Self-Improving AI
v1.1.0Captures learnings about GenAI/LLM configuration, model selection, inference optimization, fine-tuning, RAG pipelines, prompt engineering, multimodal process...
⭐ 0· 48·0 current·0 all-time
byJosé I. O.@jose-compu
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
Name/description align with what the files do: logging AI/LLM learnings, providing templates, and helper scripts to scaffold new skills. The provided hooks and scripts (activator, error detector, extract-skill) are appropriate for an onboarding/reminder-and-scaffold skill. There are no unrelated environment variables, binaries, or external credentials requested.
Instruction Scope
SKILL.md and hooks instruct the agent/user to create and update .learnings/ files in project or workspace roots and to 'promote' learnings into workspace files like SOUL.md, AGENTS.md, TOOLS.md. The code itself only injects reminder content and provides scripts; it does not automatically exfiltrate data. However the workflow explicitly asks the agent (or user) to log model outputs, parameters, token-usage and latency — and to avoid logging API keys/PII. That relies on correct redaction by the agent/user; accidental inclusion of sensitive data is possible if not carefully filtered.
Install Mechanism
There is no automated install spec; the repo suggests manual git clone or a platform-specific 'clawdhub' command. All code is included in the skill bundle (hooks and shell scripts). No network downloads or archive extraction occur during install, so install-risk is low.
Credentials
The skill declares no required env vars or credentials. One script (error-detector.sh) reads the CLAUDE_TOOL_OUTPUT environment variable — this is consistent with a PostToolUse hook and expected for error detection. Because CLAUDE_TOOL_OUTPUT may contain sensitive outputs, the SKILL.md warns not to log keys/PII; enforcement is manual. No other env vars, tokens, or config paths are requested.
Persistence & Privilege
always:false and the hooks are opt-in. The OpenClaw hook will inject a virtual reminder file during agent bootstrap if enabled; activator/error-detector scripts run with the agent's permissions when configured. This is expected for hooks, but remember hooks/scripts execute locally with the agent’s privileges — enable only where you trust the skill and its code.
Assessment
This skill is internally coherent and low-risk: it only adds local reminder files, templates, and small helper scripts. Before enabling hooks or copying scripts into your ~/.openclaw workspace: (1) review the scripts (activator.sh, error-detector.sh, extract-skill.sh) to ensure they match your expectations; (2) prefer enabling only the UserPromptSubmit activator if you want lightweight reminders and skip PostToolUse unless you want automated error detection; (3) do not log API keys, model tokens, customer data, or PII into .learnings/ — the skill asks you to record model parameters and token usage, but redaction is your responsibility; (4) if you enable hooks, consider adding matcher filters so reminders run only for relevant prompts; (5) because the scripts run with agent permissions, enable them only in trusted environments and audit any promotions that write into global workspace files (SOUL.md, AGENTS.md, TOOLS.md). If you want added assurance, run the scripts in a sandboxed/test workspace first.Like a lobster shell, security has layers — review code before you run it.
latestvk97dh31674f8dx7msjb175egj984rfxc
License
MIT-0
Free to use, modify, and redistribute. No attribution required.