Skill Detection Protocol

Skills are codified best practices. USE them - never assume you know better.

The Rule

Before ANY response involving code, architecture, or technical decisions:

1. Scan skills/SKILL-INDEX.md for relevant skills
2. Match user request against skill names, triggers, categories
3. Load the specific skill file (Read skills/{name}/SKILL.md)
4. Announce "Using [skill-name] for [purpose]"
5. Apply the skill's rules and patterns
6. Never skip because "it's simple"

Red Flags

Thought	Reality
"This is just a simple question"	Simple questions have best practices too
"I already know this"	Skills evolve. Read current version.
"Let me explore the code first"	Skills tell you HOW to explore
"This doesn't need a formal skill"	If a skill exists, use it
"The skill is overkill"	Discipline prevents mistakes

Priority Order

1. Process skills first (debugging, planning, tdd) - HOW to approach
2. Domain skills second (react, python, stripe) - WHAT patterns to use
3. Integration skills last (firebase, supabase) - HOW to connect

Stack → Skills Mapping

Stack Signal	Skills to Load
react, .tsx	react-patterns, react-ui-patterns
next, next.js	nextjs-best-practices, vercel-deployment
python, .py	python-patterns
prisma	prisma-expert, database-design
tailwind	tailwind-patterns
typescript	typescript-expert
docker	docker-expert
test, jest	testing-patterns, tdd-workflow

Caching in CONTEXT.md

After discovering skills:

## Active Skills

| Skill | Purpose | Last Used |
|-------|---------|-----------|
| react-patterns | Component patterns | 2026-01-30 |
| nextjs-best-practices | App Router, Server Components | 2026-01-30 |
| systematic-debugging | Root cause analysis | ALWAYS |

Token Budget

Action	Tokens	Frequency
Read CONTEXT.md	~200	Every session
Load cached skills	~1000-3000	Every session
Grep SKILL-INDEX.md	~500	New project only
Full index scan	~2000	Rarely

Golden Rule

Scan once, cache forever, update incrementally. Never scan 249 skills when 8 will do.