Slice

# /slice — Implement one task slice end-to-end, then stop

Delivers one atomic unit of work from `NEXT.md`: context map → failing tests → implementation → green → reviewer approval → commit. Exactly one slice per invocation, then summarize and stop.

---

## Step 1 — Gather context with the `explorer` agent

Spawn the `explorer` sub-agent to map the files relevant to this slice:
- Existing domain types, use cases, ports, and value objects the slice touches
- Adjacent tests that show the current testing patterns for this area
- Controller / adapter entry points that will need wiring

Use explorer's path-and-summary output to decide what to read in full. Do **not** read entire files yourself before knowing which ones matter.

---

## Step 2 — Write failing tests (Red)

From the slice's acceptance criteria in `NEXT.md`:

1. **Domain unit tests** (`kindred-domain/src/test/…`) for any new use-case logic.
2. **Integration tests** (`kindred-app/src/test/…`) for any new endpoint or persistence change, using Testcontainers against real PostGIS.
3. Run the tests. Confirm they **fail for the right reason** — a missing class or method, not a compile error (unless the task requires adding a new type).

Do **not** write production code yet.

---

## Step 3 — Implement to green (Green)

Write the minimum production code required to make the failing tests pass.

- **Domain layer first** (`kindred-domain`) — pure Java, zero framework imports.
- **App layer second** (`kindred-app`) — Spring wiring, JPA adapters, controller changes.
- Follow all `CLAUDE.md` conventions:
  - Java records for DTOs; sealed interfaces for domain errors
  - Constructor injection; inject `Clock`, never call `Instant.now()` inline
  - No Lombok; no speculative abstractions
- Run the affected test suite and confirm all targeted tests are green.

---

## Step 4 — Run full affected test suite

```bash
./gradlew :kindred-domain:test :kindred-app:test

All tests must be green — no regressions.

---

Step 5 — Reviewer pass

Spawn the reviewer sub-agent on the current diff.

• APPROVE → proceed to Step 6.
• CHANGES → fix every listed finding, re-run tests (Step 4), then spawn reviewer again. Repeat until APPROVE.

Note: the verify hook runs automatically on every Edit/Write and catches compile errors early. A CHANGES verdict from the reviewer catches convention violations that compile cleanly.

---

Step 6 — Update NEXT.md

• Move the completed item from "In Progress" (or "Up Next") to the Done section.
• One-line summary: what was built and the final test count (X domain + Y app).
• If a new concern surfaced (follow-up task, deferred decision, discovered debt), add it to Parking Lot.

---

Step 7 — Commit

Stage all changed files except .class files and anything under bin/ or build/ directories.

Commit message format:

T<N> — <Slice title>: <one-line summary>

Tests: <X domain + Y app>, all green.

---

Stop

Output a 3–5 line summary: what was built, final test counts, any notable decisions. Then stop. Do not pull the next item from NEXT.md.

---

Notes on future surfaces

When the React Native (kindred-mobile) or Next.js (kindred-web) modules are scaffolded, extend the verify hook (.claude/hooks/verify.sh) with:

# TS equivalent — add once RN/Next is scaffolded
if echo "$INPUT" | grep -qi "kindred-mobile\|kindred-web"; then
  npx prettier --write "$FILE_PATH" -q
  npx tsc --noEmit -p "$TS_PROJECT_ROOT/tsconfig.json"
fi