Ralph Wiggum Loop

Iteracyjnie doskonali kod lub tekst AI, wykrywając i naprawiając błędy, optymalizacje, bezpieczeństwo i styl w maksymalnie trzech krokach.

This skill appears to implement the advertised iterative improvement loop, but review and test before using on sensitive code. Specific recommendations: - Ensure LM Studio runs locally and do NOT set LMSTUDIO_URL to a remote host unless you intend to send your code there (the scripts will transmit the full code to whatever LMSTUDIO_URL is used). - Install required tools first: Python 3.9+, the 'requests' package (pip install requests), curl, and jq (the shell scripts use jq but SKILL.md didn't mention it). - Be aware of a small implementation mismatch: ralph-loop.sh passes flags (-u, -s) that generator.py's CLI doesn't define; run the Python modules directly or inspect/fix the shell script before relying on it. - Review the scripts' behavior (especially fix_code which posts code and issues to the LM endpoint) on non-sensitive examples to confirm behavior and outputs. - If you plan to run this in a production environment, run it in an isolated environment and audit network traffic to confirm LM Studio is local.

Type: OpenClaw Skill Name: ralph-wiggum-loop Version: 1.0.0 The skill bundle implements a legitimate iterative code improvement loop (Generator-Critic-Fixer-Verifier) designed to work with a local LM Studio instance. The scripts (ralph-loop.sh, critic.py, and generator.py) perform standard API interactions and include basic security linting to detect SQL injection and hardcoded secrets in the generated code. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found; the preference for 'uncensored' models is contextually aligned with providing a 'ruthless' code critic.