workflow-feature

name	workflow-feature
description	Execute one backlog User Story end-to-end on an existing harness project through the gated phase chain (BA delta, design, plan, code, review, test, verify, ship). Use when: executing a single backlog US, workflow-intake routed case 2/3, a per-US iteration inside workflow-bootstrap Stage B, or adding one scoped capability to an existing harness project.

Target US: $ARGUMENTS

One-Liner

One US, one vertical slice: carry it through BA→design→plan→code→review→test→verify→ship with a deltas-only discipline (touch only what this US changes) and passing only after review + required tests are recorded in evidence.

Execute a single backlog User Story to passing. Reached from workflow-intake case 2/3 or workflow-bootstrap per-US execution.

./harness plan <id>      # scaffold + fill plan.md
./harness start <id>     # (team: --assignee <name> --branch)
# … BA delta → design delta → plan-tasks (freeze test-strategy) → code → review → test …
./harness verify <id>    # gated by quality + review hooks → passing

Loop (each step ends in a Gate)

1. Claim & plan — ./harness plan <id>, fill docs/design-docs/<id>/plan.md. In team, ./harness start <id> --assignee <name> --branch (see workflow-team); else ./harness start <id>. Gate: plan.md placeholder is replaced and the US is in_progress.
2. BA delta (plan-ba-analysis) — only what this US changes; persona sets depth.
3. Design delta (design-*) — touch only affected architecture/schema/API/UI. Gate: the delta to requirements/design is captured and scoped to this US only.
4. Plan tasks (plan-tasks) — split into F<id>-T<n>, and freeze the test-strategy (check-test-strategy) into each task's immutable --verifications before start. Gate: child-tasks exist with frozen --verifications (immutable after start).
5. Code (dev-*) — implement per ../../resources/conventions/.
6. Review — drive check-review-loop: spawn an independent reviewer whose role is resolved from the design (scripts/role-resolver.sh --phase review → e.g. a Go Backend Tech Lead), bracket it with harness review open/record, and run the capped fix → re-review loop until no - [ ] is open; check-refactor for cleanup. Escalate to check-security-review (--phase security) when the change touches auth/data/external surface. Gate: no open - [ ] in ## Review and review provenance is CURRENT (review-provenance-gate).
7. Test — run the selected types (test-unit … test-security as the strategy requires); record to evidence ## Test. Gate: every test type the frozen strategy selected is green and recorded in ## Test.
8. Verify — ./harness verify <id> (gated by quality + review hooks). Gate: ./harness verify <id> exits 0 and the US is passing.
9. Ship (ship-commit-msg → ship-pr-create) — deploy only via workflow deploy step (always-stop). Gate: commit/PR is created; no deploy without explicit user confirmation.

Cross-cutting

• Step-gate between phases (../../resources/step-gate.md); in auto, log the decision.
• Persona (../../resources/persona-mode.md) — non-tech gets plain-language progress + auto-advance; dev gets exhaustive questions + per-phase gate.
• Scope discipline (WIP=1) — stay inside this US; cross-cutting fixes become a separate feature.
• Caveman (on by default) — apply opt-caveman to every inter-agent prompt and subagent result (reviewer/tester fan-out especially), artifacts to files (../../resources/token-budget.md).
• No silent completion — passing requires review + required tests in evidence.md.
• Task-state — expand .harness/tasks/<id>.md after start; tick boxes + update Next step at each phase boundary and commit (../../resources/task-state.md).