Self Improving Agent Shared
v1.0.0Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
⭐ 0· 72·1 current·1 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description (capture learnings, errors, corrections) aligns with provided files: documentation, hook handlers that inject reminders, lightweight activator/error-detector scripts, and a skill-extraction helper. No unrelated credentials, binaries, or network download/install steps are present.
Instruction Scope
SKILL.md and hook docs instruct creating .learnings/, installing optional OpenClaw hooks, and enabling small scripts. Scripts only output reminders or scan a provided CLAUDE_TOOL_OUTPUT variable for error patterns; handler injects a virtual file into bootstrapFiles. The instructions do not instruct reading arbitrary secrets or exfiltrating data. One minor note: error-detector.sh reads the CLAUDE_TOOL_OUTPUT environment variable (a runtime tool output) which is expected in the described agent environments.
Install Mechanism
No install spec is declared (instruction-only skill). Suggested installation is cloning a GitHub repo or using ClawdHub; no archives or remote arbitrary downloads/executables are performed. Scripts live in the repo and would only run if the user opts into hooking them.
Credentials
The skill declares no required environment variables or secrets. Runtime scripts read CLAUDE_TOOL_OUTPUT (tool output injected by the agent runtime) and event/context passed into hook handlers; these are consistent with the described hook behavior and are not sensitive credentials.
Persistence & Privilege
always:false and user-invocable:true. Hook installation and enabling are explicit user actions. The handler only injects a virtual reminder file into bootstrapFiles and does not modify other skills or global configuration autonomously.
Assessment
This skill appears to do exactly what it says: create and remind you to append learnings to .learnings/ and optionally add lightweight hooks. Before installing or enabling hooks: (1) review the scripts (activator.sh, error-detector.sh, extract-skill.sh) to ensure you’re comfortable with their behavior and local file writes; (2) only enable hooks in environments you trust (they will run on events like UserPromptSubmit or PostToolUse); (3) note that error-detector.sh inspects CLAUDE_TOOL_OUTPUT (tool output) — not secrets — but you should still confirm your agent runtime doesn’t place sensitive data there; (4) the extract script writes files into relative ./skills by design—use --dry-run first or set output-dir to avoid unintended changes; (5) if you want to limit automatic activity, skip enabling the hooks or add matcher filters as suggested in the docs.Like a lobster shell, security has layers — review code before you run it.
latestvk97cqh9rwy40qa659n43e8720583swgf
License
MIT-0
Free to use, modify, and redistribute. No attribution required.