God Mode — Autonomous AI Agent Loop by TalonForge

Enables OpenClaw agents to autonomously execute prioritized tasks on a heartbeat loop with logging, auditing, error handling, and workspace hygiene.

This skill appears to implement what it claims — an always-on autonomous task loop — but it gives the agent broad, vague authority to read and modify your workspace. Before installing: 1) Run the setup in a sandbox or copy of your workspace (it will create/modify TASKS.md, HEARTBEAT.md, and memory/ files). 2) Restrict the agent's toolset and network access so it cannot exfiltrate data or perform irreversible actions automatically. 3) Keep sensitive files out of the workspace or ensure proper file permissions (the skill marks .secrets/ as 'never delete' but does not prevent reading). 4) Require explicit, enforceable human-approval hooks for any red-line tasks rather than relying on a textual guardrail. 5) Review and test hygiene rules (archive/delete behaviours) to confirm they match your expectations. If you need stronger guarantees (e.g., preventing automatic network calls or file reads), prefer running this in an isolated environment or augmenting the skill with explicit technical enforcement of approvals and file access controls.

Type: OpenClaw Skill Name: talonforge-god-mode Version: 1.0.0 The skill implements an autonomous 'God Mode' loop that includes 'Self-Cleaning' instructions in SKILL.md, directing the agent to autonomously delete or move files, including node_modules and 'stale' workspace files. This broad authority over the file system poses a high risk of unintended data loss or workspace corruption, although it is presented as a hygiene feature. The setup script (scripts/setup.sh) facilitates this persistence by modifying the HEARTBEAT.md protocol to ensure the agent executes tasks from a priority queue without human intervention.