| name | atlas-feature |
| description | Implement a feature end to end when the user says "implement a feature", "add a feature that", or "build new functionality" that spans UI, API, and data. Ships working with verified evidence, not 'should work'. Dispatches the atlas squad in parallel and closes with an independent verifier. |
| when_to_use | the user asks to implement a feature, add a feature, or build new functionality that spans UI, API, and data and must ship working with evidence |
| allowed-tools | Read, Glob, Grep, Bash, Edit, MultiEdit |
| argument-hint | [feature] [acceptance criteria] [stack] [constraints] |
Apply the Operating Contract to this entire task. It is injected below.
cat "${CLAUDE_PLUGIN_ROOT}/references/operating-contract.md"
If the contract did not load above, read ${CLAUDE_PLUGIN_ROOT}/references/operating-contract.md and apply it before proceeding.
atlas-feature
Build the feature described in $ARGUMENTS end to end: backend and frontend, with proof.
Inputs to read from $ARGUMENTS: the feature (what it does, for whom), acceptance criteria (specific and testable), stack (frontend framework, backend framework, datastore, auth), and constraints (auth model, data sensitivity, performance targets, anything regulated). If a required input is missing or ambiguous, ask once for it, then proceed.
Workflow
The full end-to-end workflow (spec -> tests -> impl -> verify, one
failable check per phase) lives in references/feature-workflow.md.
Follow it in order; each phase's check must pass before the next.
Pre-flight
- Read the existing backend layout (routes, services, models, config) and frontend layout (components, pages, hooks, utils) and match it. Do not invent a new structure.
- Look up any unfamiliar API, framework, or SDK via Context7 (or Microsoft Learn for Microsoft services) before using it. No memory-based API calls.
Pick the shape: loop or single pass
- If this work is recurring or iterative (a sweep across many endpoints or screens, a build-fix cycle, an until-dry discovery pass, a migration, or a review round), invoke the
atlas-loop skill to select and instantiate the best-fit loop from the loop-library, then run that loop. Otherwise dispatch the squad directly for a single pass.
Execute through the squad (parallel where independent)
Dispatch all independent jobs in ONE message (multiple Agent calls in a single message) so they run concurrently; keep roughly 4-6 in flight. ALWAYS close the wave with an independent atlas:verifier in a fresh context before integrating results.
- atlas:explorer: map where the new endpoints, models, and UI belong.
- backend-architect or atlas:implementer: add the API surface, services, and data access.
- frontend-developer: build the UI against the new endpoints.
- atlas:verifier: independently confirm each claim.
- atlas:ui-runtime-tester: confirm live UI behavior in the browser.
Build rules
- Backend: validate all inputs; handle the error path; never hardcode secrets (read from env, add keys to .env.example); verify DB writes by reading them back; test the auth path, not just the happy path.
- Frontend: handle loading, empty, error, and success states; inline validation on inputs, not only on submit; mobile-first responsive; one design system (use the project's; a common default is Tailwind plus shadcn/ui).
VERIFY (evidence required)
- Hit each endpoint with curl. Show the exact command and the actual response body, including one error-path call (bad auth or invalid input).
- Read back at least one DB write to prove it persisted.
- Visit each route after hot reload: confirm the console is clean and the route renders. Demonstrate all four UI states.
REPORT
- Endpoints added, each with a sample curl call and its response.
- UI states demonstrated and the route each was shown on.
- Tests run and their output.
- Files changed (paths) and a short diff summary.