atmot-bot#

Daily Bluesky bot for AT Mot. Posts once per language each day just after the UTC puzzle rollover: invites players to today's puzzle and congratulates yesterday's solvers (by count only).

Runs as a stateless Cloudflare Worker with two Cron Triggers (EN 00:10 UTC, FR 00:11 UTC). Counts are read from the public Constellation backlink index; there is no database. The bot duplicates a small set of the app's frozen constants (src/config.ts) rather than importing the app.

Develop#

npm install
npm test          # vitest — composer + facets + puzzle math
npm run typecheck
npm run dev       # wrangler dev --test-scheduled (see Dry run below)

Dry run (no real post)#

Create a gitignored .dev.vars:

DRY_RUN = "1"
ATMOT_BOT_IDENTIFIER = "atmot.herve.bzh"
ATMOT_BOT_APP_PASSWORD = "dry-run-unused"

Then npm run dev and, in another shell:

curl "http://localhost:8787/__scheduled?cron=10+0+*+*+*"   # EN
curl "http://localhost:8787/__scheduled?cron=11+0+*+*+*"   # FR

The composed post is logged instead of published.

Deploy#

The bot posts as @atmot.herve.bzh using a Bluesky app password (Settings → Privacy and security → App passwords — not the account password).

npx wrangler login
npx wrangler secret put ATMOT_BOT_IDENTIFIER     # e.g. atmot.herve.bzh
npx wrangler secret put ATMOT_BOT_APP_PASSWORD   # the app password
npm run deploy

The free Workers plan is sufficient: each language runs in its own scheduled invocation, so each gets the full 50-subrequest budget. Solver counting samples up to SOLVER_SAMPLE_CAP (20) records per language; beyond that the count is hedged (e.g. "20+").

Each sampled record costs ~2 subrequests (DID-document resolution + the record read), so the default cap of 20 already uses most of the per-invocation budget on the free plan — keep headroom for the count, session, idempotency check, and publish (~6 more). On the paid plan you can raise SOLVER_SAMPLE_CAP in src/config.ts, but size it against the subrequest cost (~2 per record), not just the desired sample size.