- TypeScript 100%
- CSS <1%
- Python <1%
Slopgotchi #
A public AI-slop accountability pet for Tangled. Join a team "zoo," and your Tangled pull requests get scored for slop — careless overbuild, weak tests, scope sprawl — by an open model on Featherless. The score and a receipt are written as public ATProto records, and a tamagotchi-style pet renders whose health is a rolling average of your recent slop.
Slopgotchi scores carelessness, not authorship. It does not try to detect whether AI was used, and it is not a merge gate — it's a public accountability signal that lives on-protocol.
Architecture #
Three layers (see docs/plans/2026-06-24-001-feat-slopgotchi-plan.md):
- Protocol (truth):
app.slopgotchi.diagnostic(one per scored PR round) andapp.slopgotchi.pet.state(one per developer) records on the service account's ATProto repo. World-readable; any client can re-render a pet from them. - Pipeline (compute): resolve handle → DID → PDS,
listRecordsthesh.tangled.repo.pullrecords, gunzip the inline patch blob, score on Featherless, write records, recompute health. Triggered by join (backfill) and by cron (ongoing) — one idempotent loop keyed on(PR, round). - App (experience): the Next.js zoo UI and join flow.
Stack #
- Next.js (App Router) on Vercel
- Upstash Redis — the single datastore (idempotency, registry, cached pet state)
- Featherless (OpenAI-compatible) for inference
@atproto/api+@atproto/identityfor protocol reads/writes- Zod for record/score validation, Vitest for tests
Local setup #
npm install
cp .env.example .env.local # fill in the values below
npm run dev
Environment #
| Var | Purpose |
|---|---|
UPSTASH_REDIS_REST_URL / _TOKEN |
Upstash Redis. Run vercel install upstash to provision and auto-inject. |
FEATHERLESS_API_KEY |
Featherless inference. |
FEATHERLESS_MODEL |
Optional model override (default Qwen/Qwen2.5-7B-Instruct; Qwen2.5-Coder-32B is higher quality but often capacity-queued). |
SLOPGOTCHI_PDS |
PDS host for the service account (e.g. https://bsky.social). |
SLOPGOTCHI_IDENTIFIER |
Service account handle or DID. |
SLOPGOTCHI_APP_PASSWORD |
App password for the service account (never the main password). |
CRON_SECRET |
Bearer secret required by the cron poll route. |
MAX_BACKFILL_ROUNDS |
Rounds backfilled synchronously per join (default 25). |
CRON_MAX_DIDS |
DIDs processed per cron invocation (default 10). |
Tests run without any of these — unit tests use mocks, and integration tests skip when their required env vars are absent.
Testing #
npm test # vitest run
npm run typecheck # tsc --noEmit
npm run build # next build
Deploy #
vercel install upstash # provisions Redis + injects UPSTASH_* env
vercel --prod
Set the remaining env vars in the Vercel dashboard (or vercel env add).
Cron / "keeps watching" #
vercel.json ships a once-daily cron (Hobby-safe). Per-minute polling needs one
of:
- Vercel Pro: change the
cronsschedule invercel.jsonto* * * * *and raiseexport const maxDurationin the route files to300. - Hobby: point an external scheduler (cron-job.org / QStash) at
GET /api/cron/pollevery ~60s with headerAuthorization: Bearer $CRON_SECRET.
Idempotency on (PR, round) makes repeated/overlapping invocations safe.