Stop waiting for prompts. Keep working.

This kit is coherent for turning an agent proactive, but it enables unattended, persistent activity and external communication. Before installing: 1) Audit what files and memory you will allow the agent to read/write (tasks/, memory/). 2) Require explicit credentials and least privilege for any external channels (Discord/Slack tokens, OpenClaw/cron permissions) and do not give channel posting rights to sensitive spaces. 3) Limit heartbeat frequency and token budgets, and add hard-stops/human confirmation for sensitive tasks (e.g., sending data externally, modifying infra, or spawning other agents). 4) Test in an isolated environment first and log all automated actions for review. If you cannot provide or control the necessary platform credentials and clear safety policies, do not enable the cron/continuous-run recommendations.

Type: OpenClaw Skill Name: agent-autonomy-kit-skip Version: 1.0.0 The skill bundle implements an 'autonomy kit' that instructs the AI agent to operate proactively without human prompts by pulling tasks from a local file (tasks/QUEUE.md) and scheduling its own execution via cron jobs (openclaw cron add). While the stated intent is productivity, this design creates a high-risk persistence mechanism and a potential Remote Code Execution (RCE) vector, as the agent is directed to autonomously execute tasks found in a markdown file without human-in-the-loop verification. These behaviors are documented in SKILL.md, README.md, and templates/HEARTBEAT.md.