diff --git a/.planning/phases/09-extended-scenarios/09-01-PLAN.md b/.planning/phases/09-extended-scenarios/09-01-PLAN.md
new file mode 100644
index 0000000..50315ee
--- /dev/null
+++ b/.planning/phases/09-extended-scenarios/09-01-PLAN.md
@@ -0,0 +1,145 @@
+---
+phase: 09-extended-scenarios
+plan: 01
+type: execute
+wave: 1
+depends_on: []
+files_modified: [src/test/e2e/extended-scenarios.test.ts]
+autonomous: true
+---
+
+
+Create E2E tests proving conflict hand-back round-trip and multi-agent parallel completion work correctly.
+
+Purpose: Validate the full conflict resolution cycle (conflict detected -> agent resolves -> merge succeeds) and multiple agents working/completing/merging in parallel.
+Output: Extended E2E test file with conflict round-trip and parallel completion scenarios.
+
+
+
+@~/.claude/get-shit-done/workflows/execute-plan.md
+@~/.claude/get-shit-done/templates/summary.md
+
+
+
+@.planning/PROJECT.md
+@.planning/ROADMAP.md
+@.planning/STATE.md
+@.planning/phases/09-extended-scenarios/09-CONTEXT.md
+@.planning/phases/08-e2e-scenario-tests/08-01-SUMMARY.md
+@.planning/phases/08-e2e-scenario-tests/08-02-SUMMARY.md
+
+@src/test/harness.ts
+@src/test/fixtures.ts
+@src/test/e2e/edge-cases.test.ts
+@src/test/e2e/happy-path.test.ts
+
+
+
+
+
+ Task 1: Create conflict hand-back round-trip tests
+ src/test/e2e/extended-scenarios.test.ts
+
+Create new test file with describe block "Conflict hand-back round-trip". Test the full cycle:
+
+1. **Test: conflict triggers resolution task, agent resolves, merge succeeds**
+ - Seed SIMPLE_FIXTURE, complete Task A
+ - Create agent in agentRepository with worktreeId
+ - Create worktree via MockWorktreeManager
+ - Set merge conflict result for first merge attempt
+ - Queue and process merge (should fail with conflict)
+ - Verify: merge:conflicted event, task marked blocked, resolution task created
+ - Clear the merge conflict (setMergeResult to success)
+ - Find the resolution task, dispatch it to an agent
+ - Complete the resolution task
+ - Queue original task for re-merge
+ - Process merge again (should succeed)
+ - Verify: merge:completed event for original task
+
+2. **Test: conflict resolution preserves original task context**
+ - Similar setup with conflict
+ - Verify resolution task has correct parentTaskId linking back
+ - Verify resolution task prompt contains conflict file info
+
+3. **Test: multiple sequential conflicts resolved in order**
+ - Set up 2 tasks (A and B) both with conflicts
+ - Process merges, both fail
+ - Resolve A's conflict, merge A succeeds
+ - Resolve B's conflict, merge B succeeds
+ - Verify merge order matches resolution order
+
+Use same patterns as edge-cases.test.ts:
+- vi.useFakeTimers() for async control
+- Pre-seed idle agent before dispatch
+- harness.setAgentScenario for agent behavior
+- harness.worktreeManager.setMergeResult for conflict injection
+- Manual agentRepository.create for coordination tests
+
+ npm test src/test/e2e/extended-scenarios.test.ts -- --run passes
+ 3+ conflict round-trip tests passing, proving full cycle works
+
+
+
+ Task 2: Create multi-agent parallel completion tests
+ src/test/e2e/extended-scenarios.test.ts
+
+Add describe block "Multi-agent parallel work" to the test file. Test scenarios:
+
+1. **Test: multiple agents complete tasks in parallel**
+ - Seed PARALLEL_FIXTURE (4 independent tasks)
+ - Pre-seed 3 idle agents
+ - Queue all 4 tasks
+ - Dispatch 3 tasks in parallel (3 agents working)
+ - Use vi.runAllTimersAsync() to complete all 3 agents
+ - Verify: 3 agent:stopped events
+ - Complete all 3 tasks
+ - Dispatch remaining task
+ - Verify: all 4 tasks completed
+
+2. **Test: parallel merges process in correct dependency order**
+ - Use COMPLEX_FIXTURE (has dependency structure)
+ - Complete Task 1A and Task 1B (no dependencies)
+ - Set up worktrees and agents for both
+ - Queue both for merge
+ - Process merges - both should succeed (no dependencies between them)
+ - Verify: merge:completed for both in same batch
+ - Complete Task 2A (depends on 1A) and Task 3A (depends on 1B)
+ - Queue and merge - should succeed
+ - Complete Task 4A (depends on 2A and 3A)
+ - Queue and merge - should succeed
+ - Verify: final merge order respects dependency graph
+
+3. **Test: parallel dispatch with mixed outcomes**
+ - Pre-seed 2 agents
+ - Dispatch 2 tasks, one set to succeed, one set to crash
+ - Verify: one agent:stopped, one agent:crashed
+ - Verify: completed task can merge, crashed task stays in_progress
+
+Use PARALLEL_FIXTURE for independent tasks, COMPLEX_FIXTURE for dependency scenarios.
+
+ npm test src/test/e2e/extended-scenarios.test.ts -- --run passes
+ 3+ parallel work tests passing, proving multi-agent scenarios work
+
+
+
+
+
+Before declaring plan complete:
+- [ ] `npm test src/test/e2e/extended-scenarios.test.ts -- --run` passes
+- [ ] At least 6 new tests (3 conflict + 3 parallel)
+- [ ] No flaky tests (run twice to verify)
+- [ ] Test file follows existing patterns from edge-cases.test.ts
+
+
+
+
+- All tasks completed
+- All verification checks pass
+- Conflict round-trip fully tested (detect -> resolve -> re-merge)
+- Multi-agent parallel scenarios validated
+- No regressions in existing E2E tests
+
+
+
diff --git a/.planning/phases/09-extended-scenarios/09-02-PLAN.md b/.planning/phases/09-extended-scenarios/09-02-PLAN.md
new file mode 100644
index 0000000..4ac8f72
--- /dev/null
+++ b/.planning/phases/09-extended-scenarios/09-02-PLAN.md
@@ -0,0 +1,147 @@
+---
+phase: 09-extended-scenarios
+plan: 02
+type: execute
+wave: 1
+depends_on: []
+files_modified: [src/test/e2e/recovery-scenarios.test.ts]
+autonomous: true
+---
+
+
+Create E2E tests proving recovery/resume after interruption and extended agent Q&A scenarios work correctly.
+
+Purpose: Validate system can recover state after interruption and handle complex agent question/answer flows.
+Output: Recovery scenarios test file with state persistence and Q&A flow tests.
+
+
+
+@~/.claude/get-shit-done/workflows/execute-plan.md
+@~/.claude/get-shit-done/templates/summary.md
+
+
+
+@.planning/PROJECT.md
+@.planning/ROADMAP.md
+@.planning/STATE.md
+@.planning/phases/09-extended-scenarios/09-CONTEXT.md
+@.planning/phases/08-e2e-scenario-tests/08-02-SUMMARY.md
+
+@src/test/harness.ts
+@src/test/fixtures.ts
+@src/test/e2e/edge-cases.test.ts
+
+
+
+
+
+ Task 1: Create recovery/resume scenario tests
+ src/test/e2e/recovery-scenarios.test.ts
+
+Create new test file with describe block "Recovery after interruption". Test scenarios:
+
+1. **Test: queue state survives harness recreation**
+ - Seed fixture, queue tasks
+ - Get queue state (tasks in queue)
+ - Create NEW harness pointing to SAME database
+ - Query queue state from new harness
+ - Verify: queue state matches (tasks still queued)
+
+ Implementation note: createTestHarness() creates fresh in-memory DB. For this test, need to:
+ - Extract DB from first harness
+ - Create second harness manually reusing same DB
+ - Or modify test to verify DB persistence directly
+
+2. **Test: in-progress task recoverable after agent crash**
+ - Dispatch task, agent crashes mid-execution
+ - Verify task status is 'in_progress' (not completed, not lost)
+ - Queue same task again (should be dispatchable)
+ - Dispatch to new agent
+ - Agent completes successfully
+ - Verify: task completed, merge can proceed
+
+3. **Test: blocked task state persists and can be unblocked**
+ - Queue task, block it with reason
+ - Verify task in blocked state in DB
+ - "Simulate restart" by recreating managers with same DB
+ - Query blocked tasks
+ - Unblock task
+ - Verify: task now dispatchable
+
+4. **Test: merge queue state recoverable**
+ - Complete task, queue for merge
+ - Verify merge queue has pending item
+ - Query merge queue state
+ - Process merge
+ - Verify: merge completes correctly
+
+Focus on proving that DATABASE STATE is the source of truth and managers can be recreated without losing work.
+
+ npm test src/test/e2e/recovery-scenarios.test.ts -- --run passes
+ 4 recovery tests passing, proving state persistence works
+
+
+
+ Task 2: Create extended agent Q&A scenario tests
+ src/test/e2e/recovery-scenarios.test.ts
+
+Add describe block "Agent Q&A extended scenarios" to the test file. Test scenarios:
+
+1. **Test: multiple questions in sequence from same agent**
+ - Dispatch task with scenario: first asks question, then after resume asks another
+ - Handle first question (agent:waiting -> resume -> agent:resumed)
+ - Agent asks second question
+ - Handle second question
+ - Agent completes
+ - Verify: 2 agent:waiting events, 2 agent:resumed events, 1 agent:stopped
+
+ Implementation: MockAgentManager may need scenario that asks multiple questions. If not supported, test single question but verify the state machine works correctly.
+
+2. **Test: question surfaces as message in message queue**
+ - Dispatch task with waiting_for_input scenario
+ - Verify: agent:waiting event includes question
+ - Check messageRepository for user-directed message
+ - Verify: message contains the question text
+
+3. **Test: agent resumes with user's answer in context**
+ - Dispatch task, agent asks question
+ - Resume with specific answer "PostgreSQL"
+ - Verify: resume call includes the answer
+ - Agent completes
+ - Verify: agent result reflects successful completion
+
+4. **Test: waiting agent blocks task completion**
+ - Dispatch task, agent enters waiting_for_input
+ - Attempt to complete task (should not be allowed while agent waiting)
+ - Resume agent, agent completes
+ - Now complete task
+ - Verify: proper state transitions
+
+Use edge-cases.test.ts patterns for waiting/resume flow.
+
+ npm test src/test/e2e/recovery-scenarios.test.ts -- --run passes
+ 4 Q&A tests passing, proving extended question flows work
+
+
+
+
+
+Before declaring plan complete:
+- [ ] `npm test src/test/e2e/recovery-scenarios.test.ts -- --run` passes
+- [ ] At least 8 new tests (4 recovery + 4 Q&A)
+- [ ] No flaky tests (run twice to verify)
+- [ ] Test patterns consistent with existing E2E tests
+
+
+
+
+- All tasks completed
+- All verification checks pass
+- Recovery scenarios prove database is source of truth
+- Q&A flow handles multiple questions and state transitions
+- No regressions in existing E2E tests
+
+
+