fix: dispatch plan chips on double-click and send now

[kyleseaman]

Why no screenshot: Dispatch-only change — plan chips look the same; double-click and Send now now hit the plan endpoint instead of chatting the label.

Problem / Motivation

On an orchestrator plan footer, a double-click or a click on the visible Send now arrow sent the chip label (Go, Go All, Cancel) as ordinary chat instead of POSTing /api/chat/slots/{slot}/plan-action.

Cancel is the sharp edge: a typed Cancel is not special-cased server-side, so the user pressed the stop control and the plan kept running.

Why it matters

Single-click already dispatched correctly (#6040 / #5893). The two remaining send gestures skipped that gate, so the discoverable Send-now control and a habitual double-click both bypassed the plan endpoint and its per-slot single-flight latches.

What changed (motivation → approach → change)

FollowUpBar's double-click and Send-now segment call onSend, not onSelect. Hosts wired onSend straight to send() / doSend().

onSend now carries the same click-time row identity onSelect already receives, so a footer that replaces the reused chip between the two clicks of a double-click is refused instead of approving the next stage. ChatPage and ChatPane each have one dispatchPlanFollowUp helper used by every gesture. SideChat and ChatEmbed still drop followUpIsPlan on purpose (#6057 / #6754).

The autopilot spec names all three gestures and the first-click row snapshot.

Tests

• FollowUpBar: double-click after a mid-gesture footer replacement hands onSend the first-click key; Send-now hands the current key.
• ChatPane: double-click on Go and Send now on Cancel dispatch planAction; a straddled double-click after a replacement footer does not dispatch; Send now on a non-plan chip still sends the label.
• ChatPage: double-click on Go and Send now on Go All dispatch planAction.

npx vitest run src/test/FollowUpBar.test.tsx src/test/ChatPane.followUpOptions.test.tsx src/test/ChatPage.followUpToggle.test.tsx — 90 passed.

Manual verification

N/A — unit coverage drives the real ChatPage/ChatPane handlers through the real FollowUpBar gestures, including the replaced-footer race. No visual change.

Related Issues

Fixes #6240

Checklist

• At most two commits (one is the norm), with a Conventional Commits title
• Existing tests pass and new tests added for new functionality
• Self-review completed; code follows project style guidelines
• Documentation updated (autopilot frontend-rendering section)
• No secrets, credentials, or internal references in the diff

[github-actions]

UX Review (Fable 5) — ✅ PASS

UX-level review of 4856936b45c3c525048e5cacf3acb0209ecc8be6 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

The diff is behavior-only: it routes double-click and Send-now on plan chips (Go/Go All/Cancel) through the same plan-action dispatch single-click already uses, with a first-click row snapshot to refuse straddled gestures. No new user-facing strings, no visual change, no screenshots. The one UX gap on the refusal path — a refused/latched chip renders no feedback — is pre-existing, explicitly documented in usePlanActionMutation.ts as tracked (#6056), and deliberately deferred to the shared FollowUpBar; the alternative this PR prevents (silently approving a stage the user never saw, or a "Cancel" typed as chat while the plan keeps running) is strictly worse.

UX-Verdict: PASS

The visible Send-now arrow and double-click now do what the chip label promises — notably, Cancel actually stops the plan instead of chatting "Cancel".

[UX-REVIEWED] 4856936

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of 4856936b45c3c525048e5cacf3acb0209ecc8be6 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Root-cause fix at the right layer: all three chip gestures now converge on one per-host dispatch gate, with the straddled-footer race covered and spec updated in-commit.

[DESIGN-REVIEWED] 4856936

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 4856936b45c3c525048e5cacf3acb0209ecc8be6 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 4856936

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 4856936b45c3c525048e5cacf3acb0209ecc8be6: <one-sentence reason>

[github-actions]

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of 4856936b45c3c525048e5cacf3acb0209ecc8be6 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All evidence gathered. The supplied patch is stale-base drift plus the real change; the branch's own diff (verified via git merge-base = 0552cf61d, two commits da7ec055d + 4856936b4) is exactly the 8-file plan-chip fix the description declares. Sibling check: onFollowUpSend has exactly 2 hosts (ChatPage.tsx:7981, ChatPane.tsx:843), both fixed; SideChat/ChatEmbed drop plan chips by documented design (#6057/#6754, pinned by SideChat.planExclusion.test.tsx / ChatEmbed.test.tsx), so zero unfixed siblings.

First-Principles-Verdict: PASS

All three send gestures now route through the one plan gate single-click already used — cause-level, no riders, and the diff deletes more inline duplication than it adds.

What this change ships

Intent: make double-click and the Send-now arrow on a plan chip actually drive the plan (so Cancel stops it) instead of chatting the label — a FIX.

1. Double-click on a plan chip dispatches the plan action, not chat text — justified (reported defect Plan chips: double-click and the Send-now segment bypass plan dispatch and send the label as chat #6240; typed Cancel never stops a plan)
2. Send-now arrow on a plan chip dispatches the plan action — justified (same defect, the discoverable gesture)
3. A double-click straddling a replaced footer is refused — justified (extends the exact race guard onSelect already carries)
4. onSend/onFollowUpSend gain a click-time row key — declared; 2 counted consumers (ChatPage.tsx:7981, ChatPane.tsx:843)
5. Per-host inline plan gates collapsed into one dispatchPlanFollowUp each — justified subtraction (replaces two duplicated inline blocks per host)
6. autopilot.md updated in the same commit — derived (AGENTS.md spec-management rule)

Watch

• The review artifact, not the author: authentic.patch was cut against a stale base and carries ~1,800 lines of already-merged main work (ci(skills): gate this repository's markers out of shipped skill bodies #7069 skill-scope gate, fix(config): key the config_dir memo on the home it resolved #7092 config-dir memo, fix(test): stop kill_and_reap's group-kill skip from reading host process state #6955 kill_and_reap pin, apple_speech --fast, website dead-code deletions). None of it is in this branch (verified: git diff 0552cf61d..4856936b4 touches only the 8 declared files). Other automated lanes fed the same patch may report findings against code this PR does not ship.
• dispatchPlanFollowUp now has two spellings (ChatPage.tsx:5241, ChatPane.tsx:179) differing only in the slot-record-unknown branch; the host split predates this PR and merging it is genuinely larger — accepted-and-deferred, not a demand.

[FIRST-PRINCIPLES-REVIEWED] 4856936

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed 4856936b45c3c525048e5cacf3acb0209ecc8be6 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 4856936

Verdict parsed from the review's SHA-scoped output markers for commit 4856936b45c3c525048e5cacf3acb0209ecc8be6.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 4856936b45c3c525048e5cacf3acb0209ecc8be6: <one-sentence reason>