---
title: Skill routing
description: Route Expo and EAS work without implying audits or cloud actions.
---

# Skill routing

Use `$expo-development-foundation` when a task needs foundation-aware routing,
an existing consumer-app audit, an explicit catalog audit, a new Expo app, or
an existing Expo app that must be oriented before edits. It reads
`.agents/skills/manifest.json` (and the
generated `master-skill.json` catalog), applies decision trees and failure-mode
stops, distinguishes consumer, showcase, and foundation work, then reads only
shortlisted skills.

## Decision trees (required before shortlist)

Master skill references (read from the skill directory, not site pages):

- `.agents/skills/expo-development-foundation/references/decision-trees.md`
  — workspace class, Expo Go vs development client, UI stack, verification
  ladder, JS vs config vs native
- `.agents/skills/expo-development-foundation/references/foundation-invariants.json`
  — canonical runtime, security, authorization, workspace, UI, and verification policy
- `.agents/skills/expo-development-foundation/references/agent-failure-modes.md`
  — high-cost stop and redirect rules

## Workspace branches

| Class | First skill | Notes |
| --- | --- | --- |
| New Expo app | `$expo-bootstrap` | Then ordered structure, Router, UI, optional Tailwind, data, DOM/web, modules, upgrades, authorized EAS |
| Existing Expo app | `$expo-project-orientation` | Stack card + shortlist; stop unless implementation is authorized; never greenfield-restructure |
| Existing consumer-app audit | `$expo-project-orientation` → `$expo-app-audit` | Read-only; relevant domain skills; optional `$expo-design-review`; close with `$expo-verify` |
| Foundation audit | master audit path | Only with **explicit** audit language; ordinary routing is not an audit |
| Non-Expo | — | Do not force catalog skills |

Routing does not authorize implementation, telemetry, credentials, paid EAS
commands, a native build, or either audit scope. Vague requests such as “make
this app better” orient and recommend next steps; they do not silently become a
full audit or implementation.
