# Meta Skill Routing

> Strategies for dynamic skill discovery to minimize context overhead.

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

---


# Meta-Skill Routing

When a library has dozens of skills, you cannot load them all. You must "route" your attention effectively.

## The Two-Step Pattern
1.  **Index Lookup**: Read the root `README.md` or a dedicated index to identify which skills are relevant to the current request.
2.  **Explicit Load**: Only load the content of the specific `SKILL.md` files needed.

## Routing Heuristics
- **Keyword Matching**: Look for domain-specific keywords (e.g., "Flask", "TDD", "Transformers").
- **Intent Discovery**: Ask, "Is this a planning, implementing, or verifying task?" and route to the corresponding skill suite.

## Best Practices
- **Atomic Loading**: Avoid "just in case" loading. Only load a skill when you are about to perform an action related to it.
- **Reference Over Inclusion**: Link to skills in documentation instead of copying their content.