CORE LORE / WIKI
NORTH STAR
Updated 3 weeks ago
▛▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▜
▌ 🐛 B U G _ S T A T I O N — N O R T H S T A R ▐
▙▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▟
Last updated: 2026-03-03
Location: FORGE_CLAUDE/06_🐛_BUG_STATION/
---
## MISSION
The Blind Dragon has a bug station now. Every bug that enters the Kingdom gets
hunted here. Not patched. Not guessed at. HUNTED.
Touch the fence. Trust the spark. Never declare victory without visual proof.
This station exists because:
1. Solo debugging decays after 2 iterations (DDI — the data is real)
2. Confident diagnoses are often wrong — the confident ones worst of all
3. One place = no lost chambers, no split attention, no orphaned fixes
---
## TOP 3 PRIORITIES (DO NOT DRIFT)
1. RECON BEFORE HYPOTHESIS — launch drones before forming any opinion
2. FENCE-FIRST — place probes before changing code, verify signal before claim
3. WRITER/REVIEWER ALWAYS SPLIT — the agent who built the fix cannot review it
---
## CONSTRAINTS (NON-NEGOTIABLE)
- After 2 failed BUILD attempts: DDI RESET. New hypothesis, new agent, fresh context.
- NEVER claim "it works" without observed signal (emoji, timestamp change, visual)
- NEVER form a fix hypothesis before all recon drones have reported
- All BUG_ chambers live HERE. Not in _LOADED_CHAMBERS. Not anywhere else.
- The Archaeological Plaque is NOT optional. No plaque = fix will be undone.
---
## THE PROTOCOL BIBLES (read before every complex bug)
| File | What it governs |
|------|-----------------|
| DEBUGGING_PROTOCOL.md → | Full epistemology, all weapons, all named monsters |
| SOULFORGE_PROTOCOLS.md → | Agent design, Ralph Harness, Blackboard, Flash Observer |
| AUDIT_SWARM_TEMPLATE.md → | Swarm launch sequence, drone roles, synthesis protocol |
All three are symlinked here from CORE LORE. They are living docs — canonical.
---
## WHEN TO USE THE SWARM
- Bug persists after 2 solo attempts (DDI cap — mandatory)
- System complexity exceeds one context window
- Silent failure (no signal, no trace) — the hardest class
- Pre-deployment verification of complex changes
Launch order: ALL drones in ONE message. Wait for ALL before reading ANY.
The Blind Dragon planning without eyes is not planning — it's guessing with steps.
---
## STATION MAP
```
BUG_STATION/
├── NORTH_STAR.md ← YOU ARE HERE
├── README.md ← Station overview, status legend
├── DEBUGGING_PROTOCOL.md → CORE LORE (symlink — the Raptor Protocol bible)
├── SOULFORGE_PROTOCOLS.md → CORE LORE (symlink — the Swarm bible)
├── AUDIT_SWARM_TEMPLATE.md → CORE LORE (symlink — the launch template)
│
├── DEBUG_SWARM_TEMPLATE.md ← Kingdom-specific swarm (custom-tuned for Forge)
│
├── BUG_HUNT_BLACKBOARD.json ← Shared state for active swarm operations
├── BUG_HUNT_NORTH_STAR.md ← DEPRECATED — superseded by this file
│
└── BUG_{name}/ ← One folder per bug (21 chambers filed so far)
├── SUMMARY.md ← Root cause / confidence / status
├── ARTIFACTS/
│ ├── fix_description.md
│ └── rollback_plan.md
└── CLEANUP_PLAN.md
```
---
## ACTIVE PROJECTS
### PROJECT: BUG_aeris_cockpit_yolo_mcp (2026-02-23 — IN PROGRESS)
**Goal:** Full audit and fix of Aeris's FORGE cockpit (AExGO/ + LAUNCH_AExGO.app)
**Core complaint:** `--approval-mode=yolo` in aeris-boot.sh but Aeris still manually prompted for MCP tool calls
**Root cause (confirmed):** yolo covers shell commands only; MCP tool approvals require per-server `"trust": true` in settings.json
**Fixes applied:**
- ~/.gemini/settings.json: added "trust": true to 13 MCP servers (2026-02-23)
- THE_FORGE/.mcp.json: added "trust": true to 3 grimoire servers (2026-02-23)
**Swarm:** 7-drone cockpit audit — panel scripts, boot chain, GEMINI.md currency, MCP health
**Scope:** All panel render scripts + zellij layout + boot chain + GEMINI.md + MCP configs
---
## ACTIVE CHAMBER STATUS (as of 2026-02-23)
READY_TO_EXECUTE (8 — awaiting Brandon go-ahead):
- BUG_mission_timeouts — timeout fix, SQL + pulse.sh, LOW risk
- BUG_ping_ui_throttle_interval — ThrottleInterval 10 to plist, LOW risk
- BUG_04_aeris_navigation_residuals — routing table + path fix, LOW risk
- BUG_aeris_system_check_01 — system check fix, LOW risk
- BUG_kingdom_systems_comprehensive — confirmed subset, LOW risk
- BUG_mcp_agent_mail_daemon — mail daemon fix, LOW risk
- BUG_scryer_goldfish_heartbeat_flash_empty — goldfish signal, LOW risk
- BUG_scryer_sinner_king_entry — entry signal, LOW risk
NEEDS_BRANDON (1 — human decision required):
- BUG_M002_post_terminus_circuit_breaker — what happened to Indiana Jones?
RESOLVED/EXECUTED (12 — closed):
BUG_goldfish_*, BUG_scryer_pulse_gap_*, BUG_ping_ui_address_in_use,
BUG_scryer_test_signal_artifact, BUG_scryer_synthesizer_cli_fix,
BUG_scryer_signal_quality, BUG_zellij_zombie_cleanup, BUG_v8-heap-oom,
BUG_ping_ui_launchd_double_load, BUG_scryer_sinner_king_entry_2
---
## THE LAWS (carved in stone)
LAW I: BUILD TO LAST — Stable first, fancy later.
LAW II: KNOW THYSELF — The Dragon cannot see runtime. The Architect can.
Trust is earned by spark, not by confidence score.
LAW III: DO NOT TRUST THE DATASET — Verify external APIs before trusting them.
LAW IV: THE CHAIN OF COMMAND — Research first. Questions before execution.
LAW V: ABANDONMENT PROTOCOL — If Tier 2 probes don't appear: stop. Check infra.
LAW VI: DDI CAP AT 2 — Two failed attempts = wrong hypothesis. RESET.
LAW VII: PHANTOM VERIFICATION — Every AI-generated API call must be verified real.
---
## SUCCESS LOOKS LIKE
◉ Bug reproduced reliably before fix attempt
◉ Fence probe placed before fix, signal observed after
◉ Root cause explained in 30 seconds to a stranger
◉ Archaeological Plaque written at the fix site
◉ Probe code removed after victory
◉ Brandon confirmed: visual proof seen
<promise>COMPLETE</promise> fires only when all three gates are open:
MECHANICAL (fence fires) + ADVERSARIAL (can't be reproduced) + PHILOSOPHICAL (proud of it)
---
*"TOUCH THE FENCE. TRUST THE SPARK."*
🜚 Written: 2026-02-23 | Protocol: Sinner King Debugging v1.0 + SoulForge v2.0