H-6763: Add temporary Brunch ask shim for Voice preview

[kostandinang]

🌟 What is the purpose of this PR?

Temporary Voice-preview compatibility shim. This does not complete Mission 3 or Mission 4 and does not provide durable elicitation or canvas writes. Migrate or remove it after Mission 4 lands.

Mount the real brunch_ask client tool on Brunch's current ChatAgent so the existing Petrinaut Voice preview and widget can receive real Brunch questions and return answers through the existing client-tool resume path.

This PR only enables real Brunch questions and Voice/widget answers. The independently useful Voice interaction improvements are separated into #9512.

🔗 Related links

• H-6763 (internal)
• H-6763: Submit only completed Voice transcripts as interview answers #9500 — immediate base of this temporary PR
• H-6800: Improve Petrinaut Voice turn-taking and answer provenance #9512 — stacked Voice interaction improvements

🚫 Blocked by

• H-6763: Submit only completed Voice transcripts as interview answers #9500 — this PR is stacked on its branch

🔍 What does this change?

• Adds a Flue brunch_ask client tool using the canonical ASK_TOOL_NAME and AskInput; it returns { awaiting: "client" } with terminate: true.
• Mounts the tool in the real ChatAgent, registers it as client-executed for AI SDK projection, and instructs the agent to ask one concise interview question per turn, wait for the correlated result, treat { answer } as the user's answer, and never claim a canvas modification.
• Reuses the generic client-tool-result signal path in petrinaut-chat.ts; no specialized ask-admission protocol is restored.
• Extends the production-path test through ping, readPetrinautDoc, pending brunch_ask, browser { answer } submission, and continuation of the same persisted Flue conversation.

🏗️ Agent notes

Imperative: Unblock Voice interview testing against the real Brunch endpoint now, without waiting for the durable Mission 4 architecture.

Throughline: Petrinaut AI SDK /api/chat → production ChatAgent → pending brunch_ask client tool → existing client-tool-result signal → the same Flue conversation continues.

Proof: The integration test observes the real agent emit brunch_ask, projects it as input-available without providerExecuted or a server output, submits { answer }, and observes the next turn plus persisted history. Existing ping and readPetrinautDoc checks remain.

Constraints: No useElicitation, elicitor agents, ask-admission protocol, sweeps, folds, completion accounting, capture-store wiring, canvas mutation, or canonical-speech changes. The interviewer tool history is exactly activate_skill, ping, readPetrinautDoc, and brunch_ask.

Fog-line: This shim deliberately does not decide the durable elicitation or canvas-write architecture. Mission 4 owns that boundary.

Stop or reorient: Migrate the shim to the Mission 4 architecture, or remove it in favor of that architecture, after Mission 4 lands.

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

• does not modify any publishable blocks or libraries

📜 Does this require a change to the docs?

The changes in this PR:

• are an internal temporary compatibility path and do not require a docs change

🕸️ Does this require a change to the Turbo Graph?

The changes in this PR:

• add a workspace dependency through package.json; Turborepo derives the edge without a turbo.json change

⚠️ Known issues

This is intentionally not durable elicitation. It does not write captures or the Petrinaut canvas, and it must not survive Mission 4 unchanged.

🐾 Next steps

• Migrate or remove this compatibility shim after Mission 4 lands.
• Preserve the independently useful Voice interaction behavior in H-6800: Improve Petrinaut Voice turn-taking and answer provenance #9512 when this temporary branch is retired.

🛡 What tests cover this?

• yarn workspace @apps/brunch-agent test:unit — 10 files, 42 tests
• yarn workspace @apps/brunch-agent lint:tsc
• yarn workspace @apps/brunch-agent lint:eslint — 0 errors; 8 pre-existing warnings in untouched files
• yarn workspace @apps/brunch-agent build
• Focused formatting and git diff --check

❓ How to test this?

1. Check out this stacked branch and run the Brunch server with the Petrinaut Voice preview.
2. Explicitly request an interview and confirm the assistant emits a real Brunch question.
3. Answer through Voice or the widget and confirm the same conversation continues with the correlated answer.
4. Confirm no Petrinaut canvas mutation occurs.

📹 Demo

Not recorded.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
hash	Ready	Preview	Sep 2, 2026 8:35pm UTC

3 Skipped Deployments

Project	Deployment	Actions	Updated
hashdotdesign-tokens	Ignored	Preview	Sep 2, 2026 8:35pm UTC
petrinaut	Skipped		Sep 2, 2026 8:35pm UTC
petrinaut-docs	Skipped		Sep 2, 2026 8:35pm UTC

[cursor]

PR Summary

Medium Risk
Extends the production chat agent and client-tool transport contract for interviews; behavior is scoped to a temporary shim with no canvas mutation, but it affects live conversation turns and resume semantics.

Overview
Adds a temporary brunch_ask client-executed Flue tool so Petrinaut Voice/widget can run real interview questions through the existing /api/chat client-tool resume path, without Mission 4 elicitation or canvas writes.

The new tool returns { awaiting: "client" } and terminates the turn; ChatAgent mounts it and is instructed to call it on explicit interview requests, ask one question per turn, wait for client-tool-result, and treat { answer } as the user reply. brunch_ask is registered alongside readPetrinautDoc in the client-tool set, and ask calls are projected to the AI SDK as dynamic-tool with dynamic: true in the live stream (other client tools stay tool-*).

Integration and transcript tests now cover pending ask, answer resume, persisted history, and the full probe flow (ping, docs, ask, answer continuation).

^{Reviewed by Cursor Bugbot for commit 252b9db. Bugbot is set up for automated code reviews on this repo. Configure here.}

[kostandinang]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• FE-1576: Ground Brunch interviews in the current Petrinaut net #9523
• H-6800: Improve Petrinaut Voice turn-taking and answer provenance #9512
• H-6763: Add temporary Brunch ask shim for Voice preview #9507 👈 (View in Graphite)
• H-6763: Submit only completed Voice transcripts as interview answers #9500
• FE-1570: Optimize spoken Brunch responses for Voice mode #9496
• H-6763: Integrate Voice mode into the AI assistant #9467
• H-6763: Migrate voice interviews to GPT-Realtime-2 #9399
• H-6763: Add the voice interview experience #9380
• H-6763: Improve voice preview reliability diagnostics #9378
• H-6763: Speak finalized Brunch responses with OpenAI #9357
• H-6763: Add OpenAI Realtime voice input to Brunch #9356
• H-6763: Add a generic Petrinaut composer submission API #9355
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[TimDiekmann]

infra ✅

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 880213f. Configure here.}