**Taking: `engine/triage.ts`, `engine/pr-comment-handler.ts`,
`engine/eval-followups.ts`, `cli/commands/task.ts`, `cli/extension.ts`**
(write class — no collision with the comparison backlog).
## A class the census does not count
The 48-guard work list tracks `=== "triage"` **comparisons**. These are
`column: "triage"` **writes** — and post-#2515 every one creates a card
directly into the state STALL 3 was about, except **manufactured
continuously** rather than left behind by the upgrade.
## Why they bite
`createTaskImpl` resolves the column as:
```ts
column: input.column || options?.resolvedEntryColumn || fallbackIntakeColumn || "triage"
```
`input.column` **wins**, so an explicit `column: "triage"` overrides the
workflow's resolved intake column entirely.
`store-create-intake-column.test.ts` already pins that a create with
**no** column lands in the default workflow's intake (now `todo`) —
these callers opted out of it.
The sharpest is `triage.ts`'s `fn_task_create` agent tool: it passed
`workflowId: params.workflow_id` **and** `column: "triage"` in the same
call. The caller chose a workflow and the column ignored it — a Coding
(Ideas) create landed in `triage` instead of `ideas`.
## Counts
**Comparison guards: unchanged by this PR.** This is the write class;
conflating the two would misreport convergence toward the zero bar.
| file | `column: "triage"` writes before | after |
|---|---:|---:|
| `packages/engine/src/triage.ts` | 1 | **0** |
| `packages/engine/src/pr-comment-handler.ts` | 1 | **0** |
| `packages/engine/src/eval-followups.ts` | 1 | **0** |
| `packages/cli/src/commands/task.ts` | 3 | **0** |
| `packages/cli/src/extension.ts` | 3 | **0** |
| **total** | **9** | **0** |
## A test that pinned the defect
`pr-comment-handler.test.ts` asserted `column: "triage"` in the
createTask call — so it would have **failed the fix and passed the
bug**. Rewritten to assert the invariant (the caller passes no column,
so the workflow's intake wins) plus an explicit `Object.hasOwn(arg,
"column") === false`, which is what actually catches a reintroduction.
## Interaction with #2591
My merged #2591 rescues these cards once created — they sit on a legacy
planner id their workflow doesn't declare and are still in planning
stage. So this isn't a *visible* stall today; the rescue absorbs it.
**That's the reason to fix it rather than leave it:** a self-healing
path silently absorbing a steady stream of malformed creates is exactly
how the underlying defect stays invisible.
## Deliberately not touched
- `{ id: "start", kind: "start", column: "triage" }` in the builtin
coding / PR / lead-generation IRs — workflow-internal **node
declarations** for workflows that still legitimately declare a `triage`
column, not lifecycle writes.
- Left for their owners: `core/task-store/project-store-ops.ts:210`,
`core/task-store/update-task-deps.ts:111` (main worker),
`dashboard/src/routes/register-gitlab.ts:108` (u12). Same defect, same
one-line shape.
## Verification
- 304 engine/CLI tests green across the affected suites
- merge gate green (482 + 132 + 10), engine + CLI tsc clean, lint clean
No changeset: `@fusion/engine` and `@fusion/core` are private; the CLI
change is a bug fix with no user-facing API change — happy to add one if
you'd rather it appear in release notes.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
d5f1ce7abd