← Back to Skills Marketplace
81
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install self-improving-agent2
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: it helps capture errors, corrections, and feature requests into .learnings/ files and can optionally inject a reminder at agent bootstrap. Before installing or enabling hooks: (1) review the scripts (activator.sh, error-detector.sh, extract-skill.sh) yourself — they run with the agent user's permissions and will create files in your workspace; (2) only enable the PostToolUse / boot hooks if you trust the runtime, since hooks will see tool outputs (the error detector reads CLAUDE_TOOL_OUTPUT); (3) follow the guidance in SKILL.md to avoid logging secrets or full config files; (4) note minor metadata inconsistencies (naming/ownerId) in the package manifest — not a functional problem but worth sanity-checking against the repo you intend to install. If you want to be extra cautious, install manually and enable hooks one at a time (start with the activator reminder only).
Capability Analysis
Type: OpenClaw Skill
Name: self-improving-agent2
Version: 1.0.0
The skill bundle is a legitimate framework for enabling AI agents to track errors, user corrections, and new insights for continuous improvement. It includes well-structured scripts for error detection and skill extraction, featuring proactive security measures such as path traversal validation in `extract-skill.sh` and explicit instructions in `SKILL.md` to avoid logging sensitive data like secrets or private keys. The use of OpenClaw hooks and workspace files (e.g., SOUL.md, AGENTS.md) is consistent with the platform's design for managing agent behavior and long-term memory.
Capability Tags
Capability Assessment
Purpose & Capability
Name/description match the provided scripts and documentation: the skill creates and updates .learnings/ files, provides templates, optional hook integration, and a helper to extract learnings into new skills. The included scripts/handlers are consistent with that goal. Minor mismatches: the registry slug/version shown in metadata differs from some referenced repository names (repo referenced is 'self-improving-agent' while the registry entry is named 'self-improving-agent2'), and _meta.json ownerId differs from the registry ownerId — likely bookkeeping issues but not a functional mismatch.
Instruction Scope
SKILL.md and the hook scripts stay within the stated scope: creating .learnings files, injecting a reminder at bootstrap, and detecting command output errors. One small mismatch: none of the declared requires.env variables exist (skill declares no required env vars) but scripts/error-detector.sh reads CLAUDE_TOOL_OUTPUT — this is presumably a runtime variable supplied by the agent environment, not a secret the skill needs. The README explicitly warns not to log secrets and to sanitize outputs, which is good. Hooks and extract-skill.sh will create files and write scaffolds in the workspace — this is expected behavior for this skill but should be enabled only when you trust the environment.
Install Mechanism
No automated install spec is included (instruction-only with optional manual steps). Recommended installs use clawdHub or git clone from GitHub; both are standard and expected. There are no downloads from untrusted URLs, no archives/extract steps, and scripts are local and inspectable.
Credentials
The skill requests no credentials or sensitive environment variables. It does reference agent-provided functionality (e.g., CLAUDE_TOOL_OUTPUT and OpenClaw session APIs like sessions_history/sessions_send) — reasonable for an in-agent logging/reminder tool. The documentation explicitly warns about treating tool output as potentially sensitive and not forwarding secrets, which mitigates risk. Cross-session sharing features are available but are documented as opt-in and should be used only in trusted contexts.
Persistence & Privilege
always is false and the skill does not request permanent platform-wide privileges. Enabling the optional hook will cause the provided scripts/handler to run at bootstrap or on events with the same user permissions as the agent — this is expected for hooks. The skill does create files under the workspace or skills directory (extract-skill.sh writes under ./skills by design and checks paths), which is consistent with its purpose.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install self-improving-agent2 - After installation, invoke the skill by name or use
/self-improving-agent2 - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release of the self-improving agent skill for OpenClaw and general coding agents.
- Captures errors, learnings, corrections, and feature requests in structured Markdown files for continuous improvement.
- Automatic initialization/check of `.learnings/` directory and log files (`LEARNINGS.md`, `ERRORS.md`, `FEATURE_REQUESTS.md`), with logic to avoid overwriting existing data.
- Provides clear guides on when and how to log corrections, knowledge gaps, user feedback, and best practices.
- Details workflow for promoting broadly useful insights to workspace documentation (`AGENTS.md`, `SOUL.md`, etc.).
- Includes OpenClaw-focused setup and integration instructions, plus generic guidance for other environments.
- Offers standardized, privacy-respecting log formats and recommendations for agent collaboration and cross-session sharing.
Metadata
Frequently Asked Questions
What is Self Improving Agent2?
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 81 downloads so far.
How do I install Self Improving Agent2?
Run "/install self-improving-agent2" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Self Improving Agent2 free?
Yes, Self Improving Agent2 is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Self Improving Agent2 support?
Self Improving Agent2 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Self Improving Agent2?
It is built and maintained by dashi6174 (@dashi6174); the current version is v1.0.0.
More Skills