A case queue triage simulation for CodeSignal Learn and Hire. The candidate works a queue of intake cases against a written routing guide: for each case they set a category, a priority and an owner, decide whether to escalate, merge duplicates, and write short replies on the cases that need one. It measures how accurately someone applies a policy to realistic, messy requests, and how well they communicate within it.
One app serves many desks. The field labels and options, the guide and the cases all come from
the task's config.json, so the same simulation runs an IT service desk, a claims intake queue,
an eligibility desk or a fraud alert queue.
- Candidate: triages a queue of cases against a routing guide and decides which requesters need a written reply (the examples use 8 and 15 cases).
- Kind: builder; the triaged queue is the artifact. No AI in the app.
- Grading: a judged state report.
npm run reportprints every decision and reply in a fixed order; the platform's grader judges it against criteria the app never sees. - Varies per task: the desk, field labels and options, the guide, the cases (authoring guide).
- Runtime: Node.js only; the server has no runtime dependencies and the client ships prebuilt.
git submodule update --init --recursive
npm install
npm run start:dev # Vite on :3000, API on :3001, with the IT service desk exampleCONFIG_FILE=examples/claims-intake/config.json npm run start:dev loads the other example. Work
saves to .dev-state/solution.json.
As a session runs it:
npm run build
cp examples/it-service-desk/config.json config.json
npm run start:prod # serves dist/ and the API on :3000, reads ./config.json
npm run report # prints the report from ./config.json and ./solution.json| Command | What it does |
|---|---|
npm test |
headless tests: config validator, work rules, report, server API |
npm run validate -- <config.json>... |
checks task configs |
npm run report [-- --dir <dir>] |
prints the session report |
npm run build |
builds the client into dist/ |
npm run release:local |
builds, packs release.tar.gz and boots it as a session would |
A session downloads release.tar.gz and runs npm run start:prod where it unpacks; it never
installs dependencies. The tarball holds one minified bundle per entry point (server.js,
scripts/report.js, scripts/validate.js), the built client and the docs, and never a
config.json, solution.json or example solution.
Every pull request runs the tests, packs the tarball and boots it. A merge to main also
publishes a pre-release vN and moves the floating prerelease release to it:
| Channel | URL |
|---|---|
| Staging | https://github.com/CodeSignal/learn_cosmo-desk/releases/download/prerelease/release.tar.gz |
| Production | https://github.com/CodeSignal/learn_cosmo-desk/releases/latest/download/release.tar.gz |
Promote a build to production once it has passed a staging test:
gh release edit vN -R CodeSignal/learn_cosmo-desk --prerelease=false --latestserver.js HTTP server: allowlisted config, per-case saves, reset, static dist/
shared/ config format and validator, work rules (used by server and client)
lib/ file loading and saving, the session report
scripts/ validate and report entry points
client/ React app: App, components/, lib/, theme/; design system submodule
examples/ two complete tasks with strong and weak solutions (test fixtures only)
.codesignal/ the base task's session scripts (setup.sh, run_solution.sh), mirrored here
docs/authoring.md writing a task
test/ node:test, no browser