**Taking:** `packages/core/src/task-store/task-creation.ts`
| file | guards before | after |
|---|---:|---:|
| `packages/core/src/task-store/task-creation.ts` | **4** | **0** |
(4 remaining pattern matches in that file are inside the new explanatory
comments, not code.)
## What the literals meant, and why they had stopped meaning it
```
resolvedEntryColumn !== "triage" ×2 "this workflow has a MANUAL intake"
task.column === "triage" ×2 "created into the intake column"
```
The first named the **default workflow's intake id** to express *"not
the default workflow"*. Post-U11 the default's intake **is** `todo`, so
the comparison became vacuously true for the default workflow and the
guard stopped separating the two shapes it exists to separate. The real
fact is the intake trait's `autoTriage: false`, which
`resolveWorkflowIntakeFacts` now reads from the IR alongside the intake
column id.
The second was the last-resort clause for a card whose workflow could
not be resolved. `intakeFacts.intake` covers that properly — it falls
back to `DEFAULT_WORKFLOW_ID` rather than to a bare id — so an explicit
`column: "triage"` create on a workflow that still declares `triage`
(R11) is matched through the *resolved* intake instead of a coincidence
of naming.
`isUnplannedStartCreate` is also restated in terms of what it actually
detects — *"the card landed past its workflow's manual intake"*, which
is what quick-add Start does by submitting the workflow id and the
post-intake column together. That replaces `&& task.column === "todo"`,
another id standing in for a relationship.
## Two expectations the conversion legitimately inverted
Both read before changing, neither retargeted blindly.
**1. *"keeps generateSpecifiedPrompt for a direct create into todo (not
bootstrap)"***
`todo` **is** the default's intake now, so a card created there with no
spec **must** get the bootstrap seed — triage admits a card for planning
only when its `PROMPT.md` reads as a seed. Keeping the old expectation
would have pinned the FN-8587 stall: a boilerplate spec that reads as
"already planned" and is never planned.
The old behaviour survived only by **accident of resolution failing** in
the harness, which left the `=== "triage"` literal as the sole deciding
clause. Removing that literal is what surfaced it — which is the point
of the conversion. Split into two tests so the contract it was really
protecting (an explicit **non**-intake column stays a specified create)
keeps its own case.
**2. `store-reservation-atomicity`'s file-scope rollback test**
It stubs `generateSpecifiedPrompt` to inject a bad `## File Scope`, so
it needs a create that actually **calls** that generator. A `todo`
create now gets the bootstrap seed instead, and bootstrap intake prompts
deliberately skip the file-scope hard-fail because their body is
freeform operator prose where a stray `## File Scope` token is not a
real declaration.
Moved to a non-intake column so validation still runs. Left as-is the
test would have been **vacuous** — no throw, no rollback exercised —
while still reporting green.
## Verification
Core package vs the 47-failure post-merge main baseline: **47 failed —
zero new.**
Two files reported failures in the wide run and pass in isolation
(`create-task-reserved-id` 4/4, `schema-applier` 75/75) — the known
contention pattern in this suite; re-run before attributing.
Gate **482 + 10 + 71** green. Lint and core typecheck clean.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
6a33d8f8cc