# Phase 9: Extended Scenarios - Context

**Gathered:** 2026-01-31
**Status:** Ready for planning

<vision>
## How This Should Work

This phase proves the system handles real-world complexity — not just happy paths, but the messy scenarios that happen when multiple agents work in parallel, things go wrong, and the system needs to recover.

Four core scenarios to validate:

1. **Conflict hand-back** — When a merge fails, the agent receives the conflict and resolves it. The full round-trip works.

2. **Multi-agent parallel work** — Multiple agents dispatched simultaneously, each working in their worktree, merging in correct dependency order.

3. **Recovery/resume after interruption** — System can pick up where it left off after crash or interruption.

4. **Agent question/answer resume** — When an agent asks a question (AskUserQuestion), the system pauses, receives the answer, and the agent continues from where it stopped.

</vision>

<essential>
## What Must Be Nailed

- **All four scenarios proven** — Each must have E2E test coverage demonstrating the flow works
- **Conflict resolution round-trip** — Agent receives conflict, fixes it, merge succeeds
- **Question/answer resume** — Agent pauses for input, receives response, continues execution

</essential>

<specifics>
## Specific Ideas

- Use MockAgentManager scenarios to simulate waiting_for_input and crash states
- Use MockWorktreeManager to inject merge conflicts
- Test multi-agent by pre-seeding multiple idle agents and dispatching parallel tasks
- No CI pipeline needed — focus purely on test coverage

</specifics>

<notes>
## Additional Context

CI was explicitly deferred — not part of this phase scope. Phase goal is proving system resilience and coordination, not automation infrastructure.

</notes>

---

*Phase: 09-extended-scenarios*
*Context gathered: 2026-01-31*