fix(subagents): report queued persistence failures

commit: 642e3ef8b4412e429a9315e4375149d65c5dd54a
author: brice · 2mo ago · 2026-08-14 06:55 UTC
parent: 31b43ea
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

2 files · +57 -0

M packages/agent/subagents/resident_child.go +8 -0


@@ -349,6 +349,9 @@ func (c *ResidentChild) run() {
349	349		c.cancel()
350	350		if c.onCompletion != nil {
351	351		c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: active.turnID, Task: active.prompt, Err: terminalErr})
	352	+	for _, pending := range queue {
	353	+	c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: pending.turnID, Task: pending.prompt, Err: terminalErr})
	354	+	}
352	355		}
353	356		return
354	357		}

@@ -438,6 +441,11 @@ func (c *ResidentChild) run() {
438	441		}
439	442		}
440	443		c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: result.turnID, Task: active.prompt, Err: terminalErr, Summary: summary})
	444	+	if persistenceFailed {
	445	+	for _, pending := range queue {
	446	+	c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: pending.turnID, Task: pending.prompt, Err: terminalErr})
	447	+	}
	448	+	}
441	449		}
442	450		if persistenceFailed {
443	451		return


@@ -349,6 +349,9 @@ func (c *ResidentChild) run() {
349	c.cancel()	349		c.cancel()
350	if c.onCompletion != nil {	350		if c.onCompletion != nil {
351	c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: active.turnID, Task: active.prompt, Err: terminalErr})	351		c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: active.turnID, Task: active.prompt, Err: terminalErr})
		352	+	for _, pending := range queue {
		353	+	c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: pending.turnID, Task: pending.prompt, Err: terminalErr})
		354	+	}
352	}	355		}
353	return	356		return
354	}	357		}

@@ -438,6 +441,11 @@ func (c *ResidentChild) run() {
438	}	441		}
439	}	442		}
440	c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: result.turnID, Task: active.prompt, Err: terminalErr, Summary: summary})	443		c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: result.turnID, Task: active.prompt, Err: terminalErr, Summary: summary})
		444	+	if persistenceFailed {
		445	+	for _, pending := range queue {
		446	+	c.onCompletion(ResidentCompletion{ChildID: c.spec.ID, TurnID: pending.turnID, Task: pending.prompt, Err: terminalErr})
		447	+	}
		448	+	}
441	}	449		}
442	if persistenceFailed {	450		if persistenceFailed {
443	return	451		return

M packages/agent/subagents/resident_manager_test.go +49 -0


@@ -78,6 +78,55 @@ func TestResidentManagerCompletionCarriesFinalSummary(t *testing.T) {
78	78		}
79	79		}
80	80
	81	+	func TestResidentManagerReportsQueuedTurnsWhenFinalResultCannotPersist(t *testing.T) {
	82	+	started := make(chan struct{})
	83	+	release := make(chan struct{})
	84	+	completed := make(chan ResidentCompletion, 2)
	85	+	var journal *ResidentJournal
	86	+	manager := NewResidentManager(t.TempDir(), func(_ ResidentChildSpec, childJournal *ResidentJournal) (ResidentTurnRunner, error) {
	87	+	journal = childJournal
	88	+	return func(context.Context, string) error {
	89	+	close(started)
	90	+	<-release
	91	+	return nil
	92	+	}, nil
	93	+	})
	94	+	manager.SetCompletionObserver(func(completion ResidentCompletion) { completed <- completion })
	95	+	t.Cleanup(func() { _ = manager.Close(context.Background()) })
	96	+
	97	+	if _, err := manager.Spawn(context.Background(), ResidentChildSpec{ID: "persistence-failure-child", SessionID: "child-session", Provider: "openai", Model: "gpt-5"}, "initial task"); err != nil {
	98	+	t.Fatal(err)
	99	+	}
	100	+	select {
	101	+	case <-started:
	102	+	case <-time.After(time.Second):
	103	+	t.Fatal("resident child did not start")
	104	+	}
	105	+	if err := manager.Resume(context.Background(), "persistence-failure-child", "queued follow-up"); err != nil {
	106	+	t.Fatalf("Resume: %v", err)
	107	+	}
	108	+	if err := journal.Close(); err != nil {
	109	+	t.Fatalf("Close journal: %v", err)
	110	+	}
	111	+	close(release)
	112	+
	113	+	seen := make(map[string]ResidentCompletion, 2)
	114	+	for range 2 {
	115	+	select {
	116	+	case completion := <-completed:
	117	+	seen[completion.Task] = completion
	118	+	case <-time.After(time.Second):
	119	+	t.Fatalf("terminal completions = %#v, want initial and queued turns", seen)
	120	+	}
	121	+	}
	122	+	for _, task := range []string{"initial task", "queued follow-up"} {
	123	+	completion, ok := seen[task]
	124	+	if !ok \|\| completion.TurnID == "" \|\| completion.Err == nil \|\| !strings.Contains(completion.Err.Error(), "persist resident child terminal state") {
	125	+	t.Fatalf("completion for %q = %#v", task, completion)
	126	+	}
	127	+	}
	128	+	}
	129	+
81	130		func TestResidentManagerReportsAcceptedInterruptedTurn(t *testing.T) {
82	131		started := make(chan struct{})
83	132		completed := make(chan ResidentCompletion, 1)


@@ -78,6 +78,55 @@ func TestResidentManagerCompletionCarriesFinalSummary(t *testing.T) {
78	}	78		}
79	}	79		}
80		80
		81	+	func TestResidentManagerReportsQueuedTurnsWhenFinalResultCannotPersist(t *testing.T) {
		82	+	started := make(chan struct{})
		83	+	release := make(chan struct{})
		84	+	completed := make(chan ResidentCompletion, 2)
		85	+	var journal *ResidentJournal
		86	+	manager := NewResidentManager(t.TempDir(), func(_ ResidentChildSpec, childJournal *ResidentJournal) (ResidentTurnRunner, error) {
		87	+	journal = childJournal
		88	+	return func(context.Context, string) error {
		89	+	close(started)
		90	+	<-release
		91	+	return nil
		92	+	}, nil
		93	+	})
		94	+	manager.SetCompletionObserver(func(completion ResidentCompletion) { completed <- completion })
		95	+	t.Cleanup(func() { _ = manager.Close(context.Background()) })
		96	+
		97	+	if _, err := manager.Spawn(context.Background(), ResidentChildSpec{ID: "persistence-failure-child", SessionID: "child-session", Provider: "openai", Model: "gpt-5"}, "initial task"); err != nil {
		98	+	t.Fatal(err)
		99	+	}
		100	+	select {
		101	+	case <-started:
		102	+	case <-time.After(time.Second):
		103	+	t.Fatal("resident child did not start")
		104	+	}
		105	+	if err := manager.Resume(context.Background(), "persistence-failure-child", "queued follow-up"); err != nil {
		106	+	t.Fatalf("Resume: %v", err)
		107	+	}
		108	+	if err := journal.Close(); err != nil {
		109	+	t.Fatalf("Close journal: %v", err)
		110	+	}
		111	+	close(release)
		112	+
		113	+	seen := make(map[string]ResidentCompletion, 2)
		114	+	for range 2 {
		115	+	select {
		116	+	case completion := <-completed:
		117	+	seen[completion.Task] = completion
		118	+	case <-time.After(time.Second):
		119	+	t.Fatalf("terminal completions = %#v, want initial and queued turns", seen)
		120	+	}
		121	+	}
		122	+	for _, task := range []string{"initial task", "queued follow-up"} {
		123	+	completion, ok := seen[task]
		124	+	if !ok \|\| completion.TurnID == "" \|\| completion.Err == nil \|\| !strings.Contains(completion.Err.Error(), "persist resident child terminal state") {
		125	+	t.Fatalf("completion for %q = %#v", task, completion)
		126	+	}
		127	+	}
		128	+	}
		129	+
81	func TestResidentManagerReportsAcceptedInterruptedTurn(t *testing.T) {	130		func TestResidentManagerReportsAcceptedInterruptedTurn(t *testing.T) {
82	started := make(chan struct{})	131		started := make(chan struct{})
83	completed := make(chan ResidentCompletion, 1)	132		completed := make(chan ResidentCompletion, 1)