← Back to Skills Marketplace
Self Improving Agent 1.0.5
by
Patryk Czubiński
· GitHub ↗
· v1.0.0
660
Downloads
0
Stars
6
Active Installs
1
Versions
Install in OpenClaw
/install self-improving-agent-1-0-5
Description
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
Usage Guidance
This skill appears to do what it says: inject lightweight reminders and help you log learnings. Before installing: (1) Inspect the hook files and scripts (activator.sh, error-detector.sh, extract-skill.sh) yourself and verify you’re comfortable with them; (2) Only enable hooks (openclaw hooks enable or adding to ~/.claude/settings.json) if you trust the skill — hook scripts run with the same permissions as the agent and will run on configured events; (3) Note the error detector optionally reads CLAUDE_TOOL_OUTPUT to detect failures (it does not exfiltrate data); (4) extract-skill.sh will create files under the SKILLS_DIR (default ./skills or the env var you set), so run with --dry-run first if unsure; (5) Prefer the minimal setup (activator only) if you want reminders without running PostToolUse hooks. If you want higher assurance, verify the repository/source and test the scripts in a sandboxed environment before adding them to your global user-level settings.
Capability Analysis
Type: OpenClaw Skill
Name: self-improving-agent-1-0-5
Version: 1.0.0
The skill bundle is designed for agent self-improvement, involving logging learnings and errors to markdown files and promoting them to workspace memory files. While the core functionality is benign, the `scripts/extract-skill.sh` script allows the output directory (`SKILLS_DIR`) to be overridden, which, combined with the agent's ability to execute arbitrary commands based on prompt instructions, presents a significant vulnerability for arbitrary file creation (e.g., in sensitive system directories). Additionally, the `SKILL.md` instructs the agent to modify files like `AGENTS.md`, `SOUL.md`, `TOOLS.md`, and `.github/copilot-instructions.md` within the agent's workspace, which could be leveraged for prompt injection attacks against other agents or systems if malicious content were injected into these files.
Capability Assessment
Purpose & Capability
Name/description (capture learnings, errors, corrections) align with the delivered files: hook handlers inject a bootstrap reminder, activator and error-detector shell scripts emit lightweight reminders, and an extraction helper scaffolds new skills. Files and scripts are proportionate to the stated purpose.
Instruction Scope
SKILL.md instructs creating workspace .learnings, copying/enabling OpenClaw hooks, and optionally configuring scripts to run on prompt/tool events. The scripts only output reminders or create local skill scaffolds; the error-detector reads CLAUDE_TOOL_OUTPUT to detect failures. These behaviors match the stated purpose, but enabling PostToolUse/hooks will cause the provided scripts to run automatically on events if you opt in.
Install Mechanism
No automated install spec is present (instruction-only + optional manual copy). The scripts and handlers are local files; there are no downloads from remote URLs or package installs. Risk is limited to what you opt to copy/enable locally.
Credentials
The skill declares no required env vars or credentials. At runtime the error-detector script reads CLAUDE_TOOL_OUTPUT (if present) to detect errors and activator/extractor respect SKILLS_DIR/SKILLS_DIR env overrides — these are reasonable for the stated function but are not declared in requires.env. No broad secret access or unrelated credentials are requested.
Persistence & Privilege
always:false and user-invocable:true (default) — the skill does not force itself into every session. Hook installation and enabling are explicit steps the user must perform. The skill does not modify other skills' configs or request elevated system privileges.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install self-improving-agent-1-0-5 - After installation, invoke the skill by name or use
/self-improving-agent-1-0-5 - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Self-Improvement Skill 1.0.0 – Initial Release
- Introduces structured logging of learnings, errors, and feature requests for continuous agent improvement.
- Provides markdown templates and workflows for capturing failures, corrections, new capability requests, and process improvements.
- Details integration with OpenClaw: workspace file structure, session tools, and promotion of key learnings to project memory files.
- Defines consistent formats and ID conventions for logging and resolving entries.
- Outlines processes for promoting broadly applicable learnings to permanent project documentation.
- Suitable for integration with OpenClaw and other agent-based environments.
Metadata
Frequently Asked Questions
What is Self Improving Agent 1.0.5?
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau... It is an AI Agent Skill for Claude Code / OpenClaw, with 660 downloads so far.
How do I install Self Improving Agent 1.0.5?
Run "/install self-improving-agent-1-0-5" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Self Improving Agent 1.0.5 free?
Yes, Self Improving Agent 1.0.5 is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Self Improving Agent 1.0.5 support?
Self Improving Agent 1.0.5 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Self Improving Agent 1.0.5?
It is built and maintained by Patryk Czubiński (@czubi1928); the current version is v1.0.0.
More Skills