(CODUX-220) Settle preparation failures and recovery

[numnx]

🧩 Task: Settle preparation failures and recovery

Automated task PR opened by Code UX for T03.

📋 Summary

Sprint: Sprint 220 · Reduce invocation visibility latency safely · Task: T03

Make CLI task-coding execution invocations server-visible immediately after cancellation registration and before workspace preparation, while preserving truthful provider telemetry, recovery semantics, server-authoritative dashboards, and realtime Stats freshness.

🤖 Model & Provider

Provider	Codex CLI
Model	gpt-5.6-sol

⏱️ Timing

Started	2026-07-13 18:40:20 UTC
Finished	2026-07-13 18:52:25 UTC
Duration	12m 5s

📊 Token Usage

Metric	Value
Input tokens	191,179
Cached input tokens	9,247,232
Output tokens	25,216
Total tokens	9,463,627
Invocations	1
Tool calls	89
Active time	11m 36s

Included usage estimate (subscription/local login): $6.34

✅ 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
Ensure every early-created CLI execution invocation reaches an honest terminal or intentional recovery state without duplicate updates or misleading running rows.

## Scope
- `src/services/cli-workflow-service.ts`, including `runTaskWorkflow`, `updateExecutionState`, and workflow cleanup paths
- `src/services/runtime-recovery/invocation-recovery.ts`, especially interrupted task-coding reconciliation
- `src/services/runtime-startup-recovery-service.ts`, where startup recovery delegates task-coding reconciliation
- `src/services/execution-invocation-control-service.ts`, where cancellation state is closed
- `tests/backend/services/cli-workflow-service.test.ts`
- `tests/backend/services/runtime-startup-recovery-service.test.ts`
- `tests/backend/services/execution-invocation-control-service.test.ts`

## Implementation Requirements
1. Add an idempotent CLI execution-invocation finalizer for successful completion, preparation/routing/agent/memory/workspace/Git/Docker failures, and cancellation; set `finishedAt`, diagnostic fields, and one concise system audit message only while the row is still active.
2. On cancellation after dispatch registration, abort the active workflow, close task-run and dispatch state consistently, mark the execution invocation cancelled once, and prevent late provider or workflow finalizers from overwriting it.
3. Preserve shutdown semantics by leaving intentionally active rows available for startup recovery, while teaching interrupted task-coding reconciliation to distinguish valid pre-provider recovery rows from stale no-provider rows and settle stale rows with useful audit evidence.
4. Make resumed-workspace provider-completion recovery finalize the current workflow truthfully without invoking the provider again, creating duplicate provider usage, or leaving a second execution row running; cover terminal task, sprint, dispatch, and provider evidence.
5. Add focused tests for preparation failure, cancellation during preparation, shutdown interruption, stale pre-provider recovery, recovered provider completion, and idempotent late updates.

## Constraints
- Keep provider claim timing and provider usage finalization owned by `ProviderExecutionService`.
- Do not overwrite rows intentionally preserved for startup recovery.
- Preserve existing task-state, dispatch, workspace-cleanup, retry, and sprint-cancellation behavior.
- Keep documentation changes in the dedicated documentation task.

## Verification
- Run `pnpm exec vitest run tests/backend/services/cli-workflow-service.test.ts tests/backend/services/runtime-startup-recovery-service.test.ts tests/backend/services/execution-invocation-control-service.test.ts`.
- Confirm every non-shutdown early failure has a terminal execution row with `finishedAt` and an audit message, while shutdown rows remain recoverable.
- Run `pnpm run lint`.

🌿 Branch Info

Base: feature/CODUX-220-reduce-invocation-visibility-latency-safely
Head: task/feature-codux-22-t03-codex-959d9fba-mrjkhzjm

---

🤖 Generated by Code UX