# Session log

This file tracks work completed across Claude Code sessions. Every agent MUST
read this file before starting work and update it after completing work.

> Earlier sessions (2026-03-24 through 2026-03-26d) archived in
> `docs/memory/archive/sessions-through-2026-03-26.md`

## Format

Each entry follows this structure:
```
### Session [date] — [brief description]
**Agent(s):** [which agents were active]
**Work completed:**
- [bullet points of what was done]
**Decisions made:**
- [any design/architecture decisions with brief rationale]
**Open questions:**
- [anything unresolved that needs human input]
**Next steps:**
- [what should happen next]
```

---

## Sessions

### Session 2026-03-27a — Workflow improvement (5-phase infrastructure upgrade)

**Agent(s):** Claude Opus 4.6 (1M context)

**Work completed:**
- **Phase 1A:** Archived session log — moved 25 old sessions to `docs/memory/archive/sessions-through-2026-03-26.md`, trimmed active log from 1096 to 91 lines
- **Phase 1B:** Documented token access convention as D031 — `theme.palette.*` for semantic tokens, `var(--fa-*)` for component tokens. Fixed Badge.tsx colour map inconsistency (unified to CSS vars). Updated component-conventions.md.
- **Phase 2:** Added ESLint v9 (flat config) + Prettier. Installed `eslint`, `@eslint/js`, `typescript-eslint`, `eslint-plugin-react`, `eslint-plugin-react-hooks`, `eslint-plugin-jsx-a11y`, `eslint-config-prettier`, `prettier`. Ran initial format pass across all 54 .tsx files. Fixed 5 empty interface warnings (Switch, Radio, Divider, IconButton, Link). Added `lint`, `lint:fix`, `format`, `format:check` scripts.
- **Phase 3:** Created 7 new impeccable-adapted skills: `/polish`, `/harden`, `/normalize`, `/clarify`, `/typeset`, `/quieter`, `/adapt`. Downloaded Vercel reference docs (web-design-guidelines, react-best-practices). Updated `/audit` and `/review-component` with optional Vercel references. Total skills: 19.
- **Phase 4:** Added Husky v9 + lint-staged for pre-commit automation. ESLint + Prettier auto-run on staged files. Updated `/preflight` skill with ESLint and Prettier checks (now 8 checks total).
- **Phase 5:** Added Vitest v4 + jsdom + @testing-library/react. Created `vitest.config.ts` and test setup. Created `/write-tests` skill for test generation guidance. Added `test` and `test:watch` scripts. Note: `@storybook/test-runner` deferred — requires Storybook 10+ (we have 8).

**Decisions made:**
- D031: Token access convention — theme accessors for semantic, CSS vars for component (see decisions-log)
- ESLint story files exempt from `react-hooks/rules-of-hooks` and `no-console` (Storybook render pattern)
- Empty `interface extends` changed to `type =` for ESLint compliance (5 wrapper atoms)
- Storybook test-runner deferred until Storybook upgrade to v10
- Prettier printWidth set to 100 (matching observed code style)

**Also completed:**
- Created `docs/reference/component-lifecycle.md` — 10-stage quality gate sequence (build → QA → polish → present → iterate → normalize → preflight → commit)
- Created `docs/reference/retroactive-review-plan.md` — plan to review 30+ existing components (~3.5 sessions)
- Updated `/build-atom`, `/build-molecule`, `/build-organism` to include internal QA stages automatically
- Added lifecycle reference as CLAUDE.md critical rule #8

**Open questions:**
- Storybook upgrade to v10 — would unlock `@storybook/test-runner` for CI test execution
- FuneralFinder version decision (v1/v2/v3) — needed before retroactive review of organisms
- Review depth — P0/P1 only (faster) or include P2?

**Next steps (pick up in next session):**
- Start retroactive review: `/normalize atoms` → `/audit` high-priority atoms
- Interleave with new component work if preferred
- Use `/write-tests` on interactive components as time permits

---

### Session 2026-03-26f — FuneralFinder v3 build + polish

**Agent(s):** Claude Opus 4.6 (1M context)

**Work completed:**
- Created FuneralFinderV3 — clean vertical form based on user's Figma layout (5919:29445), restyled to FA design system
- Two side-by-side StatusCards (Immediate Need default-selected / Pre-planning), stack on mobile
- Standard card container (surface.raised, card shadow) — initially built with glassmorphism, user redirected to standard palette
- Overline section labels (text.secondary) for "How Can We Help", "Funeral Type", "Location"
- Standard outlined fields (white bg, neutral border, brand border on focus, no focus ring per user request)
- Location input with LocationOnOutlined pin icon
- CTA "Find Funeral Directors →" always active — validates on click, scrolls to first missing field
- Dividers after header and before CTA for visual rhythm
- Funeral type options: same as V2 + "Show all options"
- WAI-ARIA roving tabindex on radiogroup, aria-labelledby via React.useId()
- Semantic tokens throughout (border-brand, surface-warm, text-brand, interactive-focus, text-disabled)
- Error messages conditionally rendered in aria-live regions (brand copper tone — gentle validation)
- First pass scored Critique 33/40, Audit 13/20 → iterated on all findings → re-audit 18/20
- Polish pass: revised copy, fixed spacing, defaults, label rename

**Decisions made:**
- Status cards replace V2's step-circle + dropdown — simpler, more visual, side-by-side on desktop
- Standard design system styling (user clarified Figma was for structure only, not colour scheme)
- "Immediate Need" selected by default — most common use case, status error essentially unreachable
- Section renamed from "Current Status" (programmatic) to "How Can We Help" (warm, human)
- Copy: "A recent loss or one expected soon" — "expected soon" differentiates from pre-planning gently
- Copy: "Planning ahead for yourself or a loved one" — "planning ahead" reinforces non-urgency
- No sequential unlock — all fields accessible immediately
- CTA always active — validates on click, scrolls to missing field
- Form simplified to 3 fields (status, funeral type, location) vs V2's 4
- Focus ring suppressed on Select/Input per user requirement — status cards retain focus-visible
- Error colour uses text.brand (copper) not feedback.error (red) — intentional for funeral context

**Open questions:**
- Location autocomplete integration still pending across all versions
- Decision: V1 vs V2 vs V3 for production

**Next steps:**
- User to review final V3 in Storybook
- If V3 chosen: location autocomplete, possible further refinements

---

### Session 2026-03-26e — FuneralFinder v2 polish + consistency fixes

**Agent(s):** Claude Opus 4.6 (1M context)

**Work completed:**
- Fixed location input styling to match select fields (border color, hover, disabled, error states)
- Added `active` prop to StepCircle — step 1 uses brand-500 primary fill by default
- Aligned Input with full selectSx overrides (bgcolor, disabled opacity+dashed, error border)
- Investigated systemic Input vs Select visual differences — confirmed issue is component-local overrides, not theme

**Decisions made:**
- Custom field styling kept within FuneralFinderV2 (intentional divergence from theme for this component's design)
- Step 1 circle uses primary fill since it's always active — visual "start here" signal

**Next steps:**
- v2 is feature-complete and polished — ready for user review in Storybook
- Decision pending: v1 vs v2 for production
- If v2 chosen: add location autocomplete, write flow logic reference doc

---