# Compose Skills

> Use when the user says "compose skills," "which skills should I use," "select a recipe," "combine these skills," or "/agent-collab:compose-skills." Also offer this when a task needs multiple lenses and benefits from explicit context, fan-out, authority, and stop-condition limits.

- Skill: `sumitake/compose-skills` (Agent Skill)
- Install (CLI): `npx skillmds@latest add sumitake/compose-skills`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sumitake/compose-skills/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: sumitake (https://skillmd.com/u/sumitake)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/sumitake/compose-skills

---


# Compose a bounded skill set

Choose the smallest useful set: one primary lens, one supporting lens, and at
most one verifier unless the task demonstrates a larger need. Useful supporting
lenses include `brainstorm`, `context`, `delegate`, and `code-review`.

Recipes select skills, not providers, models, or authority. A runtime request
uses one of the closed semantic actions documented in the plugin README.
Architecture/review/context stay read-only; codegen returns a private patch;
governance remains a distinct authoritative artifact.

Return selected lenses, context each receives, authority, fan-out limit,
execution order, evidence, and stop conditions. Do not load every plausible
skill, turn a recipe into routing policy, or use composition to bypass an
untrusted-source audit.