ask-code-reviewer

title	Skill
type	entity
tags	["skill","library","yaml","SKILL.md","frontmatter"]
updated	"2026-04-06T00:00:00.000Z"
sources	4

Skill

A skill is the core unit of ASK — a reusable, versioned instruction set for an AI agent. Skills live in the central library and can be deployed to one or more agents.

Directory Structure (Gold Standard)

skills/<category>/<skill-name>/
├── skill.yaml     # Machine-readable metadata
├── SKILL.md       # Human+LLM instruction file
├── scripts/       # Helper scripts (required for validation gate)
└── tests/         # Tests (required for validation gate)

skill.yaml Fields

name: ask-code-reviewer
version: 1.2.0
agents: [claude, gemini]       # which agents this targets
depends_on: []                  # other skill names (resolved by SkillRegistry)

SKILL.md Frontmatter

---
name: ask-code-reviewer
description: Brief description of what the skill does
triggers: ["review my code", "check this PR"]
---

# Skill content here...

Naming Convention

• kebab-case, 2–50 characters.
• All library skills are prefixed with ask- by convention.
• Validated by ask/utils/validators.py.

Categories

Category	Purpose
coding/	Language/framework-specific dev skills
planning/	Architecture, ADRs, project management
tooling/	Meta-skills (skill creation, context, auditing)
workflows/	Multi-step process skills

Lifecycle

1. Author — create skill dir, write skill.yaml + SKILL.md.
2. Validate — ask validate checks structure; ask skill lint checks token limits.
3. Deploy — ask copy <skill> deploys to an agent via the appropriate adapter.
4. Update — bump version in skill.yaml; ask copy detects version delta.

Dependency Resolution

SkillRegistry resolves depends_on chains with cycle detection. A skill won't install unless its dependencies are satisfied.

Token Limits

ask/utils/token_analyzer.py enforces per-skill-type token limits via tiktoken. ask skill lint surfaces violations.

See skills-catalog.md for all available skills.