Skip to content

CLAUDE.md

Latest commit

 

History

History
36 lines (26 loc) · 3.78 KB

CLAUDE.md

File metadata and controls

36 lines (26 loc) · 3.78 KB
Raw

devcon-api

Express + Swagger API for Devcon data. Deployed on Render (api.devcon.org).

Commands

pnpm dev                                   # nodemon on http://localhost:4000
pnpm exec tsc --noEmit                     # type-check before considering a task complete
NODE_ENV=test pnpm exec jest <file> --runInBand   # run tests PER FILE

Hard rules

  • No database. Data is JSON files under data/, loaded into an in-memory store at boot. Git is the store of record: writes (AV enrichment, sync) commit JSON back to the repo.
  • [skip deploy] commit tag: data-only commits made by the API itself use this Render skip phrase so they don't trigger a redeploy. Keep it on automated data commits.
  • Never run the full jest suite in one process - every test file loads the full data store and node runs out of memory. Run per-file with --runInBand.
  • Speaker emails must never leave a sync or endpoint response.
  • Schedule truth flows Pretalx -> sync -> data/ -> memory. Day-of AV enrichment (YouTube IDs, sources) goes through PUT /sessions/sources/:id, not through Pretalx.
  • Community Hub sessions are not in Pretalx and have no files under data/sessions/. GET /events/:id/community-hubs/{version,bundle} builds them live from the hubs' dSheets (COMMUNITY_HUB_SHEETS env, in-memory cache). The only thing on disk is data/community-hubs/<eventId>.json, the last good read per hub, written by the API and committed in production so a redeploy starts from it; never hand-edit it. Keep hub sessions out of the Pretalx bundle. Details: ../docs/community-hubs.md.
  • AI catalogue (GET /events/:id/ai index, GET /events/:id/ai/sessions?day=…): the programme as markdown for attendees' own assistants (ChatGPT, Claude), one line per session with its CODE first (Pretalx code, or the sheet id for a hub row), main programme plus the hubs bundle. Formatter src/utils/ai-catalogue.ts (tested), assembly src/services/ai-catalogue.ts. Assistants quote the codes back into the event app's /my-interests?add=… link, so keep the line format and the codes stable; EVENT_APP_ORIGIN sets the app origin in the index. The app mirrors these pages under its own origin (/ai) because Cloudflare's bot rules on api.devcon.org can refuse an assistant's fetcher; served as text/plain for the same reason.

Supabase migrations

The "no database" rule above is about THIS API's data store. This repo also hosts the schema migrations for the shared Supabase project mealmslwugsqqyoesrxd (used by devcon ticketing/x402 and event-app), at src/supabase/migrations/.

  • Create: add a file named <YYYYMMDDHHMMSS>_<snake_case_name>.sql in src/supabase/migrations/.
  • Apply: pnpm db:migrate src/supabase/migrations/<file>.sql (wraps scripts/db-migrate.sh). It applies exactly that one file via the Supabase management API and records the version in the CLI migration history. Auth comes from the keychain token created by npx supabase login (one-time, interactive terminal only); no DB password involved.
  • NEVER run supabase db push: the migration history diverged from the remote in Feb 2026 (~14 local noon-timestamped files have real-timestamped twins on the remote), so push would try to re-apply old migrations against production. Until that's repaired (verify each stale file's changes exist in the DB, then supabase migration repair), db:migrate is the only safe apply path.
  • Inspect state with npx supabase migration list --workdir src (the supabase dir lives under src/, hence --workdir src on all CLI calls).
  • src/supabase/.temp/ is machine-local CLI state — ignored via src/supabase/.gitignore, never commit it.

Context

  • Pipeline architecture, webhook/resync behavior, AV team workflows: ../docs/av/av-stack-overview.md
  • Handover notes: docs/notes.md