self-improve
Self-Improve
Overview
Session learning loop that captures reusable knowledge and routes it to the right place: project memory, global CLAUDE.md, or a new skill. Builds on skills-management methodology for skill creation (see writing-skills reference).
Process
1. Identify Learnings
After a productive session, review what was learned:
- Patterns that worked well
- Mistakes that were caught and corrected
- Techniques that could be reused
- Debugging insights
- Tool usage patterns
- Project-specific conventions discovered
2. Classify Each Learning
Read references/classification-routing.md for the full routing table (priority 1-5) and draft format per destination. Key rule: the instinct to put things in ~/CLAUDE.md is usually wrong — route to existing skills or project CLAUDE.md first.
3. Confirm with User
Present all proposed changes grouped by destination. For each:
- What was learned (1 sentence)
- Where it goes and why
- Exact content to add/modify
Never auto-apply changes. Always get explicit approval.
4. Apply Approved Changes
Write changes to the approved destinations. For new skills, note that full TDD testing (per skills-management writing-skills reference) should happen in a dedicated session.
Common Mistakes
- Defaulting to ~/CLAUDE.md — most learnings belong in a more focused location (existing skill, project CLAUDE.md, or memory). ~/CLAUDE.md is the last resort, not the default.
- Auto-applying changes without user confirmation
- Saving session-specific context as permanent knowledge
- Creating skills for one-off solutions
- Duplicating what's already in CLAUDE.md or existing skills
More from hjewkes/agent-skills
md-render
Use when asked to render, preview, or view a markdown file in the browser. Triggers on "render markdown", "preview this", "show me this document", "open in browser".
21agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
13subagent-driven-development
Use when executing implementation plans with independent tasks in the current session
9coordinator
Use when the user says "vm", "voice mode", "team", "coordinate", or needs to orchestrate multiple agents working on related tasks in parallel
8using-superpowers
Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions
8research
Use when the user asks to research a topic, investigate options, compare technologies, analyze a problem space, or gather information before making a decision — for both coding and non-coding work
8