Improve

Improve

Purpose

Capture reusable self-improvement patterns that shape future agent work without overwriting existing learning.

When to Use

Use after a session reveals a reusable self-improvement pattern.

Workflow

1. Identify a pattern: a repeated mistake, useful tactic, repo convention, or

verification habit likely to matter again.

1. Read .agents/memory/patterns.json before writing. Preserve every existing

pattern unless the update is an intentional merge.

1. Add a concise pattern with: name, trigger (when to use), action (what to do),

evidence (why it matters), confidence (0-1), and sources (files or sessions).

1. Keep entries repository-specific and durable. Do not store secrets, transient

task status, commit hashes, or one-off TODOs.

1. Reuse relevant patterns during preflight, implementation, audit, and ship.

Pattern Store Contract

• Schema version is 1. Top-level fields: version (number) and patterns (array).
• Each pattern must have: name, trigger, action, evidence,

confidence (number 0-1), sources (array of strings), createdAt (ISO string).

• Deduplication is by id. If id is missing, it is generated from name in kebab-case.
• Merging keeps the higher confidence, combines and deduplicates sources, and keeps

the older createdAt.

• Invalid stores fail loudly: read and update operations throw rather than silently

recovering or overwriting user data.

Output

Updated .agents/memory/patterns.json with the new pattern merged in.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mverab
• Source: mverab/slashstack-skills
• License: MIT
• Homepage: https://slashstack.dev

Install and usage instructions live in the source repository linked above.