Sidebar work card: task brief never renders, and two locked visual rules are missing

[ElbertePlinio]

Goal

Close the remaining gaps between the shipped flat work card and the locked merge design, so a live card actually reads like the approved mockup.

Why

flatChatList is on locally and the card renders — but in real use it collapses to mark · project · time / title · WORKING / $1.8884. The middle of the card (the part that says what the agent is doing) is empty, and two locked visual rules never landed. This is not staleness: verified against main at 9b852be with a build from the same tree.

Reference: ~/Projects/Boards/pickforge/sidebar-work-card-merge.html (approved 2026-07-24, same board as #306).

Gaps

1. Task brief never renders in production (primary)

The renderer exists — ProjectsPane.tsx:1540 <Show when={brief()}> fed by cardBrief (src/stores/flatWorkCard.ts:116-119, reads chat.taskBriefText) — but taskBriefText has no writer outside fixtures. src/stores/workspace.ts:361 always creates chats with taskBriefText: null; the only non-null values are VRT mocks (src/lib/tauriMock.ts:424-425). The comment at flatWorkCard.ts:113-115 already admits it ("always null today").

Decided (2026-07-25): active plan step, falling back to a model-written one-liner.

• Primary — active plan step. Needs no new plumbing: PlanItemStatus already carries inProgress (src/lib/agentChat.ts:29), and latestPlanForChat (src/stores/agentChat.ts:238) is the same accessor cardPlanProgress (flatWorkCard.ts:155-163) already reads for plan M/N. The brief is the inProgress item's text. This also keeps the brief honest against the footer — the card says step 2 of 5 and what step 2 is, from one source.
• Fallback — model-written one-liner, only when there is no plan (no plan timeline item) or no inProgress item. Shares plumbing with Dynamic semantic chat titles that follow the active task #210 (dynamic semantic chat titles); build it there or extract a shared producer, do not write a second one.
• Neither available → no brief line at all. No placeholder, per the locked footer principle.

Sub-decisions for PR 2:

• Truncation: the line is already single-line-ellipsis in CSS (workbench.css:3427-3437); decide whether to also cap in the store to keep persisted values small.
• Which wins if a plan exists and a model one-liner is cached — plan step, per the ordering above; make it explicit in the resolver so it does not drift.
• All-complete plan (M === N, no inProgress item): falls through to the one-liner, then to nothing. Confirm that reads right on a just-finished card before it collapses.

2. NEEDS YOU label has no bracket glyphs

Locked rule (#306): "bracketed text only on the NEEDS YOU label". Code only recolors it — .pf-work-card-status--needsyou { color: var(--pf-warning) } (workbench.css:3423-3425), JSX at ProjectsPane.tsx:1536-1538. The four card corners are implemented (WorkCardCorners, ProjectsPane.tsx:1400-1407), the label brackets are not. The VRT at tests/vrt/flat-chat-list.spec.ts:178-182 asserts the class name, so it passes while the glyphs are absent — the test needs tightening as part of the fix. The composer chips already use the bracket idiom, so there is a house treatment to match.

3. Cost is 4-decimal in the card footer

formatCost (src/components/chat/ContextMeter.tsx:16-18) returns $1.8884; the board shows $0.41. Fine for the composer readout, too noisy at card scale. Wants a card-scale format (2 decimals, or sub-cent → <$0.01) without changing the composer.

4. Branch chip is not visually distinct

.pf-work-card-branch (workbench.css:3452-3459) sets truncation only and inherits --pf-text-low; the board renders the branch in an accent tone so it reads as an identifier next to plan M/N. Token-level change, no new chrome — see the readouts-need-an-edge principle.

Non-gaps (verified, do not "fix")

• Branch + plan M/N slots exist and work (WorkCardFooter, ProjectsPane.tsx:1414-1458); they are absent on the screenshotted chat because it has no worktree branch and no plan. Conditional-on-real-data is the locked footer rule.
• The ▮▮▮ 1/3 glyph is swarm lane ticks, not a context bar; correctly absent on non-swarm chats.
• 2px context edge, state tint, needs-you-above-working sort, QUIET · N — all implemented and matching.
• Two-line quiet rows are a recorded owner deviation (ProjectsPane.tsx:1309-1316), not a regression.

Flag

Rides the existing flatChatList flag (default off, src/stores/flags.ts:57). No new flag. The enable-in-release / remove-flag lifecycle stays tracked on #306 — this issue must land before that flag is enabled, since it is what the enabled state will look like.

PR plan

• PR 1 — visual gaps: NEEDS YOU label brackets, card-scale cost format, branch chip tone
  • Depends on: none
  • Touches: ProjectsPane.tsx, workbench.css, a card-scale cost helper, tests/vrt/flat-chat-list.spec.ts
  • Validation: bun run test:unit; VRT flat-chat-list (tighten the bracket assertion to the glyphs, not the class)
  • Link: pending
• PR 2 — brief from the active plan step
  • Depends on: none (source decided; uses existing plan plumbing)
  • Touches: flatWorkCard.ts (cardBrief resolves from latestPlanForChat instead of taskBriefText), ProjectsPane.tsx wiring, tauriMock.ts fixtures
  • Validation: unit tests — inProgress step wins; no plan → no brief; all-complete plan → no brief; VRT calm/normal/heavy with a real brief
  • Link: pending
• PR 3 — model-written one-liner fallback
  • Depends on: PR 2; coordinate with Dynamic semantic chat titles that follow the active task #210 so one producer serves both
  • Touches: shared producer + cardBrief resolver order
  • Validation: unit test for resolver precedence (plan step > one-liner > nothing)
  • Link: pending

Traceability

Acceptance criterion	PR/slice	Validation	Status
NEEDS YOU label shows bracket glyphs, working label does not	PR 1	VRT asserts glyphs, not class	pending
Card cost reads at 2 decimals; composer readout unchanged	PR 1	unit test on the card formatter	pending
Branch chip is tonally distinct from plan M/N	PR 1	VRT	pending
A chat with a plan shows its active step as the brief	PR 2	unit test + VRT	pending
Brief is absent (no placeholder) when there is nothing to say	PR 2	unit render test	pending
A chat with no plan falls back to a model-written one-liner	PR 3	resolver precedence unit test	pending
Brief and plan M/N never disagree — both read one source	PR 2	unit test	pending

Open decisions

None blocking. All three PRs can start; PR 3 should be sequenced with #210 so the one-liner producer is written once.

Refs #306, #210

[ElbertePlinio]

Brief source decided: active plan step, falling back to a model-written one-liner (then nothing).

Primary source needs no new plumbing — PlanItemStatus already carries inProgress (src/lib/agentChat.ts:29) and latestPlanForChat (src/stores/agentChat.ts:238) is the accessor cardPlanProgress already uses for plan M/N. So the brief and the footer tally read one source and cannot disagree.

Body updated: PR 2 is now the plan-step brief (unblocked), PR 3 the one-liner fallback (sequence with #210). No blocking decisions left.

[ElbertePlinio]

PR 1 is up as #384 — the three visual gaps: NEEDS YOU bracket glyphs (VRT now asserts the glyphs, not the class, which is why this gap survived), a card-scale formatCardCost kept separate from the composer's 4dp readout, and the branch chip one step up the text ramp.

PR 2 (brief from the active plan step) is unchanged and still unblocked. Two notes for whoever picks it up: PR 1's precedent argues for capping the brief in the store rather than relying on CSS ellipsis alone — a persisted value that is always truncated on screen has no reason to be long. And the all-complete-plan case (M === N, no inProgress) is worth checking against a just-finished card specifically, since #375 changed how those behave during their linger.