skill-discovery-and-selection

name	skill-discovery-and-selection
description	Use when starting a session or deciding which local skill applies; acts as the meta-skill for discovering and loading the smallest relevant skill.
aliases	["using-agent-skills","skill-routing"]

Using Agent Skills

When to use

Use this skill when starting a new task, resuming context, or choosing among overlapping local skills in this repository.

Objective

Select the smallest relevant skill or skill combination and explain why it was chosen.

Procedure

1. Read skills_docs/HOW_TO_FIND_THE_RIGHT_SKILL.md for routing by task shape (MCP Path A or filesystem Path B).
2. Open skills/MANIFEST.md or pack metadata when inventory detail is needed.
3. Identify the task shape: discovery, refinement, planning, implementation, verification or release.
4. Pick the smallest relevant local skill set.
5. Load apply-laws-of-ai only when the task involves material AI/agent harm risk, unlawful or unauthorised instructions, or explicit safety-law trade-offs.
6. Prefer the repo's native categories and naming over remote terminology unless the remote term is the clearer local analogue.
7. Load only the supporting docs needed for the task.
8. Re-evaluate if the task shape changes or new constraints appear.

Rules

• Do not treat apply-laws-of-ai as a mandatory session preamble.
• Do not load broad skill sets by default.
• Do not choose a skill based only on its name.
• Do not duplicate planning, verification or release work when an existing local skill already covers it.
• Prefer local analogues when they are more coherent with this repository's structure.

Related skills

• apply-laws-of-ai — safety-law hierarchy when the task shape warrants it
• planning-and-task-decomposition — multi-step work decomposition
• requirements-elicitation — underspecified requests

References

• Anthropic — Equipping agents for the real world with Agent Skills

Verification

• Selected skill(s) and selection rationale reported.
• apply-laws-of-ai loaded only when relevant (or explicitly skipped with rationale).
• Intentionally excluded skills noted.
• Remaining ambiguity or re-check triggers stated.