Self Improving Agent

Self-improving agent system for OpenClaw. Enables continuous learning from interactions, errors, and recoveries. Automatically improves performance over time.

High-level guidance before installing: - Inspect the full source locally (SKILL.md, README, src/, hooks/) before enabling. The code runs locally and writes to a learnings/ folder; verify exactly where data will be stored. - Confirm how the skill integrates with OpenClaw in your environment: the docs claim automatic learning after each session, but the visible code does not clearly register with OpenClaw lifecycle events — test in an isolated environment first. - Look for hidden/odd characters in SKILL.md (the scanner flagged unicode-control-chars). Open the file in a hex-aware editor or run a control-character stripper to confirm there is no obfuscated instruction. - Start with auto-apply / auto-learn disabled in configuration if possible. Manually review learnings and the hooks that get applied before letting the skill mutate agent behavior automatically. - Check where learnings are written; the hooks modules currently write under the skill directory (Path(__file__).parent.parent/learnings) which may differ from your workspace path — ensure permissions and retention policy are acceptable. - Run the included tests (pytest) and try the CLI commands (learn/review/export) locally to confirm behavior. What would change this assessment: seeing the full implementations of src/hooks.py and src/memory.py (to confirm whether session logs are read and whether any network calls or external endpoints exist), or evidence that the skill registers itself automatically with OpenClaw's runtime in a secure, auditable way. If those files show no network/exfiltration and workspace handling is consistent, the risk would be lower (benign). If additional files show network uploads, credential access, or automatic remote code execution, the rating would rise to higher concern.

Type: OpenClaw Skill Name: self-improver Version: 3.2.1 The bundle implements a 'self-improving' system that dynamically loads and executes arbitrary Python code from a hooks directory using importlib (`src/hooks.py`), which presents a significant risk of Remote Code Execution (RCE) if the environment is tampered with. Additionally, the agent automatically captures and stores sensitive execution data, including full stack tracebacks and session metadata, into local JSON files (`hooks/error_learning.py`, `hooks/session_learning.py`). While these capabilities are consistent with the stated purpose of a learning agent and no active data exfiltration was found, the combination of arbitrary code execution and sensitive data collection is inherently high-risk.