AI Steer Documents — Implementation Guide¶

How to Use¶

Each steer doc is a paste-ready Claude Code prompt covering one implementation session across both repos.

Session Flow¶

1. Open Claude Code in the target repo
2. Say: "Read specs/ai-steer/steer-0X-name.md and the referenced spec files"
3. Claude Code reads the spec for context, then follows the steer instructions
4. Run verification commands at the end of each section
5. Complete the checklist before committing

Session Ordering¶

Steer docs are numbered in dependency order:

steer-01 (EHR Builder)         ← Foundation: everything writes to this
    ↓
steer-02 (Event System)        ← SSE + QStash: real-time plumbing
    ↓
steer-03 (Intake Agent v2)     ← First agent using EHR + events
    ↓
steer-04 (Clinical Autochain)  ← Document → agent → FHIR → EHR chain
    ↓
steer-05 (Matching Pipeline)   ← Reads from EHR, uses events
    ↓
steer-06 (Chat Extraction)     ← Real-time entity extraction from conversation
    ↓
steer-07 (Explanation & Locale)← Multilingual output, final agent
    ↓
steer-08 (Demo Sweep)          ← Full journey verification + polish

Checklist (Applied to Every Session)¶

• [ ] Alembic migrations (if DB changes)
• [ ] Seed data updated
• [ ] Error codes (domain-prefixed: AGENT_, EHR_, etc.)
• [ ] Feature flags (Flagsmith key registered)
• [ ] CORS (if new endpoints)
• [ ] Env vars (documented in .env.example)
• [ ] Swagger/OpenAPI updated
• [ ] Unit tests (pytest)
• [ ] Playwright E2E (if frontend changes)
• [ ] Accessibility (WCAG 2.1 AA)
• [ ] Performance (no N+1 queries, lazy loading)
• [ ] PostHog events (frontend analytics)
• [ ] SEO meta tags (if new pages)
• [ ] Security headers
• [ ] Health check endpoint still passes
• [ ] Loading states (skeleton/spinner)
• [ ] Mobile responsive
• [ ] CLAUDE.md updated (both repos)
• [ ] API changelog entry
• [ ] Rollback plan documented

Repos¶