ai-features
Installation
SKILL.md
AI Features
Overview
An AI feature earns its place when it removes a step the user performs by hand today, or answers a question the interface cannot. Everything else is a sparkle button. The skill reads the app first, proposes only features tied to a real column and a real user moment, shows each as a static mock built on example data, and builds the shared plumbing once the user has chosen.
Two principles drive the workflow:
- The app's data decides. A proposal names the table and column it reads. No column, no feature.
- Narrow beats conversational. Narrowly scoped AI features are easier to understand and see better adoption (NN/g); a chat box in front of a working feature is the last thing to propose.
Navigation
| Load | When |
|---|---|
| references/CATALOG.md | Phase 3 — the feature catalog: shipped features by app category, what the user sees, data needed, model tier, and the ones users rejected |
| references/MODELS.md | Phase 3 and 6 — which model for which job, local vs hosted rules, prices, memory and context, dated |
| references/SETUP.md | Phase 6 — runtime detection, the runtime interface, configuration keys, structured output per provider, vectors in SQLite, packaging, hygiene |
| references/EVAL.md | Tuning the description |