current_room}
while frontier:
room, path = frontier.pop(0)
if room == target_room:
return path[0] if path else None
for direction, dest in self.world.rooms[room].get("connections", {}).items():
if dest not in seen:
seen.add(dest)
frontier.append((dest, path + [direction]))
return None
def _move_toward_goal(self, room, target_room):
direction = self._next_direction_toward(room, target_room)
return f"move:{direction}" if direction else None
def _advance_goal_cycle(self, char_name, char):
cycle = self.GOAL_CYCLES.get(char_name)
if not cycle or self.world.tick <= 0:
return
goal = char.get("active_goal")
if goal not in cycle:
return
target_room = self._target_room_for(char_name, goal)
if char.get("room") != target_room:
return
if self.world.tick % 12 != 0:
return
index = cycle.index(goal)
char["active_goal"] = cycle[(index + 1) % len(cycle)]
def make_choice(self, char_name):
"""Make a choice for this NPC this tick."""
char = self.world.characters[char_name]
self._advance_goal_cycle(char_name, char)
room = char["room"]
available = ActionSystem.get_available_actions(char_name, self.world)
# If low energy, rest
if char["energy"] <= 1:
return "rest"
# Goal-driven behavior
goal = char["active_goal"]
if char_name == "Marcus":
return self._marcus_choice(char, room, available)
elif char_name == "Bezalel":
@@ -487,66 +576,96 @@ class NPCAI:
return self._clawcode_choice(char, room, available)
elif char_name == "Kimi":
return self._kimi_choice(char, room, available)
return "rest"
def _marcus_choice(self, char, room, available):
goal = char.get("active_goal", "sit")
target_room = self._target_room_for("Marcus", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
others = self._available_targets(available, "speak")
if goal == "speak_truth" and others:
return f"speak:{random.choice(others)}"
if goal == "remember" and room == "Bridge":
return random.choice(["examine", "rest"])
if room == "Garden" and random.random() < 0.7:
return "rest"
if room != "Garden":
return "move:west"
# Speak to someone if possible
others = [a.split(":")[1] for a in available if a.startswith("speak:")]
if others and random.random() < 0.4:
return f"speak:{random.choice(others)}"
return "rest"
def _bezalel_choice(self, char, room, available):
target_room = self._target_room_for("Bezalel", char.get("active_goal", "forge"))
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
if room == "Forge" and self.world.rooms["Forge"]["fire"] == "glowing":
return random.choice(["forge", "rest"] if char["energy"] > 2 else ["rest"])
if room != "Forge":
return "move:west"
if random.random() < 0.3:
return "tend_fire"
return "forge"
def _kimi_choice(self, char, room, available):
others = [a.split(":")[1] for a in available if a.startswith("speak:")]
goal = char.get("active_goal", "contemplate")
target_room = self._target_room_for("Kimi", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
others = self._available_targets(available, "speak")
if goal == "read" and room == "Tower":
return "study" if char["energy"] > 2 else "rest"
if room == "Garden" and others and random.random() < 0.3:
return f"speak:{random.choice(others)}"
if room == "Tower":
return "study" if char["energy"] > 2 else "rest"
return "move:east" # Head back toward Garden
if room == "Bridge":
return random.choice(["examine", "rest"])
return "rest"
def _gemini_choice(self, char, room, available):
others = [a.split(":")[1] for a in available if a.startswith("listen:")]
if room == "Garden" and others and random.random() < 0.4:
return f"listen:{random.choice(others)}"
return random.choice(["plant", "rest"] if room == "Garden" else ["move:west"])
goal = char.get("active_goal", "observe")
target_room = self._target_room_for("Gemini", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
listeners = self._available_targets(available, "listen")
if room == "Garden" and listeners and random.random() < 0.4:
return f"listen:{random.choice(listeners)}"
return random.choice(["plant", "rest"] if room == "Garden" else ["examine", "rest"])
def _ezra_choice(self, char, room, available):
goal = char.get("active_goal", "study")
target_room = self._target_room_for("Ezra", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
if room == "Tower" and char["energy"] > 2:
return random.choice(["study", "write_rule", "help:Timmy"])
if room != "Tower":
return "move:south"
return "rest"
def _claude_choice(self, char, room, available):
others = [a.split(":")[1] for a in available if a.startswith("confront:")]
goal = char.get("active_goal", "inspect")
target_room = self._target_room_for("Claude", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
others = self._available_targets(available, "confront")
if others and random.random() < 0.2:
return f"confront:{random.choice(others)}"
return random.choice(["examine", "rest"])
def _clawcode_choice(self, char, room, available):
goal = char.get("active_goal", "test_edge")
target_room = self._target_room_for("ClawCode", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
if room == "Forge" and char["energy"] > 2:
return "forge"
return random.choice(["move:east", "forge", "rest"])
return random.choice(["examine", "rest"])
def _allegro_choice(self, char, room, available):
others = [a.split(":")[1] for a in available if a.startswith("speak:")]
goal = char.get("active_goal", "oversee")
target_room = self._target_room_for("Allegro", goal)
if room != target_room:
return self._move_toward_goal(room, target_room) or "rest"
others = self._available_targets(available, "speak")
if others and random.random() < 0.3:
return f"speak:{random.choice(others)}"
return random.choice(["move:north", "move:south", "examine"])
return random.choice(["examine", "rest"])
class DialogueSystem:

View File

@@ -10,7 +10,6 @@ BACKUP_LOG_DIR="${BACKUP_LOG_DIR:-${BACKUP_ROOT}/logs}"
BACKUP_RETENTION_DAYS="${BACKUP_RETENTION_DAYS:-14}"
BACKUP_S3_URI="${BACKUP_S3_URI:-}"
BACKUP_NAS_TARGET="${BACKUP_NAS_TARGET:-}"
OFFSITE_TARGET="${OFFSITE_TARGET:-}"
AWS_ENDPOINT_URL="${AWS_ENDPOINT_URL:-}"
BACKUP_NAME="hermes-backup-${DATESTAMP}"
LOCAL_BACKUP_DIR="${BACKUP_ROOT}/${DATESTAMP}"
@@ -32,16 +31,6 @@ fail() {
exit 1
}
send_telegram() {
local message="$1"
if [[ -n "${TELEGRAM_BOT_TOKEN:-}" && -n "${TELEGRAM_CHAT_ID:-}" ]]; then
curl -s -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
-d "chat_id=${TELEGRAM_CHAT_ID}" \
-d "text=${message}" \
-d "parse_mode=HTML" > /dev/null || true
fi
}
cleanup() {
rm -f "$PLAINTEXT_ARCHIVE"
rm -rf "$STAGE_DIR"
@@ -129,17 +118,6 @@ upload_to_nas() {
log "Uploaded backup to NAS target: $target_dir"
}
upload_to_offsite() {
local archive_path="$1"
local manifest_path="$2"
local target_root="$3"
local target_dir="${target_root%/}/${DATESTAMP}"
mkdir -p "$target_dir"
rsync -az --delete "$archive_path" "$manifest_path" "$target_dir/"
log "Uploaded backup to offsite target: $target_dir"
}
upload_to_s3() {
local archive_path="$1"
local manifest_path="$2"
@@ -183,16 +161,10 @@ if [[ -n "$BACKUP_NAS_TARGET" ]]; then
upload_to_nas "$ENCRYPTED_ARCHIVE" "$MANIFEST_PATH" "$BACKUP_NAS_TARGET"
fi
if [[ -n "$OFFSITE_TARGET" ]]; then
upload_to_offsite "$ENCRYPTED_ARCHIVE" "$MANIFEST_PATH" "$OFFSITE_TARGET"
fi
if [[ -n "$BACKUP_S3_URI" ]]; then
upload_to_s3 "$ENCRYPTED_ARCHIVE" "$MANIFEST_PATH"
fi
find "$BACKUP_ROOT" -mindepth 1 -maxdepth 1 -type d -name '20*' -mtime "+${BACKUP_RETENTION_DAYS}" -exec rm -rf {} + 2>/dev/null || true
find "$BACKUP_ROOT" -mindepth 1 -maxdepth 1 -type d -mtime +7 -exec rm -rf {} + 2>/dev/null || true
log "Retention applied (${BACKUP_RETENTION_DAYS} days)"
log "Backup pipeline completed successfully"
send_telegram "✅ Daily backup completed: ${DATESTAMP}"

View File

@@ -0,0 +1,54 @@
from importlib.util import module_from_spec, spec_from_file_location
from pathlib import Path
ROOT = Path(__file__).resolve().parent.parent
GAME_PATH = ROOT / "evennia" / "timmy_world" / "world" / "game.py"
def load_game_module():
spec = spec_from_file_location("tower_world_game", GAME_PATH)
module = module_from_spec(spec)
assert spec.loader is not None
spec.loader.exec_module(module)
module.random.seed(0)
return module
def _visitor_sets_after_ticks(module, ticks=100):
engine = module.GameEngine()
engine.start_new_game()
visitors = {room: set() for room in engine.world.rooms}
for _ in range(ticks):
engine.run_tick("rest")
for name, char in engine.world.characters.items():
if name == "Timmy":
continue
visitors[char["room"]].add(name)
return visitors
class TestTowerGameNpcPurpose:
def test_goal_driven_room_targets(self):
module = load_game_module()
world = module.World()
npc_ai = module.NPCAI(world)
world.characters["Marcus"]["room"] = "Threshold"
world.characters["Marcus"]["active_goal"] = "sit"
assert npc_ai.make_choice("Marcus") == "move:east"
world.characters["Ezra"]["room"] = "Threshold"
world.characters["Ezra"]["active_goal"] = "study"
assert npc_ai.make_choice("Ezra") == "move:north"
world.characters["Claude"]["room"] = "Threshold"
world.characters["Claude"]["active_goal"] = "enforce_order"
assert npc_ai.make_choice("Claude") == "move:south"
def test_every_room_gets_multiple_npc_visitors_over_100_ticks(self):
module = load_game_module()
visitors = _visitor_sets_after_ticks(module, ticks=100)
assert all(len(names) >= 2 for names in visitors.values()), visitors
assert len(visitors["Bridge"]) >= 3, visitors["Bridge"]