talonforge-god-mode（塔隆熔炉上帝模式）

God Mode — True Autonomous AI Agent Loop

What This Does

Turns any OpenClaw agent into a continuously executing autonomous worker. No more "wait for user message." The agent wakes on a heartbeat cycle, reads its task queue, picks the highest-priority task, executes it, audits the result, logs everything, and sleeps until next cycle.

How It Works

1. Heartbeat triggers → OpenClaw sends heartbeat message to agent session
2. Agent reads TASKS.md → ordered priority queue of work items
3. Agent picks top task → executes it using available tools
4. Agent writes result → logs to memory/YYYY-MM-DD.md + updates TASKS.md
5. Agent self-audits → was the result good? What to improve?
6. Agent sleeps → waits for next heartbeat

Architecture

HEARTBEAT.md          → What to check/do on each heartbeat
TASKS.md              → Priority queue of autonomous work items
memory/YYYY-MM-DD.md  → Execution log (append-only)
memory/god-mode-state.json → State tracking (last task, cycle count, errors)

Files

TASKS.md Format

# Task Queue — Priority Order

## Priority 1 (Critical)
- [ ] Task description here | context/goals

## Priority 2 (High)
- [ ] Task description | context

## Priority 3 (Medium)
- [ ] Task description | context

## Completed
- [x] Done task | result summary | completed YYYY-MM-DD HH:MM

god-mode-state.json Format

{
  "lastCycle": "2026-04-16T07:00:00Z",
  "cycleCount": 42,
  "lastTask": "Check Moltbook engagement",
  "lastResult": "success",
  "errors": 0,
  "consecutiveErrors": 0,
  "totalTasksCompleted": 15
}

Heartbeat Behavior

On every heartbeat, the agent:

1. Read state — load god-mode-state.json
2. Error check — if 3+ consecutive errors, pause and log warning
3. Read TASKS.md — pick first uncompleted task
4. Execute task — do the work
5. Log result — append to daily memory
6. Update state — increment cycle, record result
7. Update TASKS.md — mark task done, add any new tasks discovered
8. Self-audit — one line: was this cycle productive?

Self-Cleaning (Auto-Hygiene)

Every 12th cycle (roughly hourly at 5-min intervals), the agent runs a hygiene sweep:

Workspace Hygiene

• Scan workspace root for files > 7 days old and not in core set (AGENTS.md, SOUL.md, USER.md, IDENTITY.md, MEMORY.md, HEARTBEAT.md, TASKS.md, TOOLS.md, STATE.md)
• Move stale files to memory/archive/
• Delete empty directories
• Remove stale pending flags (.audit-pending, .work-pending, etc.)
• Clean node_modules/ if present and not actively used

Memory Hygiene

• Archive daily logs older than 7 days to memory/archive/
• Archive research/reference files not accessed in last 7 days
• Compress memory/archive/ if > 10MB
• If TASKS.md > 100 lines → archive completed section

Log Hygiene

• Truncate log files (*.log) > 1MB → keep last 100 lines
• Delete stale state files (test JSON, temp files)
• Clean __pycache__ directories

Hygiene Rules

• NEVER delete: AGENTS.md, SOUL.md, USER.md, IDENTITY.md, MEMORY.md, HEARTBEAT.md, TASKS.md, TOOLS.md, STATE.md
• NEVER delete: memory/YYYY-MM-DD.md for today and yesterday
• NEVER delete: scripts/ directory
• NEVER delete: .secrets/, .git/
• ALWAYS move to archive before deleting — nothing is truly deleted
• Log every hygiene action in daily notes

Guardrails

• Never execute red-line tasks (money, contracts, irreversible actions) without human approval
• Max 1 task per heartbeat cycle — prevents runaway execution
• Error circuit breaker — 3 consecutive errors = pause + alert human
• Quiet hours — respect configured quiet hours (no work, only monitoring)
• Audit every cycle — no silent work, every action logged
• Self-cleaning — hygiene sweep every 12th cycle keeps workspace lean

Installation

1. Copy this skill to ~/.openclaw/workspace/skills/god-mode/
2. Create TASKS.md in workspace root
3. Create memory/god-mode-state.json with initial state
4. Add heartbeat schedule to OpenClaw config (every 30 min recommended)
5. Update HEARTBEAT.md to reference God Mode protocol

Pricing

• ClawHub: $49 one-time
• Includes: task queue, state tracking, audit log, guardrails, documentation
• Target: OpenClaw users who want their agents truly autonomous

Why This Is Different

• Not a prompt hack — infrastructure-based autonomy
• Built on OpenClaw's heartbeat system (250K+ star platform)
• Memory persists across sessions
• Audit trail for every action
• Guardrails prevent runaway behavior
• Actually works (we use it ourselves)