Using agent skills

Overview

This meta-skill teaches you how to use the AI Agent Skills framework. Load it first in any new session.

When to Use

Load this skill at the start of every session when working in a project that has this framework installed.

Process

Step 1: Identify your task type

Map what you're about to do to one of the 8 core commands:

• Defining/scoping → /spec
• Planning/decomposing → /plan
• Implementing → /build
• Testing/verifying → /test
• Reviewing code → /review
• Simplifying code → /simplify
• Deploying → /ship
• Responding to incidents → /incident

Step 2: Load the primary skill

Read the skill file for your task: skills/<name>/SKILL.md

Step 3: Identify supporting skills

Check AGENTS.md for the skill composition table. Load supporting skills if they apply.

Step 4: Check for a relevant pack

If working in a specific domain (AI/ML, data, frontend), load the matching pack from packs/.

Step 5: Execute with verification gates

Follow each step in the skill. Do not skip steps. At each checkpoint, produce evidence.

Anti-Rationalizations

"The task is simple enough that I don't need a skill" Simple tasks still benefit from structured thinking. The overhead of loading a skill is seconds; the cost of a missed step can be days.

"I already know how to do this" Skills encode accumulated engineering wisdom beyond what any individual knows. They also prevent bad days from becoming bad commits.

Verification Requirements

• Identified the correct primary skill for the current task
• Loaded all relevant supporting skills
• Have clear verification requirements before starting work