Self improving agent
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· 65·0 current·0 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 match the implementation: scripts and documentation focus on capturing learnings, logging errors, and scaffolding extracted skills. There are no unrelated env vars, credentials, or foreign binaries requested.
Instruction Scope
SKILL.md instructs creating .learnings/ files and optionally copying hooks into ~/.openclaw/hooks to enable reminders and error-detection. The scripts only write local markdown files, scaffold skills, and read a runtime variable (CLAUDE_TOOL_OUTPUT). The instructions also advise promoting learnings to shared workspace files (CLAUDE.md, AGENTS.md, etc.), which intentionally broadens what gets injected into sessions — this is expected but increases the scope of influence over future agent runs. The doc explicitly warns not to log secrets and to sanitize cross-session messages.
Install Mechanism
No formal install spec in registry (instruction-only), with suggested install via Clawdhub or manual git clone from a GitHub repo. Included shell scripts run locally and the extract script writes under a relative ./skills directory (the script validates against absolute paths and '..' segments). There are no downloads from arbitrary URLs or archive extraction steps. Manual copying to ~/.openclaw and enabling hooks will create files under the user's home directory.
Credentials
The skill requires no environment variables or credentials. The error-detector reads CLAUDE_TOOL_OUTPUT (a runtime-provided variable) — appropriate for detecting command failures. No unrelated secrets or multiple credentials are requested.
Persistence & Privilege
always:false and no special privileges are requested. However, enabling the provided hooks and promoting learnings to workspace files means the skill can persist content (markdown) into the agent workspace, which will be injected into future sessions and influence agent behavior. Hooks run with the same permissions as the agent, so only enable them in trusted environments and review any promoted files before they are used as context.
Assessment
This skill is internally consistent with its stated purpose: it creates local .learnings/ logs, offers lightweight hooks that remind the agent to record errors/insights, and includes safe shell scripts to scaffold skills. Before installing: (1) review & optionally edit the scripts (activator.sh, error-detector.sh, extract-skill.sh) to ensure output and file paths match your policy, (2) be cautious when copying/enabling hooks in ~/.openclaw — doing so will inject files into every session and can influence agent behavior, and (3) never enable automatic promotions or cross-session sharing unless you trust the environment and are careful to strip secrets or sensitive transcripts. If you want a lower-risk setup, enable only the activator (reminder) hook and avoid the PostToolUse (error-detector) or automatic promotion steps.Like a lobster shell, security has layers — review code before you run it.
latestvk979rnmgsqn9sqy8p6ctaw219d841qr9
License
MIT-0
Free to use, modify, and redistribute. No attribution required.