Done Session

## /done-session

Run this checklist in order when you believe a session is complete.
Do not skip steps. Do not declare the session done until all pass.

---

### Step 1 — Verify (session tests)

```bash
make verify

If it fails: fix the root cause and re-run. If the same error appears twice, stop and explain. Do not proceed to Step 2 until this exits 0.

---

Step 2 — Regression gate (all tests)

make test

This runs the full test suite — not just this session's tests. If a prior phase breaks, fix it before continuing. A regression here means this session changed something shared (a model, a schema, a shared package) in a breaking way.

If make test does not exist yet, run each service's tests explicitly:

pytest backend/ -v --tb=short -q
pnpm --filter web test --run
pnpm --filter mobile test --run

---

Step 3 — FIXME count

grep -rn "# FIXME" --include="*.py" --include="*.ts" --include="*.tsx" . \
  | grep -v node_modules | grep -v ".git"

Note the count and delta from the start of this session. Flag as risk if delta > 5.

---

Step 4 — Write handoff note

Create .claude/sessions/{phase}/{session-id}-handoff.md using the template at .claude/templates/handoff-template.md. Fill in every section. Write "none" if empty.

---

Step 5 — Commit

Stage only the files this session was responsible for (never git add .):

git add <specific files>
git commit -m "feat(phase-{n}): {description}

Implements SPEC.md § {section reference}.
make verify: passing
make test: passing
FIXME count: {n} (delta: +{delta})"

---

Step 6 — Open PR

gh pr create \
  --title "feat(phase-{n}): {session title}" \
  --body "$(cat <<'EOF'
## What this builds
{one paragraph}

## SPEC section
{exact section name from SPEC.md}

## Verification
\`\`\`
make verify   # exits 0
make test     # exits 0 (full regression)
\`\`\`

## FIXME count
{n} total ({+delta} this session)

## Handoff note
See \`.claude/sessions/{phase}/{session-id}-handoff.md\`
EOF
)"

---

Step 7 — AI code review

/code-review medium --comment

This posts findings as inline PR comments. Read the output. If any finding matches a rule in QUICK-REF.md or the guardrails, fix it before asking the human to review — do not leave guardrail violations for the human to catch.

---

Step 8 — Mark index done

In .claude/sessions/SESSION-INDEX.md, change [ ] to [x] for this session.

---

Step 9 — Report

Tell the human:

• PR URL
• make verify and make test status
• FIXME count + delta
• Any findings from the code review that were fixed
• Any risks or open questions for the next session