Claiming **TaskCard.tsx**, the largest app-side cluster in the census.
39 convert; 3 are flagged and left **counted**, one of them a live bug.
## Census
| | before | after |
|---|---:|---:|
| `TaskCard.tsx` | **42** | **3** |
Baseline shrinks by exactly the 39 converted.
## The shape of it
`taskColumnFlags` was **already threaded into this component** and
already consumed by `canEdit` and `isTaskAgentActive` — but the terminal
/ mid-flight / review questions were still answered by comparing
`task.column` to a literal, **39 times in one component**. That is how a
card ends up rendering as live work by one question and terminal by the
next on the same board.
Four booleans now resolve once, beside the existing intake/hold pair and
before the first `useState` that reads them:
```ts
const isWipColumn = isWipColumnRole(taskColumnFlags, task.column);
const isReviewColumn = isReviewColumnRole(taskColumnFlags, task.column);
const isCompleteColumn = isCompleteColumnRole(taskColumnFlags, task.column);
const isArchivedColumn = isArchivedColumnRole(taskColumnFlags, task.column);
```
Flags-first with the legacy id as the documented no-metadata fallback —
identical in shape to the intake/hold pair directly above. No new data
flow, no new abstraction.
## A live bug, flagged rather than converted
The in-review card menu pushes move targets:
```ts
for (const column of ["done", "triage"] as const) {
```
**`triage` is the column #2515/U11 deleted** when it merged intake and
hold into a single `todo` lane. On a post-U11 board this pushes a "Move
to triage" entry for a column that does not exist, and
`taskActionColumnLabel("triage")` labels a target the board cannot show.
Converting it to a role would have been the *worst* outcome — it would
have **hidden the staleness** by resolving the dead target to a live
column. Removing a visible menu entry is exactly the UI-affordance
change AGENTS requires a Surface Enumeration for (the workflow-row
chevron took FN-6115 → FN-6118 → FN-6123 for skipping it), and what it
should offer instead is a product call. Recorded at the site with the
cause.
The other two flagged: `getInReviewCompletionMs` is module-scope with
only a `Task` and no flags to consult (same class as
`project-engine.ts:2555` and `github-tracking-comments.ts:165`), and the
`isHoldColumn` fallback arm, which *is* the degraded answer.
## Revert proof — both directions, because only one is reachable by
renaming
| reverted | result |
|---|---|
| `task.column === "done"` on the archive guard | Archive **appears** on
a mid-flight card: `expect(element).not.toBeInTheDocument()` |
| same | Archive **missing** on a renamed complete lane: `Unable to find
… name "Archive"` |
The pure-rename direction is only half the property. The other half —
traits say mid-flight, column still *named* `done` — is what an
unconditional id comparison actually gets wrong, and it is reachable by
repurposing a default column rather than renaming one.
## Two process findings
**1. `git stash` is shared across worktrees, and a concurrent worker's
stash cost me this cluster once.** I stashed to measure a baseline, and
between my push and my `git stash pop` another agent working in a
different worktree of this repo pushed a stash — so `pop` (which is
positional) applied **their** `self-healing.ts` changes into my tree and
my TaskCard work vanished from it. Their entry was kept rather than
dropped, so nothing was lost; I reverted their application, left
`stash@{0}` untouched, and recovered mine with `git stash apply <sha>`.
**Positional stash refs are unsafe in this repo** — the stack is in the
common git dir, so every worktree shares it. Use an explicit SHA.
**2. Running `--strict` regenerated 26 lines, not 1.** The writer
deliberately omits the derived aggregate blocks (my own earlier change,
to stop every fleet PR conflicting on the same totals lines) but main's
baseline still carries them from an older write — so a regeneration here
would have silently deleted `totals`, `byColumnId` and `properties` as a
side effect of converting one file. I hand-edited the single entry
instead, so the aggregate removal stays owned by the PR that introduced
it. Worth knowing: **`--strict` auto-rewrites and prints "COMMIT IT", so
this rides along invisibly** for anyone who does.
## Verification
`pnpm test:gate` **GREEN** (158 + 10 + 487 + 71) · **684 passed** across
TaskCard / role-invariance / workflow-resolved-columns / ListView /
columnRoles · dashboard `tsc -p tsconfig.app.json` clean · `pnpm lint`
clean · census `--strict` exits 0.
The **3 `TaskCard` failures are pre-existing** — verified twice against
a stashed clean `origin/main`, same three names. They assert CSS-var
geometry (`expected '0' to be 'var(--space-xs) var(--space-sm)'`) and
are untouched by this change.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
e46cc7f1be