1215656 Self Improving Agent@3.0.6
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· 69·1 current·1 all-time
by@1215656
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 contents: logging learnings to .learnings/*, promoting entries to workspace docs, and optional hook-based reminders. The included hooks, activator, error-detector, and extract-skill scripts are coherent with a skill whose job is to capture and scaffold learnings.
Instruction Scope
SKILL.md asks operators to copy hooks into OpenClaw/Claude settings and to run small helper scripts. Those instructions intentionally inject reminder files into agent bootstrap context and run lightweight local scripts (activator and error-detector). This is expected for the described purpose, but enabling hooks means the skill's content becomes part of the agent's injected context and the hook scripts will run automatically when configured.
Install Mechanism
No install spec in registry; the skill is instruction-only in the registry bundle and includes local scripts and hook handlers. The manual install instructions reference a GitHub repo (reasonable). There are no downloads from arbitrary URLs or extract operations in the bundle itself.
Credentials
The skill requests no credentials or env vars. Scripts reference CLAUDE_TOOL_OUTPUT (an agent-provided environment variable) and otherwise operate on local files. No unrelated secrets or external service tokens are requested.
Persistence & Privilege
always:false (normal). The skill can be enabled as a hook which adds virtual bootstrap files and runs scripts with the same permissions as the agent — this is expected but means the skill can influence agent context across sessions if you opt in. extract-skill.sh writes files to a relative ./skills directory (has checks against absolute paths and '..').
Assessment
This skill appears to do what it says: inject lightweight reminders and provide small helper scripts for logging and extracting learnings. Before enabling it: 1) Inspect the scripts (activator.sh, error-detector.sh, extract-skill.sh) yourself — they run locally with your agent's permissions. 2) If you enable hooks, prefer the minimal activator-only setup first (less automatic execution). 3) Be cautious enabling PostToolUse hooks that execute scripts on every tool call (they could run often); use matcher filters if available. 4) Use extract-skill.sh with --dry-run initially to review what it would create. 5) If you share a machine or workspace, remember these files become part of the agent's injected workspace and can affect future sessions. Overall the bundle is coherent and doesn't request disproportionate access, but only enable its automatic hooks after you review the scripts and are comfortable with local script execution in your environment.Like a lobster shell, security has layers — review code before you run it.
latestvk9703btpvwygpp32sdxydhkn2h83kmvh
License
MIT-0
Free to use, modify, and redistribute. No attribution required.