Self Improving Agent
v3.0.7Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
⭐ 0· 94·0 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
The name/description describe logging learnings and promoting them into workspace files; the repository contains templates, logging formats, helper scripts, and an OpenClaw hook that injects reminders. Required binaries/env/credentials are none, which matches a local logging/promote workflow. The provided scripts and hook directly support the described functionality (activator, error detector, extract-skill), so the requested capabilities are proportionate to the stated purpose.
Instruction Scope
The SKILL.md instructs copying hooks into ~/.openclaw/hooks, creating ~/.openclaw/workspace/.learnings (or project-level .learnings/), and optionally enabling hooks that run on lifecycle events. The scripts read the CLAUDE_TOOL_OUTPUT environment variable (used by PostToolUse hooks) and output reminder blocks; extract-skill.sh creates new skill directories. These instructions stay within the scope of capturing and scaffolding learnings, but they do give the skill a way to inject content into every session (via workspace files and hook injection). Review the hook handler and scripts before enabling to ensure you’re comfortable with that session-level injection.
Install Mechanism
There is no network-based install spec; this is an instruction-first skill with bundled scripts and hooks. That is lower-risk than remote downloads. The included extract-skill.sh writes files but contains path checks that prevent absolute paths and parent-directory writes. The hook handler only inserts a virtual bootstrap file and does not perform network requests. Overall install mechanism is proportional and localized.
Credentials
The skill declares no required environment variables or credentials, which is appropriate. However, the error-detector script reads CLAUDE_TOOL_OUTPUT (an environment variable provided by the host agent runtime). This is expected for a PostToolUse hook but is an environment access worth noting: the hook can inspect tool output text. No secrets (API keys, AWS, tokens) are requested or required by the skill.
Persistence & Privilege
always:false (no forced/global inclusion). The skill instructs users to copy hooks and workspace files into their OpenClaw home (e.g., ~/.openclaw/), which, if enabled, gives it recurring presence and the ability to inject reminders into each session. This is expected for a self-improvement hook, but you should treat enabling hooks as granting ongoing influence over session context (opt-in rather than automatic).
Assessment
This skill is coherent with its stated goal of logging and promoting learnings and appears safe to inspect and opt into, but take these precautions before enabling it globally: 1) Review the hook handler (hooks/openclaw/handler.{js,ts}) and the scripts (scripts/activator.sh, error-detector.sh, extract-skill.sh) so you understand what will run and when. 2) Note that enabling hooks writes or references files under your ~/.openclaw (or project) workspace and will inject reminder content into sessions — enable only if you use OpenClaw/Claude-style hooks. 3) The error detector reads CLAUDE_TOOL_OUTPUT (tool output); ensure you’re comfortable with hooks inspecting tool outputs in your environment. 4) If you enable scripts, prefer project-level (not global) configuration for testing; test with dry-run / minimal setup first. 5) The extract-skill.sh includes path checks preventing absolute/parent directory writes, but avoid running it with elevated privileges and verify generated files before promoting them. If you don’t use OpenClaw or hooks, you can still use the logging templates manually without installing hooks.Like a lobster shell, security has layers — review code before you run it.
latestvk97f14a6z8q5s1cwp4w263j7dx83hb1t
License
MIT-0
Free to use, modify, and redistribute. No attribution required.