Self Improving Agent

Most agent "memory" features just dump notes into a file and hope. This skill takes a different line: it treats a failure, a user correction, or a repeated workaround as a candidate lesson that has to earn its way into durable guidance. The CLI captures redacted, deduplicated candidates, runs an executable behavior eval against a representative scenario, and only then lets you apply the change to one named owner, a script, a skill, or project instructions, with a rollback path recorded. That separation of validation from application is the right call, and the ceiling is clear: it's a disciplined process rather than automatic learning, and its hook integration is limited to Claude Code and only when you opt in with `apb init --hooks`. Worth reading the lifecycle reference before you trust it with anything touching real rules.

skills.sh installs: 35.6K
GitHub stars: 87
last updated: Oct 26, 2026

View on GitHub

Install

$ npx -y skills add zhaono1/agent-playbook --skill self-improving-agent --agent claude-code

Installs into .claude/skills of the current project.

Files

Self-Improving Agent

Turn evidence from completed work into a small, auditable behavior change. The default result is a candidate or no change—not an automatic rewrite of skills.

Use This Skill When

A tool or workflow failed in a way that may recur.
The user corrected an assumption, requirement, or operating rule.
The same workaround appeared more than once.
A focused test proved a better reusable method.
The user asks to review or consolidate learning candidates.

Do not use it for routine session summaries, raw transcript storage, speculative ideas without evidence, or project facts that belong in project documentation.