fix(subagents): remove lifetime spawn budget

commit: 755e82eb988ea908426b6123378630e25d0a0fa9
author: brice · 2mo ago · 2026-08-06 19:39 UTC
committer: brice · 2mo ago · 2026-08-06 19:41 UTC
parent: 21b66ba
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

10 files · +105 -46

M README.md +1 -1


@@ -387,7 +387,7 @@ Background subagents that run alongside your main session. Each one is a separat
387	387
388	388		`/session export` does NOT bundle subagents. A `.zutsession` is just the main chat transcript; per-agent state (session file, unix-socket inbox) is machine-local and doesn't round-trip through a JSONL file. To share what an agent said, copy it out of the transcript view manually.
389	389
390		-	Auto-subagents. With `/settings` -> auto-subagents on, the main agent gets the canonical `subagent_spawn` tool and a system-prompt nudge to delegate independent work. The companion `subagent_status` tool accepts an optional `agent_id`: omit it to list visible workers, or pass a full id / unique prefix to query one without waiting for completion. It returns privacy-safe JSON metadata with the worker id, lifecycle state (`starting`, `running`, `completed`, `failed`, `cancelled`, or `detached`), underlying process/turn state, timestamps, a bounded first-line task summary, and terminal-result metadata when available; it does not return prompts, transcripts, result output, credentials, provider settings, or filesystem paths. The spawn tool returns the child id and initial state immediately; completion is a durable `turn.result` with a `subagent://<id>/result` reference. It accepts named profiles, timeout, `max_turns`, and `isolation:"shared"` or `isolation:"worktree"`. Omit `max_turns` to use the supervisor policy; the default ceiling is 3 and `subagents.max_turns` changes it. Global concurrency and total-spawn budgets apply equally to slash commands, tool calls, and batches. Off by default; toggle from `/settings`.
	390	+	Auto-subagents. With `/settings` -> auto-subagents on, the main agent gets the canonical `subagent_spawn` tool and a system-prompt nudge to delegate independent work. The companion `subagent_status` tool accepts an optional `agent_id`: omit it to list visible workers, or pass a full id / unique prefix to query one without waiting for completion. It returns privacy-safe JSON metadata with the worker id, lifecycle state (`starting`, `running`, `completed`, `failed`, `cancelled`, or `detached`), underlying process/turn state, timestamps, a bounded first-line task summary, and terminal-result metadata when available; it does not return prompts, transcripts, result output, credentials, provider settings, or filesystem paths. The spawn tool returns the child id and initial state immediately; completion is a durable `turn.result` with a `subagent://<id>/result` reference. It accepts named profiles, timeout, `max_turns`, and `isolation:"shared"` or `isolation:"worktree"`. Omit `max_turns` to use the supervisor policy; the default ceiling is 3 and `subagents.max_turns` changes it. Global concurrency limits apply equally to slash commands, tool calls, and batches. Off by default; toggle from `/settings`.
391	391
392	392		### Named subagent profiles
393	393


@@ -387,7 +387,7 @@ Background subagents that run alongside your main session. Each one is a separat
387			387
388		`/session export` does NOT bundle subagents. A `.zutsession` is just the main chat transcript; per-agent state (session file, unix-socket inbox) is machine-local and doesn't round-trip through a JSONL file. To share what an agent said, copy it out of the transcript view manually.	388		`/session export` does NOT bundle subagents. A `.zutsession` is just the main chat transcript; per-agent state (session file, unix-socket inbox) is machine-local and doesn't round-trip through a JSONL file. To share what an agent said, copy it out of the transcript view manually.
389			389
390	-	Auto-subagents. With `/settings` -> auto-subagents on, the main agent gets the canonical `subagent_spawn` tool and a system-prompt nudge to delegate independent work. The companion `subagent_status` tool accepts an optional `agent_id`: omit it to list visible workers, or pass a full id / unique prefix to query one without waiting for completion. It returns privacy-safe JSON metadata with the worker id, lifecycle state (`starting`, `running`, `completed`, `failed`, `cancelled`, or `detached`), underlying process/turn state, timestamps, a bounded first-line task summary, and terminal-result metadata when available; it does not return prompts, transcripts, result output, credentials, provider settings, or filesystem paths. The spawn tool returns the child id and initial state immediately; completion is a durable `turn.result` with a `subagent://<id>/result` reference. It accepts named profiles, timeout, `max_turns`, and `isolation:"shared"` or `isolation:"worktree"`. Omit `max_turns` to use the supervisor policy; the default ceiling is 3 and `subagents.max_turns` changes it. Global concurrency and total-spawn budgets apply equally to slash commands, tool calls, and batches. Off by default; toggle from `/settings`.	390	+	Auto-subagents. With `/settings` -> auto-subagents on, the main agent gets the canonical `subagent_spawn` tool and a system-prompt nudge to delegate independent work. The companion `subagent_status` tool accepts an optional `agent_id`: omit it to list visible workers, or pass a full id / unique prefix to query one without waiting for completion. It returns privacy-safe JSON metadata with the worker id, lifecycle state (`starting`, `running`, `completed`, `failed`, `cancelled`, or `detached`), underlying process/turn state, timestamps, a bounded first-line task summary, and terminal-result metadata when available; it does not return prompts, transcripts, result output, credentials, provider settings, or filesystem paths. The spawn tool returns the child id and initial state immediately; completion is a durable `turn.result` with a `subagent://<id>/result` reference. It accepts named profiles, timeout, `max_turns`, and `isolation:"shared"` or `isolation:"worktree"`. Omit `max_turns` to use the supervisor policy; the default ceiling is 3 and `subagents.max_turns` changes it. Global concurrency limits apply equally to slash commands, tool calls, and batches. Off by default; toggle from `/settings`.
391			391
392		### Named subagent profiles	392		### Named subagent profiles
393			393

M docs/subagents.md +1 -1


@@ -111,4 +111,4 @@ When auto-subagents is enabled, the model also receives the read-only `subagent_
111	111
112	112		## Resource policy
113	113
114		-	The persisted `subagents` config object supports `max_concurrent`, `max_concurrent_per_parent`, `max_total_spawned`, `queue_timeout`, `default_timeout`, `max_turns`, output caps, allowed tools/roots, heartbeat and idle timeouts. A missing or non-positive configured `max_turns` uses the default ceiling of 3. The `max_turns` field in `subagent_spawn` is optional: omit it to use the policy ceiling, or provide a value from 1 through the configured maximum. Limits apply to slash commands, `subagent_spawn`, and batch operations. A child cannot create descendants in v1. Per-agent timeouts are retained across reload/resume. Packaged `zut run` agents keep their declared capability ceiling by disabling subagent delegation; profiles are not a substitute for that permission boundary.
	114	+	The persisted `subagents` config object supports `max_concurrent`, `max_concurrent_per_parent`, `queue_timeout`, `default_timeout`, `max_turns`, output caps, allowed tools/roots, heartbeat and idle timeouts. Previously saved `max_total_spawned` values are accepted but ignored. A missing or non-positive configured `max_turns` uses the default ceiling of 3. The `max_turns` field in `subagent_spawn` is optional: omit it to use the policy ceiling, or provide a value from 1 through the configured maximum. Concurrency limits apply to slash commands, `subagent_spawn`, and batch operations. A child cannot create descendants in v1. Per-agent timeouts are retained across reload/resume. Packaged `zut run` agents keep their declared capability ceiling by disabling subagent delegation; profiles are not a substitute for that permission boundary.


@@ -111,4 +111,4 @@ When auto-subagents is enabled, the model also receives the read-only `subagent_
111			111
112		## Resource policy	112		## Resource policy
113			113
114	-	The persisted `subagents` config object supports `max_concurrent`, `max_concurrent_per_parent`, `max_total_spawned`, `queue_timeout`, `default_timeout`, `max_turns`, output caps, allowed tools/roots, heartbeat and idle timeouts. A missing or non-positive configured `max_turns` uses the default ceiling of 3. The `max_turns` field in `subagent_spawn` is optional: omit it to use the policy ceiling, or provide a value from 1 through the configured maximum. Limits apply to slash commands, `subagent_spawn`, and batch operations. A child cannot create descendants in v1. Per-agent timeouts are retained across reload/resume. Packaged `zut run` agents keep their declared capability ceiling by disabling subagent delegation; profiles are not a substitute for that permission boundary.	114	+	The persisted `subagents` config object supports `max_concurrent`, `max_concurrent_per_parent`, `queue_timeout`, `default_timeout`, `max_turns`, output caps, allowed tools/roots, heartbeat and idle timeouts. Previously saved `max_total_spawned` values are accepted but ignored. A missing or non-positive configured `max_turns` uses the default ceiling of 3. The `max_turns` field in `subagent_spawn` is optional: omit it to use the policy ceiling, or provide a value from 1 through the configured maximum. Concurrency limits apply to slash commands, `subagent_spawn`, and batch operations. A child cannot create descendants in v1. Per-agent timeouts are retained across reload/resume. Packaged `zut run` agents keep their declared capability ceiling by disabling subagent delegation; profiles are not a substitute for that permission boundary.

M packages/agent/cli.go +0 -1


@@ -1241,7 +1241,6 @@ func subagentPolicyFromConfig(cfg SubagentsConfig) subagents.SubagentPolicy {
1241	1241		return subagents.SubagentPolicy{
1242	1242		MaxConcurrent: cfg.MaxConcurrent,
1243	1243		MaxConcurrentPerParent: cfg.MaxConcurrentPerParent,
1244		-	MaxTotalSpawned: cfg.MaxTotalSpawned,
1245	1244		QueueTimeout: parseDuration(cfg.QueueTimeout),
1246	1245		DefaultTimeout: parseDuration(cfg.DefaultTimeout),
1247	1246		MaxTurns: cfg.MaxTurns,


@@ -1241,7 +1241,6 @@ func subagentPolicyFromConfig(cfg SubagentsConfig) subagents.SubagentPolicy {
1241		return subagents.SubagentPolicy{	1241	return subagents.SubagentPolicy{
1242		MaxConcurrent: cfg.MaxConcurrent,	1242	MaxConcurrent: cfg.MaxConcurrent,
1243		MaxConcurrentPerParent: cfg.MaxConcurrentPerParent,	1243	MaxConcurrentPerParent: cfg.MaxConcurrentPerParent,
1244	-	MaxTotalSpawned: cfg.MaxTotalSpawned,
1245		QueueTimeout: parseDuration(cfg.QueueTimeout),	1244	QueueTimeout: parseDuration(cfg.QueueTimeout),
1246		DefaultTimeout: parseDuration(cfg.DefaultTimeout),	1245	DefaultTimeout: parseDuration(cfg.DefaultTimeout),
1247		MaxTurns: cfg.MaxTurns,	1246	MaxTurns: cfg.MaxTurns,

M packages/agent/config.go +4 -1


@@ -27,7 +27,6 @@ type QuickModelShortcut struct {
27	27		type SubagentsConfig struct {
28	28		MaxConcurrent int `json:"max_concurrent,omitempty"`
29	29		MaxConcurrentPerParent int `json:"max_concurrent_per_parent,omitempty"`
30		-	MaxTotalSpawned int `json:"max_total_spawned,omitempty"`
31	30		QueueTimeout string `json:"queue_timeout,omitempty"`
32	31		DefaultTimeout string `json:"default_timeout,omitempty"`
33	32		MaxTurns int `json:"max_turns,omitempty"`

@@ -39,6 +38,10 @@ type SubagentsConfig struct {
39	38		IdleTimeout string `json:"idle_timeout,omitempty"`
40	39		ReconnectTimeout string `json:"reconnect_timeout,omitempty"`
41	40		CancelGracePeriod string `json:"cancel_grace_period,omitempty"`
	41	+
	42	+	// MaxTotalSpawned is retained only to accept existing configuration.
	43	+	// Lifetime total-spawn budgets are no longer enforced.
	44	+	MaxTotalSpawned int `json:"max_total_spawned,omitempty"`
42	45		}
43	46
44	47		// Config is the persisted user configuration.

 type SubagentsConfig struct {
 	MaxConcurrent          int      `json:"max_concurrent,omitempty"`
 	MaxConcurrentPerParent int      `json:"max_concurrent_per_parent,omitempty"`
-	MaxTotalSpawned        int      `json:"max_total_spawned,omitempty"`
 	QueueTimeout           string   `json:"queue_timeout,omitempty"`
 	DefaultTimeout         string   `json:"default_timeout,omitempty"`
 	MaxTurns               int      `json:"max_turns,omitempty"`
 	IdleTimeout            string   `json:"idle_timeout,omitempty"`
 	ReconnectTimeout       string   `json:"reconnect_timeout,omitempty"`
 	CancelGracePeriod      string   `json:"cancel_grace_period,omitempty"`
+	// MaxTotalSpawned is retained only to accept existing configuration.
+	// Lifetime total-spawn budgets are no longer enforced.
+	MaxTotalSpawned int `json:"max_total_spawned,omitempty"`
 }
 // Config is the persisted user configuration.

A packages/agent/subagent_policy_test.go +44 -0

@@ -0,0 +1,44 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"context"
	5	+	"os"
	6	+	"testing"
	7	+
	8	+	"github.com/bnema/zut/packages/agent/subagents"
	9	+	)
	10	+
	11	+	func TestSubagentPolicyIgnoresLegacyMaxTotalSpawned(t *testing.T) {
	12	+	t.Setenv("ZUT_HOME", t.TempDir())
	13	+	if err := os.WriteFile(ConfigPath(), []byte(`{"subagents":{"max_concurrent":1,"max_total_spawned":1}}`), 0o600); err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+	cfg, err := LoadConfig()
	17	+	if err != nil {
	18	+	t.Fatal(err)
	19	+	}
	20	+	if got := cfg.Subagents.MaxTotalSpawned; got != 1 {
	21	+	t.Fatalf("legacy max_total_spawned = %d, want 1", got)
	22	+	}
	23	+
	24	+	supervisor := subagents.New(subagents.Config{
	25	+	Root: t.TempDir(),
	26	+	RepoRoot: t.TempDir(),
	27	+	Policy: subagentPolicyFromConfig(cfg.Subagents),
	28	+	NewRunner: func(*subagents.Agent) subagents.Runner {
	29	+	return subagents.RunnerFunc(func(context.Context, subagents.Sink) error { return nil })
	30	+	},
	31	+	})
	32	+	t.Cleanup(supervisor.StopAll)
	33	+
	34	+	for _, task := range []string{"first", "second"} {
	35	+	agent, err := supervisor.Spawn(context.Background(), task)
	36	+	if err != nil {
	37	+	t.Fatalf("Spawn(%q): %v", task, err)
	38	+	}
	39	+	agent.Wait()
	40	+	if got := agent.Status(); got != subagents.StatusDone {
	41	+	t.Fatalf("%s status = %s, want %s", task, got, subagents.StatusDone)
	42	+	}
	43	+	}
	44	+	}

@@ -0,0 +1,44 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"context"
	5	+	"os"
	6	+	"testing"
	7	+
	8	+	"github.com/bnema/zut/packages/agent/subagents"
	9	+	)
	10	+
	11	+	func TestSubagentPolicyIgnoresLegacyMaxTotalSpawned(t *testing.T) {
	12	+	t.Setenv("ZUT_HOME", t.TempDir())
	13	+	if err := os.WriteFile(ConfigPath(), []byte(`{"subagents":{"max_concurrent":1,"max_total_spawned":1}}`), 0o600); err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+	cfg, err := LoadConfig()
	17	+	if err != nil {
	18	+	t.Fatal(err)
	19	+	}
	20	+	if got := cfg.Subagents.MaxTotalSpawned; got != 1 {
	21	+	t.Fatalf("legacy max_total_spawned = %d, want 1", got)
	22	+	}
	23	+
	24	+	supervisor := subagents.New(subagents.Config{
	25	+	Root: t.TempDir(),
	26	+	RepoRoot: t.TempDir(),
	27	+	Policy: subagentPolicyFromConfig(cfg.Subagents),
	28	+	NewRunner: func(*subagents.Agent) subagents.Runner {
	29	+	return subagents.RunnerFunc(func(context.Context, subagents.Sink) error { return nil })
	30	+	},
	31	+	})
	32	+	t.Cleanup(supervisor.StopAll)
	33	+
	34	+	for _, task := range []string{"first", "second"} {
	35	+	agent, err := supervisor.Spawn(context.Background(), task)
	36	+	if err != nil {
	37	+	t.Fatalf("Spawn(%q): %v", task, err)
	38	+	}
	39	+	agent.Wait()
	40	+	if got := agent.Status(); got != subagents.StatusDone {
	41	+	t.Fatalf("%s status = %s, want %s", task, got, subagents.StatusDone)
	42	+	}
	43	+	}
	44	+	}

M packages/agent/subagents/batch.go +0 -6


@@ -80,12 +80,6 @@ func (f Supervisor) SpawnBatch(ctx context.Context, req BatchRequest) (Batch,
80	80		if len(req.Tasks) == 0 {
81	81		return nil, fmt.Errorf("subagents: batch requires at least one task")
82	82		}
83		-	f.mu.Lock()
84		-	remaining := f.cfg.Policy.MaxTotalSpawned - f.totalSpawned
85		-	f.mu.Unlock()
86		-	if len(req.Tasks) > remaining {
87		-	return nil, fmt.Errorf("subagents: batch needs %d spawn budget entries, only %d remain", len(req.Tasks), remaining)
88		-	}
89	83		if ctx == nil {
90	84		ctx = context.Background()
91	85		}


@@ -80,12 +80,6 @@ func (f Supervisor) SpawnBatch(ctx context.Context, req BatchRequest) (Batch,
80		if len(req.Tasks) == 0 {	80	if len(req.Tasks) == 0 {
81		return nil, fmt.Errorf("subagents: batch requires at least one task")	81	return nil, fmt.Errorf("subagents: batch requires at least one task")
82		}	82	}
83	-	f.mu.Lock()
84	-	remaining := f.cfg.Policy.MaxTotalSpawned - f.totalSpawned
85	-	f.mu.Unlock()
86	-	if len(req.Tasks) > remaining {
87	-	return nil, fmt.Errorf("subagents: batch needs %d spawn budget entries, only %d remain", len(req.Tasks), remaining)
88	-	}
89		if ctx == nil {	83	if ctx == nil {
90		ctx = context.Background()	84	ctx = context.Background()
91		}	85	}

M packages/agent/subagents/batch_test.go +31 -2


@@ -3,6 +3,7 @@ package subagents
3	3		import (
4	4		"context"
5	5		"encoding/json"
	6	+	"fmt"
6	7		"os"
7	8		"path/filepath"
8	9		"testing"

@@ -13,7 +14,7 @@ func TestBatchIDsRemainUniqueWithFixedClock(t *testing.T) {
13	14		now := time.Unix(123, 0)
14	15		f := New(Config{
15	16		Root: t.TempDir(), RepoRoot: t.TempDir(), Now: func() time.Time { return now },
16		-	Policy: SubagentPolicy{MaxConcurrent: 2, MaxTotalSpawned: 4},
	17	+	Policy: SubagentPolicy{MaxConcurrent: 2},
17	18		NewRunner: func(*Agent) Runner {
18	19		return RunnerFunc(func(context.Context, Sink) error { return nil })
19	20		},

@@ -104,7 +105,7 @@ func TestBatchSpawnWaitAndCollect(t *testing.T) {
104	105		root := t.TempDir()
105	106		f := New(Config{
106	107		Root: root, RepoRoot: root,
107		-	Policy: SubagentPolicy{MaxConcurrent: 2, MaxConcurrentPerParent: 2, MaxTotalSpawned: 4, DefaultTimeout: 0, IdleTimeout: 0},
	108	+	Policy: SubagentPolicy{MaxConcurrent: 2, MaxConcurrentPerParent: 2, DefaultTimeout: 0, IdleTimeout: 0},
108	109		NewRunner: func(*Agent) Runner {
109	110		return RunnerFunc(func(context.Context, Sink) error { return nil })
110	111		},

@@ -125,3 +126,31 @@ func TestBatchSpawnWaitAndCollect(t *testing.T) {
125	126		t.Fatalf("batch status = %s", batch.Status())
126	127		}
127	128		}
	129	+
	130	+	func TestBatchAllowsMoreThanFormerLifetimeLimit(t *testing.T) {
	131	+	const workersBeyondFormerLifetimeLimit = 33
	132	+	tasks := make([]string, workersBeyondFormerLifetimeLimit)
	133	+	for i := range tasks {
	134	+	tasks[i] = fmt.Sprintf("worker-%d", i)
	135	+	}
	136	+	f := New(Config{
	137	+	Root: t.TempDir(), RepoRoot: t.TempDir(),
	138	+	Policy: SubagentPolicy{MaxConcurrent: 1, MaxConcurrentPerParent: 1},
	139	+	NewRunner: func(*Agent) Runner {
	140	+	return RunnerFunc(func(context.Context, Sink) error { return nil })
	141	+	},
	142	+	})
	143	+	t.Cleanup(f.StopAll)
	144	+
	145	+	batch, err := f.SpawnBatch(context.Background(), BatchRequest{Tasks: tasks, MaxConcurrent: 1})
	146	+	if err != nil {
	147	+	t.Fatal(err)
	148	+	}
	149	+	result, err := f.WaitBatch(batch.ID)
	150	+	if err != nil {
	151	+	t.Fatal(err)
	152	+	}
	153	+	if result.Status != BatchSucceeded \|\| len(result.ChildIDs) != len(tasks) \|\| len(result.Results) != len(tasks) {
	154	+	t.Fatalf("batch result = %#v", result)
	155	+	}
	156	+	}


@@ -3,6 +3,7 @@ package subagents
3		import (	3		import (
4		"context"	4		"context"
5		"encoding/json"	5		"encoding/json"
			6	+	"fmt"
6		"os"	7		"os"
7		"path/filepath"	8		"path/filepath"
8		"testing"	9		"testing"

@@ -13,7 +14,7 @@ func TestBatchIDsRemainUniqueWithFixedClock(t *testing.T) {
13		now := time.Unix(123, 0)	14		now := time.Unix(123, 0)
14		f := New(Config{	15		f := New(Config{
15		Root: t.TempDir(), RepoRoot: t.TempDir(), Now: func() time.Time { return now },	16		Root: t.TempDir(), RepoRoot: t.TempDir(), Now: func() time.Time { return now },
16	-	Policy: SubagentPolicy{MaxConcurrent: 2, MaxTotalSpawned: 4},	17	+	Policy: SubagentPolicy{MaxConcurrent: 2},
17		NewRunner: func(*Agent) Runner {	18		NewRunner: func(*Agent) Runner {
18		return RunnerFunc(func(context.Context, Sink) error { return nil })	19		return RunnerFunc(func(context.Context, Sink) error { return nil })
19		},	20		},

@@ -104,7 +105,7 @@ func TestBatchSpawnWaitAndCollect(t *testing.T) {
104		root := t.TempDir()	105		root := t.TempDir()
105		f := New(Config{	106		f := New(Config{
106		Root: root, RepoRoot: root,	107		Root: root, RepoRoot: root,
107	-	Policy: SubagentPolicy{MaxConcurrent: 2, MaxConcurrentPerParent: 2, MaxTotalSpawned: 4, DefaultTimeout: 0, IdleTimeout: 0},	108	+	Policy: SubagentPolicy{MaxConcurrent: 2, MaxConcurrentPerParent: 2, DefaultTimeout: 0, IdleTimeout: 0},
108		NewRunner: func(*Agent) Runner {	109		NewRunner: func(*Agent) Runner {
109		return RunnerFunc(func(context.Context, Sink) error { return nil })	110		return RunnerFunc(func(context.Context, Sink) error { return nil })
110		},	111		},

@@ -125,3 +126,31 @@ func TestBatchSpawnWaitAndCollect(t *testing.T) {
125		t.Fatalf("batch status = %s", batch.Status())	126		t.Fatalf("batch status = %s", batch.Status())
126		}	127		}
127		}	128		}
			129	+
			130	+	func TestBatchAllowsMoreThanFormerLifetimeLimit(t *testing.T) {
			131	+	const workersBeyondFormerLifetimeLimit = 33
			132	+	tasks := make([]string, workersBeyondFormerLifetimeLimit)
			133	+	for i := range tasks {
			134	+	tasks[i] = fmt.Sprintf("worker-%d", i)
			135	+	}
			136	+	f := New(Config{
			137	+	Root: t.TempDir(), RepoRoot: t.TempDir(),
			138	+	Policy: SubagentPolicy{MaxConcurrent: 1, MaxConcurrentPerParent: 1},
			139	+	NewRunner: func(*Agent) Runner {
			140	+	return RunnerFunc(func(context.Context, Sink) error { return nil })
			141	+	},
			142	+	})
			143	+	t.Cleanup(f.StopAll)
			144	+
			145	+	batch, err := f.SpawnBatch(context.Background(), BatchRequest{Tasks: tasks, MaxConcurrent: 1})
			146	+	if err != nil {
			147	+	t.Fatal(err)
			148	+	}
			149	+	result, err := f.WaitBatch(batch.ID)
			150	+	if err != nil {
			151	+	t.Fatal(err)
			152	+	}
			153	+	if result.Status != BatchSucceeded \|\| len(result.ChildIDs) != len(tasks) \|\| len(result.Results) != len(tasks) {
			154	+	t.Fatalf("batch result = %#v", result)
			155	+	}
			156	+	}

M packages/agent/subagents/model.go +0 -4


@@ -53,7 +53,6 @@ const (
53	53		type SubagentPolicy struct {
54	54		MaxConcurrent int
55	55		MaxConcurrentPerParent int
56		-	MaxTotalSpawned int
57	56		QueueTimeout time.Duration
58	57		DefaultTimeout time.Duration
59	58		MaxTurns int

@@ -74,9 +73,6 @@ func (p *SubagentPolicy) normalize() {
74	73		if p.MaxConcurrentPerParent <= 0 {
75	74		p.MaxConcurrentPerParent = 4
76	75		}
77		-	if p.MaxTotalSpawned <= 0 {
78		-	p.MaxTotalSpawned = 32
79		-	}
80	76		if p.QueueTimeout <= 0 {
81	77		p.QueueTimeout = 30 * time.Minute
82	78		}


@@ -53,7 +53,6 @@ const (
53		type SubagentPolicy struct {	53	type SubagentPolicy struct {
54		MaxConcurrent int	54	MaxConcurrent int
55		MaxConcurrentPerParent int	55	MaxConcurrentPerParent int
56	-	MaxTotalSpawned int
57		QueueTimeout time.Duration	56	QueueTimeout time.Duration
58		DefaultTimeout time.Duration	57	DefaultTimeout time.Duration
59		MaxTurns int	58	MaxTurns int

@@ -74,9 +73,6 @@ func (p *SubagentPolicy) normalize() {
74		if p.MaxConcurrentPerParent <= 0 {	73	if p.MaxConcurrentPerParent <= 0 {
75		p.MaxConcurrentPerParent = 4	74	p.MaxConcurrentPerParent = 4
76		}	75	}
77	-	if p.MaxTotalSpawned <= 0 {
78	-	p.MaxTotalSpawned = 32
79	-	}
80		if p.QueueTimeout <= 0 {	76	if p.QueueTimeout <= 0 {
81		p.QueueTimeout = 30 * time.Minute	77	p.QueueTimeout = 30 * time.Minute
82		}	78	}

M packages/agent/subagents/scheduler_test.go +24 -13


@@ -2,9 +2,9 @@ package subagents
2	2
3	3		import (
4	4		"context"
	5	+	"fmt"
5	6		"os"
6	7		"path/filepath"
7		-	"strings"
8	8		"sync/atomic"
9	9		"testing"
10	10		"time"

@@ -32,7 +32,7 @@ func TestSchedulerBoundsConcurrencyAndReleasesCapacity(t *testing.T) {
32	32		var active, maxActive int32
33	33		f := New(Config{
34	34		Root: root, RepoRoot: root,
35		-	Policy: SubagentPolicy{MaxConcurrent: 1, MaxConcurrentPerParent: 4, MaxTotalSpawned: 3, QueueTimeout: time.Minute, DefaultTimeout: 0, IdleTimeout: time.Hour},
	35	+	Policy: SubagentPolicy{MaxConcurrent: 1, MaxConcurrentPerParent: 4, QueueTimeout: time.Minute, DefaultTimeout: 0, IdleTimeout: time.Hour},
36	36		NewRunner: func(a *Agent) Runner {
37	37		return RunnerFunc(func(ctx context.Context, _ Sink) error {
38	38		current := atomic.AddInt32(&active, 1)

@@ -95,20 +95,31 @@ func TestSchedulerBoundsConcurrencyAndReleasesCapacity(t *testing.T) {
95	95		}
96	96		}
97	97
98		-	func TestSchedulerRejectsChildSpawningAndTotalBudget(t *testing.T) {
99		-	f := New(Config{Root: t.TempDir(), RepoRoot: t.TempDir(), Policy: SubagentPolicy{MaxTotalSpawned: 1}, NewRunner: func(*Agent) Runner {
	98	+	func TestSchedulerRejectsChildSpawning(t *testing.T) {
	99	+	f := New(Config{Root: t.TempDir(), RepoRoot: t.TempDir(), NewRunner: func(*Agent) Runner {
100	100		return RunnerFunc(func(context.Context, Sink) error { return nil })
101	101		}})
102	102		if _, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "child", RequesterAgentID: "worker-1"}); err == nil {
103	103		t.Fatal("child-originated spawn was accepted")
104	104		}
105		-	first, err := f.Spawn(context.Background(), "one")
106		-	if err != nil {
107		-	t.Fatal(err)
108		-	}
109		-	first.Wait()
110		-	if _, err := f.Spawn(context.Background(), "two"); err == nil \|\| !strings.Contains(err.Error(), "budget") {
111		-	t.Fatalf("second spawn error = %v, want budget rejection", err)
	105	+	}
	106	+
	107	+	func TestSchedulerAllowsWorkersAfterManyCompletions(t *testing.T) {
	108	+	f := New(Config{Root: t.TempDir(), RepoRoot: t.TempDir(), Policy: SubagentPolicy{MaxConcurrent: 1}, NewRunner: func(*Agent) Runner {
	109	+	return RunnerFunc(func(context.Context, Sink) error { return nil })
	110	+	}})
	111	+	t.Cleanup(f.StopAll)
	112	+
	113	+	const workersBeyondFormerLifetimeLimit = 33
	114	+	for worker := 0; worker < workersBeyondFormerLifetimeLimit; worker++ {
	115	+	a, err := f.Spawn(context.Background(), fmt.Sprintf("worker-%d", worker))
	116	+	if err != nil {
	117	+	t.Fatalf("Spawn(worker-%d): %v", worker, err)
	118	+	}
	119	+	a.Wait()
	120	+	if got := a.Status(); got != StatusDone {
	121	+	t.Fatalf("worker-%d status = %s, want %s", worker, got, StatusDone)
	122	+	}
112	123		}
113	124		}
114	125

@@ -117,7 +128,7 @@ func TestStopQueuedAgentBeforeRunnerAdmission(t *testing.T) {
117	128		started := make(chan string, 2)
118	129		f := New(Config{
119	130		Root: root, RepoRoot: root,
120		-	Policy: SubagentPolicy{MaxConcurrent: 1, MaxTotalSpawned: 3, IdleTimeout: time.Hour},
	131	+	Policy: SubagentPolicy{MaxConcurrent: 1, IdleTimeout: time.Hour},
121	132		NewRunner: func(a *Agent) Runner {
122	133		return RunnerFunc(func(ctx context.Context, _ Sink) error {
123	134		started <- a.ID

@@ -165,7 +176,7 @@ func TestCanceledQueuedAgentIsRemovedAndGetsResult(t *testing.T) {
165	176		started := make(chan struct{})
166	177		f := New(Config{
167	178		Root: root, RepoRoot: root,
168		-	Policy: SubagentPolicy{MaxConcurrent: 1, MaxTotalSpawned: 4, IdleTimeout: time.Hour},
	179	+	Policy: SubagentPolicy{MaxConcurrent: 1, IdleTimeout: time.Hour},
169	180		NewRunner: func(*Agent) Runner {
170	181		return RunnerFunc(func(ctx context.Context, _ Sink) error {
171	182		select {


@@ -2,9 +2,9 @@ package subagents
2			2
3		import (	3		import (
4		"context"	4		"context"
			5	+	"fmt"
5		"os"	6		"os"
6		"path/filepath"	7		"path/filepath"
7	-	"strings"
8		"sync/atomic"	8		"sync/atomic"
9		"testing"	9		"testing"
10		"time"	10		"time"

@@ -32,7 +32,7 @@ func TestSchedulerBoundsConcurrencyAndReleasesCapacity(t *testing.T) {
32		var active, maxActive int32	32		var active, maxActive int32
33		f := New(Config{	33		f := New(Config{
34		Root: root, RepoRoot: root,	34		Root: root, RepoRoot: root,
35	-	Policy: SubagentPolicy{MaxConcurrent: 1, MaxConcurrentPerParent: 4, MaxTotalSpawned: 3, QueueTimeout: time.Minute, DefaultTimeout: 0, IdleTimeout: time.Hour},	35	+	Policy: SubagentPolicy{MaxConcurrent: 1, MaxConcurrentPerParent: 4, QueueTimeout: time.Minute, DefaultTimeout: 0, IdleTimeout: time.Hour},
36		NewRunner: func(a *Agent) Runner {	36		NewRunner: func(a *Agent) Runner {
37		return RunnerFunc(func(ctx context.Context, _ Sink) error {	37		return RunnerFunc(func(ctx context.Context, _ Sink) error {
38		current := atomic.AddInt32(&active, 1)	38		current := atomic.AddInt32(&active, 1)

@@ -95,20 +95,31 @@ func TestSchedulerBoundsConcurrencyAndReleasesCapacity(t *testing.T) {
95		}	95		}
96		}	96		}
97			97
98	-	func TestSchedulerRejectsChildSpawningAndTotalBudget(t *testing.T) {	98	+	func TestSchedulerRejectsChildSpawning(t *testing.T) {
99	-	f := New(Config{Root: t.TempDir(), RepoRoot: t.TempDir(), Policy: SubagentPolicy{MaxTotalSpawned: 1}, NewRunner: func(*Agent) Runner {	99	+	f := New(Config{Root: t.TempDir(), RepoRoot: t.TempDir(), NewRunner: func(*Agent) Runner {
100		return RunnerFunc(func(context.Context, Sink) error { return nil })	100		return RunnerFunc(func(context.Context, Sink) error { return nil })
101		}})	101		}})
102		if _, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "child", RequesterAgentID: "worker-1"}); err == nil {	102		if _, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "child", RequesterAgentID: "worker-1"}); err == nil {
103		t.Fatal("child-originated spawn was accepted")	103		t.Fatal("child-originated spawn was accepted")
104		}	104		}
105	-	first, err := f.Spawn(context.Background(), "one")	105	+	}
106	-	if err != nil {	106	+
107	-	t.Fatal(err)	107	+	func TestSchedulerAllowsWorkersAfterManyCompletions(t *testing.T) {
108	-	}	108	+	f := New(Config{Root: t.TempDir(), RepoRoot: t.TempDir(), Policy: SubagentPolicy{MaxConcurrent: 1}, NewRunner: func(*Agent) Runner {
109	-	first.Wait()	109	+	return RunnerFunc(func(context.Context, Sink) error { return nil })
110	-	if _, err := f.Spawn(context.Background(), "two"); err == nil \|\| !strings.Contains(err.Error(), "budget") {	110	+	}})
111	-	t.Fatalf("second spawn error = %v, want budget rejection", err)	111	+	t.Cleanup(f.StopAll)
			112	+
			113	+	const workersBeyondFormerLifetimeLimit = 33
			114	+	for worker := 0; worker < workersBeyondFormerLifetimeLimit; worker++ {
			115	+	a, err := f.Spawn(context.Background(), fmt.Sprintf("worker-%d", worker))
			116	+	if err != nil {
			117	+	t.Fatalf("Spawn(worker-%d): %v", worker, err)
			118	+	}
			119	+	a.Wait()
			120	+	if got := a.Status(); got != StatusDone {
			121	+	t.Fatalf("worker-%d status = %s, want %s", worker, got, StatusDone)
			122	+	}
112		}	123		}
113		}	124		}
114			125

@@ -117,7 +128,7 @@ func TestStopQueuedAgentBeforeRunnerAdmission(t *testing.T) {
117		started := make(chan string, 2)	128		started := make(chan string, 2)
118		f := New(Config{	129		f := New(Config{
119		Root: root, RepoRoot: root,	130		Root: root, RepoRoot: root,
120	-	Policy: SubagentPolicy{MaxConcurrent: 1, MaxTotalSpawned: 3, IdleTimeout: time.Hour},	131	+	Policy: SubagentPolicy{MaxConcurrent: 1, IdleTimeout: time.Hour},
121		NewRunner: func(a *Agent) Runner {	132		NewRunner: func(a *Agent) Runner {
122		return RunnerFunc(func(ctx context.Context, _ Sink) error {	133		return RunnerFunc(func(ctx context.Context, _ Sink) error {
123		started <- a.ID	134		started <- a.ID

@@ -165,7 +176,7 @@ func TestCanceledQueuedAgentIsRemovedAndGetsResult(t *testing.T) {
165		started := make(chan struct{})	176		started := make(chan struct{})
166		f := New(Config{	177		f := New(Config{
167		Root: root, RepoRoot: root,	178		Root: root, RepoRoot: root,
168	-	Policy: SubagentPolicy{MaxConcurrent: 1, MaxTotalSpawned: 4, IdleTimeout: time.Hour},	179	+	Policy: SubagentPolicy{MaxConcurrent: 1, IdleTimeout: time.Hour},
169		NewRunner: func(*Agent) Runner {	180		NewRunner: func(*Agent) Runner {
170		return RunnerFunc(func(ctx context.Context, _ Sink) error {	181		return RunnerFunc(func(ctx context.Context, _ Sink) error {
171		select {	182		select {

M packages/agent/subagents/supervisor.go +0 -17


@@ -127,7 +127,6 @@ type Supervisor struct {
127	127		active int
128	128		activeByParent map[string]int
129	129		activeByBatch map[string]int
130		-	totalSpawned int
131	130		batches map[string]*Batch
132	131
133	132		// activeSession is the host session id the dashboard is

@@ -369,15 +368,8 @@ func (f Supervisor) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, er
369	368		now := f.cfg.Now()
370	369		baseID := newAgentID(task, now)
371	370
372		-	// Reserve a spawn budget before touching disk. This makes the total
373		-	// budget race-free across slash commands, tools, and batch callers.
374	371		f.mu.Lock()
375	372		id := f.uniqueAgentIDLocked(baseID)
376		-	if f.totalSpawned >= f.cfg.Policy.MaxTotalSpawned {
377		-	f.mu.Unlock()
378		-	return nil, fmt.Errorf("subagents: total spawn budget exhausted (%d)", f.cfg.Policy.MaxTotalSpawned)
379		-	}
380		-	f.totalSpawned++
381	373		dir := f.cfg.RepoRoot
382	374		fastMode := f.cfg.FastMode
383	375		configProvider := f.cfg.Provider

@@ -397,14 +389,6 @@ func (f Supervisor) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, er
397	389		baseURL = strings.TrimSpace(configBaseURL)
398	390		}
399	391		insecureTLS := req.InsecureTLS \|\| (inheritProviderSettings && configInsecureTLS)
400		-	reserved := true
401		-	defer func() {
402		-	if reserved {
403		-	f.mu.Lock()
404		-	f.totalSpawned--
405		-	f.mu.Unlock()
406		-	}
407		-	}()
408	392
409	393		stateDir := f.agentStateDir(id)
410	394		if err := os.MkdirAll(stateDir, 0o700); err != nil {

@@ -542,7 +526,6 @@ func (f Supervisor) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, er
542	526		// supervisor can reconstruct work that never reached a process.
543	527		f.armQueueTimeout(a)
544	528		f.schedule()
545		-	reserved = false
546	529		return a, nil
547	530		}
548	531


@@ -127,7 +127,6 @@ type Supervisor struct {
127		active int	127	active int
128		activeByParent map[string]int	128	activeByParent map[string]int
129		activeByBatch map[string]int	129	activeByBatch map[string]int
130	-	totalSpawned int
131		batches map[string]*Batch	130	batches map[string]*Batch
132			131
133		// activeSession is the host session id the dashboard is	132	// activeSession is the host session id the dashboard is

@@ -369,15 +368,8 @@ func (f Supervisor) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, er
369		now := f.cfg.Now()	368	now := f.cfg.Now()
370		baseID := newAgentID(task, now)	369	baseID := newAgentID(task, now)
371			370
372	-	// Reserve a spawn budget before touching disk. This makes the total
373	-	// budget race-free across slash commands, tools, and batch callers.
374		f.mu.Lock()	371	f.mu.Lock()
375		id := f.uniqueAgentIDLocked(baseID)	372	id := f.uniqueAgentIDLocked(baseID)
376	-	if f.totalSpawned >= f.cfg.Policy.MaxTotalSpawned {
377	-	f.mu.Unlock()
378	-	return nil, fmt.Errorf("subagents: total spawn budget exhausted (%d)", f.cfg.Policy.MaxTotalSpawned)
379	-	}
380	-	f.totalSpawned++
381		dir := f.cfg.RepoRoot	373	dir := f.cfg.RepoRoot
382		fastMode := f.cfg.FastMode	374	fastMode := f.cfg.FastMode
383		configProvider := f.cfg.Provider	375	configProvider := f.cfg.Provider

@@ -397,14 +389,6 @@ func (f Supervisor) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, er
397		baseURL = strings.TrimSpace(configBaseURL)	389	baseURL = strings.TrimSpace(configBaseURL)
398		}	390	}
399		insecureTLS := req.InsecureTLS \|\| (inheritProviderSettings && configInsecureTLS)	391	insecureTLS := req.InsecureTLS \|\| (inheritProviderSettings && configInsecureTLS)
400	-	reserved := true
401	-	defer func() {
402	-	if reserved {
403	-	f.mu.Lock()
404	-	f.totalSpawned--
405	-	f.mu.Unlock()
406	-	}
407	-	}()
408			392
409		stateDir := f.agentStateDir(id)	393	stateDir := f.agentStateDir(id)
410		if err := os.MkdirAll(stateDir, 0o700); err != nil {	394	if err := os.MkdirAll(stateDir, 0o700); err != nil {

@@ -542,7 +526,6 @@ func (f Supervisor) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, er
542		// supervisor can reconstruct work that never reached a process.	526	// supervisor can reconstruct work that never reached a process.
543		f.armQueueTimeout(a)	527	f.armQueueTimeout(a)
544		f.schedule()	528	f.schedule()
545	-	reserved = false
546		return a, nil	529	return a, nil
547		}	530	}
548			531