skill-foundry

Router skill: dispatch one build/certify request to exactly ONE specialist playbook. [DOC]

When to use

Use when the user wants to create, design, audit, certify, benchmark, index, or search an agentic asset (skill / agent / command / prompt / hook / MCP server / workflow): verb = build-an-asset, noun = one of the topics → route here. Do NOT use to run an existing asset, edit unrelated files, or answer domain questions. [DOC]

Procedure

In: topic (one enum value) + depth (quick|deep) + intent. Out: the chosen playbook's artifact (definition / audit / cert verdict / index), tagged and gates-passed. [INFERENCE]

1. Resolve topic against the enum. Map intent → topic via routes.json desc fields (each lists its own trigger phrases). [DOC]
2. Read EXACTLY ONE playbook from routes:. Never load the whole cluster — one route per invocation keeps context lean. [DOC]
3. Apply by depth: quick → essentials path only; deep → exhaustive, with verification at each step. [DOC]
4. Validate against the gate below before reporting done. [DOC]

Spine: Discover → Analyze → Execute → Validate. Gates: constitution v6.0.0 (enforcement), evidence tags, script-first rule. [DOC]

Routing aids live in assets/ — assets/quality-rubric.json (routing scorecard) and assets/checklist.md (spine checklist). [DOC]

Tie-breakers (ambiguous topic) [INFERENCE]

• create a skill → meta-skill-creator; design a skill (spec only) → design-skill.
• improve/production-ize end-to-end → assembly-skill; audit only → x-ray-skill; certify only → certify-skill.
• create a prompt file → prompt-creator; create/review/port a system prompt → prompt-forge.
• workflow YAML/steps/RACI → workflow-creator; slash-command phases → workflow-forge.
• Still ambiguous → ask one disambiguating question; do not guess.

Validation gate (Done = all true)

• Topic matched a valid enum value; exactly one playbook was read. [DOC]
• The playbook's own acceptance criteria/rubric passed (incl. constitution v6.0.0). [DOC]
• Every non-obvious claim carries one Alfa-set tag ([DOC]/[INFERENCE]/[ASSUMPTION], EN spelling); no foreign taxonomy mixed in. [DOC]
• Output is single-brand; no invented prices; no green-as-success. [DOC]

Anti-patterns

• Reading multiple route files "to compare" — pick one, route, done. [DOC]
• Inventing a topic outside the enum, or silently defaulting when intent is unclear. [DOC]
• Skipping the playbook gate and self-declaring success. [ASSUMPTION]

Self-correction triggers

• No enum fits → out of foundry scope; redirect, don't force a route. [INFERENCE]
• Gate fails → fix the artifact and re-run the gate; never ship a failing asset. [DOC]