fix: clear stale dispatch gate after context reset (#3210) (#3223)

2026-08-31 22:37:19 +00:00
parent 4e3dc91607
commit 9f2bb78aa2
4 changed files with 200 additions and 0 deletions
@@ -810,6 +810,7 @@ func (m *Manager) ResetAgentContext(ctx context.Context, executionID string) err
 		case <-exec.promptDoneCh:
 		default:
 		}
+		exec.dispatchedPromptPending.Store(false)
 	})

 	// Restore the complete captured configuration. A strict-mode model that is
@@ -1151,6 +1152,7 @@ func (m *Manager) resetAgentRestartState(executionID string, commands agentComma
 		case <-exec.promptDoneCh:
 		default:
 		}
+		exec.dispatchedPromptPending.Store(false)
 	})
 }

@@ -250,6 +250,10 @@ func newRestartMockAgentctlServer(t *testing.T, failStop, failSessionNew bool) *
 						"success": true,
 					})
 				}
+			case "agent.prompt":
+				resp, _ = ws.NewResponse(msg.ID, msg.Action, map[string]interface{}{
+					"success": true,
+				})
 			case "agent.stderr":
 				stderrLines := m.stderrLines
 				if len(stderrLines) == 0 {
@@ -406,6 +410,7 @@ func TestManager_RestartAgentProcess_Success(t *testing.T) {
 	exec.assistantHistoryBuffer.WriteString("old response")
 	exec.needsResumeContext = true
 	exec.resumeContextInjected = true
+	exec.dispatchedPromptPending.Store(true)
 	exec.promptDoneCh <- PromptCompletionSignal{StopReason: "stale"}

 	mgr.executionStore.Add(exec)
@@ -443,6 +448,9 @@ func TestManager_RestartAgentProcess_Success(t *testing.T) {
 		t.Fatalf("expected stale prompt signal to be drained")
 	default:
 	}
+	if exec.dispatchedPromptPending.Load() {
+		t.Fatal("expected stale dispatch gate to be cleared")
+	}

 	httpActions := mock.getHTTPActions()
 	if !slices.Equal(httpActions, []string{"stop", "configure", "start"}) {
@@ -854,6 +862,47 @@ func TestManager_ResetAgentContext_ReappliesSessionMode(t *testing.T) {
 	require.Zero(t, exec.assistantHistoryBuffer.Len())
 }

+// @covers AC-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002.3
+func TestManager_ResetAgentContext_ClearsIdleDispatchGate(t *testing.T) {
+	mgr := newTestManager(t)
+	mock := newRestartMockAgentctlServer(t, false, false)
+
+	client := createTestClient(t, mock.server.URL)
+	t.Cleanup(client.Close)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
+	defer cancel()
+
+	require.NoError(t, client.StreamUpdates(ctx, func(agentctl.AgentEvent) {}, nil, nil))
+
+	exec := &AgentExecution{
+		ID:                 "exec-idle-reset",
+		TaskID:             "task-1",
+		SessionID:          "session-1",
+		AgentProfileID:     "profile-1",
+		ACPSessionID:       "old-session",
+		AgentCommand:       "auggie --model test",
+		Status:             v1.AgentStatusRunning,
+		WorkspacePath:      "/workspace",
+		sessionInitialized: true,
+		agentctl:           client,
+		promptDoneCh:       make(chan PromptCompletionSignal, 1),
+	}
+	exec.dispatchedPromptPending.Store(true)
+	exec.promptDoneCh <- PromptCompletionSignal{StopReason: "end_turn"}
+	require.NoError(t, mgr.executionStore.Add(exec))
+
+	require.NoError(t, mgr.ResetAgentContext(ctx, exec.ID))
+	require.False(t, exec.dispatchedPromptPending.Load(),
+		"reset must clear the gate when it drains the completion signal")
+
+	followUpCtx, followUpCancel := context.WithTimeout(context.Background(), 3*time.Second)
+	defer followUpCancel()
+	_, err := mgr.PromptAgent(followUpCtx, exec.ID, "follow-up", nil, true)
+	require.NoError(t, err, "follow-up prompt must reach the new agent context")
+	require.Contains(t, mock.getWSActions(), "agent.prompt")
+}
+
 // TestManager_ResetAgentContext_ReappliesSessionModel is the regression test
 // for an ACP fast-path reset replacing the task's selected model with the
 // provider default from the freshly-created session.
@@ -0,0 +1,62 @@
+---
+created: 2026-08-31
+status: done
+requirements:
+  - REQ-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002
+system_design:
+  - ../../specs/tasks/system-design/workflow-step-agent-start-ownership.md
+legacy_specs: []
+---
+
+# Implementation Plan: Clear the Idle Context-Reset Dispatch Gate
+
+## Overview
+
+Clear the dispatch-only completion gate when an idle context reset removes its buffered completion signal. Cover the ACP reset path and the process-restart fallback.
+
+## Scope
+
+### In scope
+
+- Reproduce the idle dispatch-only state in lifecycle manager tests.
+- Clear `dispatchedPromptPending` after each successful reset path drains `promptDoneCh`.
+- Prove that the next prompt can reach agentctl after an ACP session reset.
+- Prove that the process-restart fallback also clears the old gate.
+
+### Out of scope
+
+- Change the timeout behavior in `waitForPendingDispatchedPrompt`.
+- Change active-turn quiescence or cancellation ownership.
+- Add UI recovery for sessions that entered this state before the correction.
+- Explain unrelated differences between prompt-completion log counts.
+
+## Technical approach
+
+Update `Manager.ResetAgentContext` in `manager_interaction.go`. Clear the atomic pending flag in the same locked state update that drains the old completion signal.
+
+Update `Manager.resetAgentRestartState` in the same file. Apply the same state transition to the process-restart fallback.
+
+Add focused tests in `manager_interaction_test.go`. The ACP test will reset an idle execution and send a follow-up dispatch-only prompt through the real manager path.
+
+Extend the existing restart-state test with a pending gate. The test will prove that process replacement clears the signal and its matching flag.
+
+## Tests
+
+- `AC-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002.3`: `TestManager_ResetAgentContext_ClearsIdleDispatchGate` proves that the follow-up prompt reaches the new ACP session.
+- `AC-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002.3`: `TestManager_RestartAgentProcess_Success` proves that the process-restart fallback clears the prior gate.
+- `AC-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002.5`: Existing pending-wait tests remain unchanged because the timeout still preserves active predecessor ownership.
+
+## Work orders
+
+- [x] [Task 01: Clear the idle reset dispatch gate](task-01-clear-idle-reset-dispatch-gate.md)
+
+## Verification results
+
+- RED: Both focused regression tests failed because the reset paths kept `dispatchedPromptPending` set.
+- GREEN: The focused tests passed after both reset paths cleared the flag after the signal drain.
+- Final: The race-enabled verification passed 2 tests. `git diff --check` also passed.
+
+## Risks
+
+- Clearing the gate before the old turn is quiescent can admit concurrent prompts. The existing orchestrator reset boundary must remain unchanged.
+- Clearing only the ACP fast path leaves adapters that use process replacement exposed to the same error.
@@ -0,0 +1,87 @@
+---
+id: "01-clear-idle-reset-dispatch-gate"
+title: "Clear the idle reset dispatch gate"
+status: done
+wave: 1
+depends_on: []
+plan: "plan.md"
+requirements:
+  - REQ-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002
+acceptance_criteria:
+  - AC-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002.3
+  - AC-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002.5
+system_design:
+  - ../../specs/tasks/system-design/workflow-step-agent-start-ownership.md
+---
+
+# Task 01: Clear the Idle Reset Dispatch Gate
+
+## Summary
+
+Clear the pending dispatch-only gate when a successful idle context reset removes its completion signal. Apply the correction to both provider-reset paths.
+
+## In scope
+
+- Add a failing regression test for an idle ACP reset with a buffered dispatch-only completion.
+- Prove that a follow-up dispatch-only prompt reaches agentctl after the reset.
+- Add pending-gate coverage to the existing process-restart state test.
+- Clear the old pending flag after both reset paths drain the old signal.
+
+## Out of scope
+
+- Clear the gate when `waitForPendingDispatchedPrompt` times out.
+- Change cancellation escalation or active-turn reset behavior.
+- Add frontend behavior or browser tests.
+
+## Acceptance
+
+- An ACP session reset removes the old completion signal and clears its matching pending flag.
+- The next prompt reaches agentctl without a predecessor timeout.
+- The process-restart fallback clears the same signal and flag pair.
+
+## Verification
+
+```bash
+go test -race ./internal/agent/runtime/lifecycle -run 'TestManager_(ResetAgentContext_ClearsIdleDispatchGate|RestartAgentProcess_Success)$' -count=1
+git diff --check
+```
+
+Run the commands from `apps/backend` and the repository root, respectively.
+
+## Files likely touched
+
+- `apps/backend/internal/agent/runtime/lifecycle/manager_interaction.go`
+- `apps/backend/internal/agent/runtime/lifecycle/manager_interaction_test.go`
+- `docs/plans/idle-context-reset-dispatch-gate/plan.md`
+- `docs/plans/idle-context-reset-dispatch-gate/task-01-clear-idle-reset-dispatch-gate.md`
+
+## Dependencies
+
+None.
+
+## Risks
+
+- A reset path that drains the signal without clearing the flag will keep the execution unusable.
+- A flag clear before successful context replacement can release a valid active generation.
+
+## Parallelism
+
+`sequential`
+
+## Inputs
+
+- `REQ-TASKS-WORKFLOW-STEP-AGENT-START-OWNERSHIP-002`
+- `docs/specs/tasks/system-design/workflow-step-agent-start-ownership.md`
+- `docs/decisions/2026-08-30-context-reset-quiesces-active-turn.md`
+- GitHub issue `#3210`
+- Existing reset and pending-prompt tests in the lifecycle package.
+
+## Results
+
+- RED: The focused command failed 2 tests before the production correction.
+- `TestManager_RestartAgentProcess_Success` found that process replacement kept the stale gate.
+- `TestManager_ResetAgentContext_ClearsIdleDispatchGate` found the same state after an ACP reset.
+- GREEN: The focused command passed 2 tests after both reset paths cleared the gate.
+- Final: The race-enabled verification passed 2 tests. `git diff --check` passed.
+- Changed files: `manager_interaction.go`, `manager_interaction_test.go`, and this plan package.
+- No known blockers remain.