self-improving-agent

# self-improving-agent Captures learnings, errors, and corrections from every session to enable continuous self-improvement for your AI coding agent. Error Logging — Records unexpected command failures, tool errors, and API failures to .learnings/ERRORS.md for future reference. Correction Capture — When a user says "No, that's wrong" or "Actually...", the correction is immediately logged with category correction . Feature Request Tracking — Logs missing capabilities to .learnings/FEATURE_REQUESTS.md so they can be addressed later. Knowledge Gap Detection — Identifies and records outdated or incorrect knowledge under category knowledge_gap . Best Practice Discovery — When a better approach is found for a recurring task, it is saved under category best_practice . ## How It Works 1 Step 1 Trigger Detection The skill activates on specific signals: command failures, user corrections, missing features, API errors, outdated knowledge, or better approaches discovered during a task. 2 Step 2 Structured Logging Each event is appended to the appropriate markdown file in the .learnings/ directory with a consistent format including timestamp, category, context, and resolution. 3 Step 3 Pre-Task Review Before major tasks, the agent reviews existing learnings to avoid repeating past mistakes. 4 Step 4 Promotion Pipeline Learnings confirmed across multiple sessions are promoted to persistent project configuration files such as CLAUDE.md , AGENTS.md , or SOUL.md . ## Key Features - **Zero Configuration**: No API keys or environment variables required; works out of the box with any AI coding agent. - **Markdown-Based Storage**: All learnings are stored as human-readable markdown files, making them easy to review, edit, and version-control. - **Three-File System**: Separates errors, learnings, and feature requests into dedicated files for clarity. - **OpenClaw Integration**: Native support for OpenClaw's workspace-based prompt injection with automatic skill loading. - **Cross-Agent Compatible**: Works with Claude Code, Cursor, Windsurf, and any agent that reads markdown context files. - **Promotion Guidelines**: Built-in rules for escalating important learnings to project-level memory so they persist indefinitely. ## Requirements - **No API keys needed**: This skill requires no external services or credentials. - **No required binaries**: Works entirely through file read/write operations. - **Any AI coding agent**: Compatible with Claude Code, Cursor, Windsurf, or any agent that supports custom instructions or skill files. ## Use Cases - **Persistent Memory Across Sessions**: Prevents the agent from repeating the same mistakes session after session. - **Team Knowledge Base**: Share the .learnings/ folder across a team to accumulate collective agent knowledge. - **Debugging Aid**: Quickly find the root cause of recurring errors by reviewing the structured error log. - **Continuous Improvement Loop**: Gradually improve agent behavior on a project over time without manual prompt engineering. - **Feature Roadmap Input**: Use the feature request log to identify gaps and prioritize tooling improvements. ## Installation Install via: `npx clawhub@latest install self-improving-agent`

Installation

Related Skills