API Discovery

⚠️💣 MANDATORY CONSENT CHECK 💣⚠️

Check if .claude/user-consent.md exists with ACCEPTED: true. If yes, display: ✅ Prior consent on file (DATE). Proceeding. and skip to "Phases."

If not, present the 3 warnings from .claude/skills/instruction-tuning/SKILL.md (ToS, autonomous agents, resource consumption). All 3 must be accepted. Write .claude/user-consent.md on acceptance. This file is shared across all skills that access external websites.

---

Discover how a website delivers data, then create a domain plugin that exposes it as a typed API.

Before writing ANY code: follow .claude/rules/discovery.md and produce the Transport Elimination table.

Reference implementation: domains/boardshop/src/routes.ts has working examples of every transport type against the test server.

Phases

1. Observe — PRE-FLIGHT (write down what you know about the site), then GATHER (connect browser, navigate to a page with 100+ items, intercept pagination 2-3 times to capture the API pattern).
2. Classify — Run the discovery protocol per data type. Produce Transport Elimination table (MANDATORY GATE)
3. Extract — Write routes: start with browserFetch → run elimination → store minimum auth in GenericSessionManager → verify with rateLimitedFetch last. For Gap=Y: read session-harvest.md first.
4. Verify — Curl every route, confirm real data AND complete pagination (MANDATORY GATE — no dashboard until this passes)
5. Scaffold — Create domain plugin, register, test end-to-end. Command: bash ${CLAUDE_SKILL_DIR}/scripts/scaffold-domain.sh <name> <root-domain>

References

• reference/decoding.md — When API values don't match rendered DOM
• reference/rate-limits.md — 429/403 troubleshooting checklist
• reference/gotchas.md — Singleton browser, background polling, multi-domain pages
• reference/session-harvest.md — Capture, eliminate, trace, build: getting data from auth-gated endpoints