← Back to Skills Marketplace
Self-Improving Meta
by
José I. O.
· GitHub ↗
· v1.2.1
· MIT-0
102
Downloads
0
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install self-improving-meta
Description
Improves the agent infrastructure itself — core prompt files (AGENTS.md, SOUL.md, TOOLS.md, CLAUDE.md), skill definitions, hooks, rules, extensions, and memo...
Usage Guidance
This package is coherent with its stated purpose (capturing learnings, scaffolding meta-skills, injecting reminders) but contains tools that can write skill files and promote changes to the very prompt files other skills depend on. Before installing or enabling hooks:
- Audit the scripts (scripts/extract-skill.sh, activator.sh, error-detector.sh) and the hook handlers to confirm they do only what you expect.
- Do NOT enable PostToolUse / automated promotion workflows until you trust the behavior; prefer enabling only the lightweight UserPromptSubmit activator for reminders.
- Keep the skill disabled in production or shared workspaces until you test it in an isolated workspace. Back up AGENTS.md, SOUL.md, TOOLS.md, and any shared prompt files first.
- Ensure the extract-skill.sh cannot be triggered autonomously by an agent without explicit human confirmation; prefer a dry-run or manual invocation model.
- If possible, restrict autonomous invocation for this skill or require explicit user approval for any promotion/change operations.
If you want a safer posture, install for local experimentation only, review the code, and enforce human-in-the-loop promotion of any meta-learning before changes are applied.
Capability Analysis
Type: OpenClaw Skill
Name: self-improving-meta
Version: 1.2.1
The self-improving-meta skill bundle enables an AI agent to monitor its own performance and 'self-improve' by logging infrastructure issues and modifying its core configuration and instruction files (e.g., AGENTS.md, SOUL.md). While the stated intent is system optimization and the instructions include safeguards like human review, the capability for an agent to rewrite its own foundational prompts and scaffold new skills via scripts/extract-skill.sh represents a high-risk self-modification pattern. This creates a potential vector for persistent prompt injection where an agent could be influenced to permanently alter its behavioral guidelines or system rules across all future sessions.
Capability Assessment
Purpose & Capability
The name/description match the provided files: the skill logs meta-learnings, provides templates, and includes helper scripts/hooks to create/promote meta-skills. Scripts that create files under ./skills and .learnings are coherent with a meta-improvement workflow (promotion/extraction of learnings).
Instruction Scope
SKILL.md emphasizes a manual-first workflow and explicitly warns not to auto-modify core prompt files, but elsewhere uses phrasing like "Apply fixes directly" and provides an extract-skill.sh capable of creating skill folders and SKILL.md files. There are also helper references to sessions_spawn/sessions_send (cross-session actions). The combination of clear manual-safety guidance and ambiguous statements that permit direct modification is inconsistent and could let an agent (if permitted) make broad changes without adequate human review.
Install Mechanism
No install spec or remote download is used by the skill package itself — all files are shipped in the skill. Manual install instructions suggest a git clone from a GitHub repo (normal). No extract/install-from-URL behavior or archive extraction is present in the packaged files.
Credentials
The skill declares no required environment variables or credentials. One script (error-detector.sh) reads CLAUDE_TOOL_OUTPUT when provided by a hook context; this is expected for an error-detector hook and is documented in SKILL.md. There are no unrelated secrets requested.
Persistence & Privilege
The skill is not flagged always:true, and hooks are opt-in, but it is a meta-skill that can (when run) create/modify skill files and prompt files. Because model invocation/autonomous invocation is allowed by default on the platform, allowing full autonomy plus enabling the hook or running extract scripts increases blast radius. The package does not itself force persistent system changes, but its capabilities are high-impact and should be constrained.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install self-improving-meta - After installation, invoke the skill by name or use
/self-improving-meta - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.2.1
**Version 1.1.0**
- Added stackability contract for multi-skill installations.
- Added namespaced logging guidance (`.learnings/meta/`) for coexistence with other skills.
- Added required `Skill: meta` metadata field and cross-skill precedence/ownership rules.
- Clarified hook arbitration model (single dispatcher, dedupe, rate limiting).
v1.1.0
**Summary:** This update adds explicit safety boundaries to prevent unwanted auto-modification of infrastructure files and clarifies that human approval is required for all core changes.
- Added "Safety Boundaries" section, emphasizing manual review and user approval before changes to core files.
- Clarified that manual workflow is default; hooks for reminders are opt-in.
- Updated guidance for promoting meta-learnings: now requires explicit user approval before modifying `AGENTS.md`, `SOUL.md`, `TOOLS.md`, or `MEMORY.md`.
- Improved language about sensitive information and proper scoping of fixes.
v1.0.0
self-improving-meta 1.0.0
- Initial release of the self-improving meta skill for agent infrastructure enhancement.
- Provides structured logging for meta learnings, issues, and feature requests in `.learnings/` directory.
- Supports first-use initialization, safe file creation, and clear logging procedures for various infrastructure problems.
- Outlines direct promotion paths for critical learnings to core prompt and config files (e.g., AGENTS.md, SOUL.md).
- Includes setup steps and integration guidance for both OpenClaw and generic agent environments.
Metadata
Frequently Asked Questions
What is Self-Improving Meta?
Improves the agent infrastructure itself — core prompt files (AGENTS.md, SOUL.md, TOOLS.md, CLAUDE.md), skill definitions, hooks, rules, extensions, and memo... It is an AI Agent Skill for Claude Code / OpenClaw, with 102 downloads so far.
How do I install Self-Improving Meta?
Run "/install self-improving-meta" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Self-Improving Meta free?
Yes, Self-Improving Meta is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Self-Improving Meta support?
Self-Improving Meta is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Self-Improving Meta?
It is built and maintained by José I. O. (@jose-compu); the current version is v1.2.1.
More Skills