← Back to Skills Marketplace
282
Downloads
0
Stars
11
Active Installs
1
Versions
Install in OpenClaw
/install self-improving-agent-3
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 coherent and low-risk for its stated purpose. Before enabling it: 1) review and, if needed, run the shell scripts in a safe/test workspace (they only print reminders or scan local tool output for error patterns); 2) enable hooks only at the project level (not globally) unless you trust the behavior site-wide; 3) note that promoting learnings or using sessions_history/sessions_send uses OpenClaw session features that can access other session transcripts—only use those when appropriate; 4) if you plan to use extract-skill.sh, keep it in a project workspace (it refuses absolute and .. paths but will create files under the given relative path); and 5) set executable permissions intentionally (chmod +x) and inspect scripts before running. Overall this skill is internally consistent with its description.
Capability Analysis
Type: OpenClaw Skill
Name: self-improving-agent-3
Version: 0.1.0
The 'self-improving-agent' skill bundle is designed to help AI agents log errors, user corrections, and new insights into a structured `.learnings/` directory to improve future performance. It utilizes shell scripts (e.g., `activator.sh`, `error-detector.sh`) and OpenClaw hooks to inject reminders into the agent's context and automate the creation of new skill templates via `extract-skill.sh`. The logic is transparent, lacks any data exfiltration or unauthorized network activity, and includes basic security checks in the extraction script to prevent path traversal.
Capability Assessment
Purpose & Capability
Name/description (self-improvement / capture learnings) align with the provided files: README-style SKILL.md, templates, logging files, hook handlers, and small helper scripts. Required env/config/credentials are none, which matches the non-privileged nature of the task.
Instruction Scope
SKILL.md instructs agents to create and update local .learnings files, promote entries to workspace docs, and optionally use OpenClaw session utilities (sessions_list, sessions_history, sessions_send, sessions_spawn). Those inter-session utilities are platform features — their mention is in-scope for a cross-session learning workflow but you should be aware they read or send session transcripts when used. The scripts/hooks themselves only inject text or inspect tool output (CLAUDE_TOOL_OUTPUT) and do not attempt network exfiltration.
Install Mechanism
No install spec is present (instruction-only), which is low risk. The repository contains optional hook files and three small shell scripts plus hook handlers; manual copy instructions target user-owned paths (~/.openclaw, project dirs). The extract-skill.sh helper writes new skill scaffolds but includes checks (rejects absolute paths and '..') and requires explicit invocation.
Credentials
The skill declares no environment variables, no credentials, and no config paths. The scripts read CLAUDE_TOOL_OUTPUT (documented) for error detection — that is proportional to the stated purpose of detecting command errors. There are no unexpected SECRET/TOKEN environment requirements.
Persistence & Privilege
always is false and the skill does not request permanent/system-wide privileges. Hooks are opt-in and must be explicitly enabled. The provided activator and error-detector scripts run with the same user permissions as the agent — normal for hooks — and the extract script writes only to a relative ./skills directory (with checks).
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install self-improving-agent-3 - After installation, invoke the skill by name or use
/self-improving-agent-3 - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.0
Initial release of the self-improving-agent-3 skill for OpenClaw and generic coding agents.
- Captures learnings, errors, and feature requests in structured markdown logs for continuous self-improvement.
- Provides clear guidelines for when and how to log corrections, failures, knowledge gaps, and best practices.
- Includes workflow for promoting important insights to project-wide memory files (e.g., AGENTS.md, SOUL.md, TOOLS.md).
- Supports both OpenClaw workspace integration and generic agent/project setups.
- Defines standardized formats for logging learnings, errors, and feature requests, including metadata and ID generation.
- Explains workflow for resolving and promoting logged entries to enhance agent and project effectiveness over time.
Metadata
Frequently Asked Questions
What is Self Improvement?
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 282 downloads so far.
How do I install Self Improvement?
Run "/install self-improving-agent-3" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Self Improvement free?
Yes, Self Improvement is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Self Improvement support?
Self Improvement is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Self Improvement?
It is built and maintained by Buradly (@buradly); the current version is v0.1.0.
More Skills