The harness is https://fagsem-h2026-harness.scienta.cloud, and the rules are on the printed
rules sheet. Your team slug is in .fagdag/team. In the first minutes:
-
Sign in to the harness with your Scienta Google account.
-
Open your team's pages:
/teams/<team>/claims(where a person claims; agents can't) and/feedback(check results). Copy the team token from/feedbackinto.fagdag/harness-token, which is gitignored. Never commit or share it. -
Push an empty commit (
git commit --allow-empty -m "chore: Test deploy" && git push) and watch the deploy workflow in the Actions tab go green. Your service is atTEAM_URL,https://fagsem-h2026-<team>.scienta.cloud. -
Run the examples against it:
examples/run.sh https://fagsem-h2026-<team>.scienta.cloud. -
Fetch a spec. The slugs are on
/roadmap. The base features have precise specs; the drip features come as feature requests, and you write the spec and the tests. Hurl files you add toexamples/are picked up byrun.sh:H=$(cat .fagdag/harness-url) TOKEN=$(cat .fagdag/harness-token) curl -fsS -H "Authorization: Bearer $TOKEN" "$H/specs/speakers/spec.md"
Add request logging early. The base app logs no requests. The checks' traffic is your best debugging aid, next to the transcripts on
/feedback. Read the logs with.github/fagdag/logs.sh. -
Set up your workflow. Start
REVIEW.mdnow and keep it as a running log.AGENTS.mdhas the facts agents need (contract, commands, gotchas), andCLAUDE.mdimports it: keep that line, and put your own agent instructions in.claude/rules/. The organisers' optional starter kit for Claude Code is linked from/roadmapand/specs.
GitHub Actions. Workflow catalogue: .github/workflows/README.md
(deploy, black-box, logs).
Deploy. Every push to main runs .github/workflows/deploy.yml on a GitHub-hosted runner. It
builds the root Dockerfile with the commit's SHA baked in, pushes it to your registry as
app:<sha> and app:latest, and runs .github/fagdag/rollout.sh. That redeploys your container,
fails unless Scaleway reports it ready, then waits up to 60 s for GET /health to answer 200 with
the new SHA as its version. If that fails, app:latest goes back to the version that was live
before (which keeps serving) and the job fails, with the container's error_message and its logs.
You may edit the workflow and add your own gates; if you break it, you lose your own ticks. The
organisers set its config: the secrets SCW_ACCESS_KEY and SCW_SECRET_KEY (your push key) and the
variables SCW_REGISTRY, SCW_CONTAINER_ID, SCW_ORGANIZATION_ID and TEAM_URL. Don't change them. To roll back by
hand, re-tag a good app:<sha> as app:latest and redeploy, or git revert and push.
Deploy contract (anything else, including the stack, is yours to change):
- a
Dockerfileat the repo root, built forlinux/amd64, which takes--build-arg GIT_SHAand bakes it in; - the service listens on port 8080;
GET /healthreturns 200{"status": "ok", "version": "<GIT_SHA>"}only when the service can reach its database within 5 s, and otherwise 503{"status": "db-unavailable", "version": …};- the database comes from
PGHOST,PGPORT,PGDATABASE,PGUSERandPGPASSWORD: your team's own Postgres database, which keeps its data across deploys; - the filesystem is ephemeral: anything written to disk is lost on redeploy;
- no other external dependencies;
- one instance, which may scale to zero when idle: beyond 80 concurrent requests the platform
queues requests and then answers
503itself, which counts as your failure.
Your project is your sandbox. Your push key can change more than your image, but scaling, size
and the domain are fixed by the organisers, and their next apply reverts any drift. Don't touch
anything outside your own service. The database is yours alone: you may connect to it directly, as
a debugging aid, with the PG* values your container gets. The connection pool is capped at 5
(spring.datasource.hikari.maximum-pool-size); tune it if you like. Two things to know: advisory
locks are "not guaranteed" on this database (use row locks, SELECT … FOR UPDATE), and a session
SET leaks to whoever gets that pooled connection next (set things inside the transaction).
application.properties turns sslmode off for the local Compose Postgres; the Dockerfile sets
PGSSLMODE=require for the container.
Logs. Your container's logs (stdout and stderr, timestamped) are in Scaleway Cockpit. From a
laptop or an agent, .github/fagdag/logs.sh prints them, oldest first, with the URL and read-only
token committed in .fagdag/logs-url and .fagdag/logs-token:
.github/fagdag/logs.sh --tail 200 --since 2026-11-06T13:00:00Z--tail defaults to 200 (max 5000) and --since to the last 24 hours. The token can read your
project's logs and nothing else. To follow them live, use
logcli:
LOKI_ADDR=$(cat .fagdag/logs-url) LOKI_BEARER_TOKEN=$(cat .fagdag/logs-token) \
logcli query --tail '{resource_type="serverless_container"}'In GitHub, run the logs workflow from the Actions tab. After a failed deploy, the deploy job's
last step prints the logs since the job started. A crash on start-up shows up in the logs, an image
pull error only in the error_message, and a build error in the Actions log.
Examples. examples/run.sh [base-url] runs the base specs' published Hurl examples (needs
Hurl). They're the happy path, not a gate. The harness's checks go further.
Black-box tests. Deep HTTP coverage lives in black-box-tests/ (TypeScript + Vitest), separate
from Maven/Testcontainers. Hurl stays the light smoke path; Vitest covers base features plus
co-speakers (status codes, limits, concurrency). Details and commands: black-box-tests/README.md.
docker compose up -d # scripts assume Postgres is already up
cd black-box-tests && npm ci
npm run test:local # build/start jar → vitest → stop (or npm test if the app is already up)Use CFP_BASE_URL if you point at a non-default origin (not bare BASE_URL). CI runs on every PR and on every push to main via
black-box.yml (see workflows README);
mark that check required in branch protection. Agent rule: .claude/rules/black-box-api.md — review
and extend this suite on any API change.
Build and run locally. Docker is needed for both: the tests start Postgres with
Testcontainers, and docker compose up -d starts a local Postgres matching the defaults in
application.properties (localhost:5432, database, user and password cfp). Then ./mvnw package
(Java 11) and java -jar target/cfp.jar, or docker build . to build what gets deployed.
REVIEW.md, workflow/ and .claude/. Keep REVIEW.md as a running log from the start; it
has the showcase brief. Your workflow artifacts go in workflow/ and .claude/.
Backend for the conference call for papers. Speakers register and submit talks, the programme committee votes on them.
- Java 8
- Maven 3.5+
mvn spring-boot:run
The service starts on http://localhost:8081. It needs the local Postgres from docker-compose.yml
(docker-compose up -d); the schema is created on start-up.
| Method | Path | Description |
|---|---|---|
| POST | /speakers |
Register a speaker |
| GET | /speakers/{id} |
Get a speaker |
| POST | /talks |
Submit a talk (max 5 per speaker) |
| GET | /talks/{id} |
Get a talk |
| POST | /talks/{id}/votes |
Vote for a talk (1-3 stars) |
| GET | /talks/top |
Top 10 talks by votes |
Formats: LIGHTNING, PRESENTATION, WORKSHOP, KEYNOTE.
mvn test
Push to master, Jenkins builds and deploys to the test server. Ask Kjetil for access.