Self Improving Agent (Alex)
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· 60·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
Capability signals
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
OpenClaw
Benign
high confidencePurpose & Capability
The name/description (capture learnings, errors, promote to workspace memory) aligns with the supplied SKILL.md, hook handlers, and scripts. The files implement logging templates, workspace injection, an activator reminder, an error-detector that pattern-matches tool output, and a helper to scaffold extracted skills — all coherent with a self-improvement/logging skill.
Instruction Scope
Runtime instructions focus on creating ~/.openclaw workspace files, enabling an optional hook, and appending entries to .learnings/*.md. The hooks/scripts only inject reminders or scan tool output and output guidance; they do not attempt to read unrelated system files or transmit data externally. The error-detector reads CLAUDE_TOOL_OUTPUT (a platform-provided context variable) which is expected for PostToolUse hooks.
Install Mechanism
There is no automated install spec; installation is manual (cloning from a GitHub repo or copying files). That is reasonable for this type of skill, but cloning code from an external repository is an action the user should review (the repo URL is not a well-known project site in the metadata). The included scripts are plain shell/JS and are extractable for review before use.
Credentials
The skill requests no credentials, no config paths, and no special binaries. The only environment access is to CLAUDE_TOOL_OUTPUT in error-detector.sh (used to detect command errors) and to SKILLS_DIR/SKILLS_DIR env var in extract-skill.sh for output location — both are proportional and not secrets. No unexpected tokens/keys are requested.
Persistence & Privilege
always:false (opt-in). The skill can inject a virtual reminder at agent bootstrap if you copy/enable the provided OpenClaw hook; enabling that hook gives the skill presence at session startup but requires explicit user action. The scripts create files in the user's skills/workspace directories when used (e.g., extract-skill.sh), which is expected behavior.
Assessment
This skill appears coherent and implements what it claims: logging learnings, injecting optional reminders, and helping extract skills. Before installing or enabling hooks, review the repository and included scripts (hooks/openclaw/handler.{js,ts}, scripts/*.sh) to confirm you’re comfortable with the files being copied into ~/.openclaw and with enabling hook execution. Note: enabling the hook or adding the activator/error-detector to your agent settings will run the small scripts in each session (they only output reminders and perform pattern-matching on tool output), and extract-skill.sh will create files in the configured skills directory. If you don’t want session-wide reminders, skip copying/enabling the hook or use the minimal setup described in the docs.Like a lobster shell, security has layers — review code before you run it.
agentvk97dm8xnfj3jg8c0r6ph2qx4ch84ksb0latestvk97dm8xnfj3jg8c0r6ph2qx4ch84ksb0learningvk97dm8xnfj3jg8c0r6ph2qx4ch84ksb0self-improvementvk97dm8xnfj3jg8c0r6ph2qx4ch84ksb0
License
MIT-0
Free to use, modify, and redistribute. No attribution required.