fix(workhub): admit quoted Session names and spoken Chinese stops

[ARE404]

Summary

Three independent WorkHub fixes, each self-contained.

Two of them are user-facing bugs where a request was not just missed but acted on as something else.

1. A quoted Session name refused the request

Create a new Session called "Payments" resolved as non-executable, while the same sentence without quotes was admitted. Quoting the name is the natural way to write it, so naming a Session by quoting it did not work at all, in either language.

The intent reader masks quoted spans before the text is read for commands, so a quoted word can never be mistaken for one. The title is the one place quotes mark the object rather than hide a command, and the caller already resolves it from the unmasked source; the imperative check was re-deriving it from masked text, where every quoted title reads as unusable, and refusing on that. It now takes the caller's verdict. The mask still decides everything else — a question, a hedge, a negation and an unterminated quote are all refused as before, which the test pins alongside the four sentences that now work.

2. A spoken Chinese stop was not read as a stop

停掉支付任务 carried no stop cue at all. The Chinese pattern matched only the written verbs 停止 / 取消 / 终止 / 中止, so the most ordinary spoken form fell through to routing and was delivered to Payments as new work: asking to stop it started more of it.

停掉 and 停下 are the spoken forms of 停止. English already covers its own range with stop / cancel / terminate / halt, so this restores the same range rather than claiming a wider one. 关掉 stays out — it reads as "switch off", which is usually work to do inside a Session, and English admits no equivalent. Anaphora now behaves as it does in English: 停掉它 reads the cue and claims no target, so the surface asks which work instead of guessing. The test pins that against Stop it rather than restating the shape, so the two languages cannot drift apart. A cue alone still stops nothing — the reference must resolve to a real Session, which is what keeps 停掉这个定时器 ordinary work.

3. Bridge surface with no callers

workHub.answer and workHub.createSession are declared on the Desktop bridge, implemented in preload, and handled in the main process, and no renderer code calls either. Answering happens through act({ disposition: 'answer_here' }) and Session creation through the create_new disposition, both served by the Action Gate.

This removes the bridge members, their preload implementations, the workhub:answer IPC handler and the Host client method behind it. The workhub.coordination.answer operation stays: it is a Host protocol contract with a live server implementation, and retiring it is a wire change with its own epoch, not a consequence of the renderer being silent.

Verification

Each fix carries a regression that fails on the old behaviour: reverting the production change turns the corresponding test red, verified for both bugs.

Run locally, all green:

• npm run build, npm run typecheck, npm run lint, npm run format:check, npm run check:asf-headers
• @maka/core 801 pass / 0 fail
• Desktop WorkHub controller 72 pass / 0 fail
• Runtime Host WorkHub suites 65 pass / 0 fail

Compatibility

No wire change and no schema change. The removed bridge members had no callers, and the Host protocol surface is untouched.

AI use

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code wrote the changes and tests in this PR under human direction and review.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — two refused requests are now admitted, and one unused bridge surface is removed
• No

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[Astro-Han]

LGTM.