(CODUX-183) Execute gated wakeups

[numnx]

🧩 Task: Execute gated wakeups

Automated task PR opened by Code UX for T03.

📋 Summary

Sprint: Sprint 183 · Scheduler Wakeups and Guidance · Task: T03

Extend scheduler wakeups so project-manager chat workers can schedule delayed follow-up messages that fire only after the originating invocation has completed, and update project-manager guidance to use that capability for sensible delayed follow-ups.

🤖 Model & Provider

Provider	Codex CLI
Model	gpt-5.5

⏱️ Timing

Started	2026-07-07 13:43:17 UTC
Finished	2026-07-07 13:49:20 UTC
Duration	6m 3s

📊 Token Usage

Metric	Value
Input tokens	98,613
Cached input tokens	1,486,592
Output tokens	12,626
Total tokens	1,597,831
Invocations	1
Tool calls	49
Active time	4m 53s

Included usage estimate (subscription/local login): $1.62

✅ QA Review

⏳ QA review runs after this PR is opened. Check the task card in the dashboard for the outcome once it lands.

📝 Full Task Prompt

## Objective
Update scheduler execution so wakeup entries post through the chat runtime after their due time, but defer if they are gated on an invocation that has not completed yet.

## Scope
- src/services/scheduler-service.ts
- src/app/dependency-factory/dashboard-factory.ts
- tests/backend/services/scheduler-service.test.ts
- src/repositories/execution-repository.ts only if type imports or public methods are needed for existing `getExecutionInvocation`

## Implementation Requirements
1. Add `executionRepository?: ExecutionRepository` to `SchedulerServiceDeps` and wire `coreDeps.executionRepository` into the `SchedulerService` constructor in `dashboard-factory.ts`.
2. Before `markRunSucceeded` in `runDueEntries`, detect wakeup entries with `wakeupTarget.resumeAfterInvocationCompletion !== false` and a `sourceInvocationId`; if the execution repository reports the invocation is missing or has status other than `completed`, skip execution and leave the entry `scheduled` with its existing due `nextRunAt`.
3. For a source invocation with terminal non-completed status (`failed`, `cancelled`, or `paused` if treated as terminal by existing invocation contracts), mark the wakeup failed with a clear `lastError` instead of posting stale follow-up text.
4. Add wakeup execution to `executeEntry` so it calls `ChatThreadRuntimeService.postMessage` with `bodyMarkdown`, optional thread/title/connection fields, and metadata including `source: "scheduler"`, `schedulerEntryId`, `scheduledFor`, `targetType: "wakeup"`, and `sourceInvocationId` when present.
5. Keep the current in-flight guard, due-entry recheck, recurrence advancement, and one-time completion behavior intact for all target types.
6. Add service tests for immediate ungated wakeups, gated wakeups deferred while the source invocation is running, gated wakeups posting after completion, and failed/cancelled source invocations marking the wakeup failed.

## Constraints
- Do not mark a gated wakeup succeeded before it actually posts to chat.
- Do not busy-loop by creating new entries or mutating `nextRunAt` on every deferred tick.
- Do not block or delay unrelated due scheduler entries when one gated wakeup is waiting.
- Preserve current behavior that successful one-time entries become `completed` and recurring entries advance via `computeNextRunAfterOccurrence`.

## Verification
- `pnpm exec vitest run tests/backend/services/scheduler-service.test.ts`
- `pnpm run lint`
- Success means due wakeups post through the chat runtime only when allowed, deferred wakeups remain scheduled, and existing sprint/quicksprint/chat scheduler tests still pass.

🌿 Branch Info

Base: feature/CODUX-183-scheduler-wakeups-and-guidance
Head: task/feature-codux-183-scheduler-wakeups-and-guidance-t03-codex-mrap9bw4

---

🤖 Generated by Code UX