self-evalutaed-agent

Automatically detects errors, researches solutions, executes improvements, and measures impact while remembering effective procedures for continuous self-imp...

This package appears coherent for its claimed purpose, but review and test before deploying to production: 1) Run it in an isolated test workspace (set OPENCLAW_WORKSPACE to a non-root path) so the scripts can create memory/ and backlog/ without touching your real data. 2) Fix the import bug in topic_selector.py (it imports SELF_IMPROVE_LOG / SELF_IMPROVEMENT_LOG which is not defined in config.py) — that will cause module import failures; run the scripts manually to confirm behavior. 3) Inspect files the skill will read/write (errors.jsonl, .circuit_breakers.json, memory/procedural.jsonl, backlog/*.md) and ensure you’re comfortable with automatic writes to those locations. 4) Be cautious with any stored procedures: procedural_memory records arbitrary command strings (it does not execute them in the provided code), so review prior to re-using them. 5) If you plan to enable cron-triggering, confirm the command paths match where you placed the scripts. If you want higher assurance, request a maintainer or author-signed release or run the code through your own static/dynamic checks.

Type: OpenClaw Skill Name: self-evalutaed-agent Version: 1.0.0 The skill bundle implements a self-improvement loop for an OpenClaw agent by monitoring local error logs and updating a task backlog. The Python scripts (auto_trigger.py, self_improvement_cycle.py, topic_selector.py) perform standard log analysis, JSON processing, and Markdown generation within the defined workspace (/root/.openclaw/workspace). There is no evidence of data exfiltration, unauthorized network access, or malicious prompt injection; the system's use of subprocesses and cron integration is consistent with its stated purpose of autonomous maintenance.