Automatically evaluates and approves agent outputs based on clarity, conciseness, actionability, and structure using a rule-based system.

This skill appears to be a simple, local Node.js script that scores text by a few heuristics — generally safe and coherent with its description. Before installing: (1) ensure Node.js is available where the agent runs (the manifest omits this), (2) review any agent integration points (the README mentions calling it in HEARTBEAT or before sending responses), and (3) be aware future versions may add history logging or LLM API calls (README TODOs); if you don't trust the author, run the script in an isolated environment first.

Type: OpenClaw Skill Name: self-review Version: 1.0.0 The skill bundle is designed for self-review of agent output quality using a simple rule-based engine. The `index.js` script reads input from stdin, performs basic regex checks on its length, presence of action words, and structure, then exits with a status code indicating approval or a need for improvement. There is no evidence of data exfiltration, malicious execution, persistence mechanisms, or prompt injection attempts in any of the files. The documentation (`SKILL.md`, `README.md`) and skill definition (`skill.json`) clearly describe its benign purpose and functionality, with future 'TODO' items in `README.md` being plans, not active instructions or malicious code.