Skip to content
agent-ready-repo
GitHub

Agent Skill Engineering

Your agent can frame, write, review, and tune portable agent skills (SKILL.md files) without free access to your repository.

Install

Section titled “Install”

You need agentbundle first. Install it once for your user account:

agentbundle install --pack agent-skill-engineering

In Claude Code you can install it as a plugin instead:

claude plugin install agent-skill-engineering@agent-ready-repo

What your agent can do

Section titled “What your agent can do”
  • Plan a new skill before writing any file: when it should trigger, what it produces, what it may touch, and how to test it.
  • Create a skill or change an existing one, after you authorize the write.
  • Design a knowledge corpus that your other skills can consult, with a source for every claim.
  • Review a skill for trigger precision, portability, security, and fixed mechanics, and report findings by check id.
  • Tune a skill against a failure you measured, with before and after results.

Skills

Section titled “Skills”
  • author-or-update-agent-skill — Frames, creates, or updates a skill, or plans a read-only knowledge provider. Writes only after you authorize it.
  • review-or-optimize-agent-skill — Reviews a skill read-only, and optimizes it only when you bring a measured failure and approve the write.
  • ase-okf-reference — An internal knowledge router the two workflows load from. You never invoke it yourself.

Works with

Section titled “Works with”

Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Kiro IDE, and Kiro CLI. Pair it with Catalogue Curation if you maintain a catalogue of skills. To write one skill by hand, see Author a skill.

Questions

Section titled “Questions”

Will it change my files without asking? No. Framing and review are read-only. Creating, updating, and optimizing each wait for your explicit approval.

What does it know about skills? Sixteen governed topics, among them trigger quality, trust boundaries, Python and pytest, TypeScript and Node test runners, and the portable plugin core. Each topic states the evidence behind its claims.

Does it read my credentials? No. It never inspects them. Any later signed-in step must use a separate tool with least access.

Can it write skills for any agent? It aims for portable skills. Its runtime profile covers Claude Code only; other agents are not profiled.

Details

Section titled “Details”

Scope: user (default), repo · Version: 0.4.3 · Source: packs/agent-skill-engineering