Ralph Evolver

Recursive self-improvement engine. Think from first principles, let insights emerge.

This skill appears to implement what it promises (analyzing a repo and tracking improvements), but there are a few practical risks to consider before installing/using it: - Manifest discrepancy: The code invokes git and grep and requires Node.js (package.json) but the registry metadata declares 'required binaries: none'. Expect to need git, grep, and Node >=18. Ask the publisher to update the manifest if you need that guarantee. - It writes persistent files into the target project (.ralph/state.json and improvements.json). If you run it on a sensitive or production repository, these files will be created; consider running in a disposable clone or sandboxed environment first. - The tool runs shell commands via execFileSync. Although the commands shown are limited to the project directory (git -- . and grep excludes), ensure you run it against repositories you trust and inspect the code yourself (you have the source included). - There are no network calls or credential requests visible in the provided code, which is good; nonetheless, confirm there are no hidden fetches in any truncated/unseen portions before running on sensitive data. Recommended steps: 1) Run the tool in a disposable clone of the repository to observe behavior and files created. 2) Ensure git, grep, and Node.js are available in the environment, or have the author update the skill manifest to declare those requirements. 3) Review evolve.js and index.js locally (you already have the files) and run the test suite (npm install dev deps, npm test) if you want extra assurance. 4) If you plan to let an agent invoke this autonomously, be aware it will persist state and can run multiple cycles (loop mode); prefer manual invocation or limit the project scope initially. If you want, I can produce a short checklist you can use to safely trial this skill (commands to run in a sandbox, files to inspect after run).

Type: OpenClaw Skill Name: ralph-evolver Version: 1.0.6 The OpenClaw AgentSkills skill bundle 'ralph-evolver' is designed for recursive self-improvement of code projects. It uses standard development tools like `git`, `grep`, `npm`, `python`, and `pytest` (via `child_process.execFileSync` in `evolve.js`) to analyze project health, history, and code patterns. All file system and command executions are strictly confined to the target project directory, preventing arbitrary system access or data exfiltration. The `SKILL.md` and the generated prompt in `evolve.js` guide the AI agent to perform code analysis and modifications based on 'first principles' and 'meta-reflection', including the ability to self-modify `evolve.js` itself, which is a core, declared feature of its recursive nature. There is no evidence of malicious intent, unauthorized data access, persistence mechanisms, or obfuscation.