@ours.network/fleet 0.17.7 → 0.18.0-nightly.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/runner.js CHANGED
@@ -59,15 +59,10 @@ const defaultDeps = () => ({
59
59
  },
60
60
  });
61
61
  const MONITOR_OWNER_FILE = '.monitor-owner';
62
- /** Fleet roles consume the operator-owned daemon; a role session never starts it. */
63
- const FLEET_OURS_AUTOSTART = '0';
64
62
  /** Environment injected only into the managed harness process. */
65
63
  export function managedFleetProxyEnv(role, stateDir) {
66
64
  return {
67
65
  ...(role.env ?? {}),
68
- // This must win over both inherited/configured auto-start. ACP agents run
69
- // directly rather than through ours-codex, so the runner owns this fence.
70
- OURS_AUTOSTART: FLEET_OURS_AUTOSTART,
71
66
  [FLEET_PROXY_STATE_DIR_ENV]: stateDir,
72
67
  [FLEET_PROXY_CALLER_ENV]: role.name,
73
68
  };
@@ -111,14 +106,11 @@ async function executeManagedSpawn(caller, configPath, requested, log) {
111
106
  session: preview.session,
112
107
  ...(preview.model ? { model: preview.model } : {}),
113
108
  monitor: { mode: preview.monitor.mode, interrupt: preview.monitor.interrupt },
114
- permissionMode: effectivePermissionMode(preview),
115
109
  inherited,
116
110
  creationActionId,
117
111
  };
118
112
  log(`[${caller.name}] managed fleet proxy spawned ${result.lifetime} role ${result.role} `
119
- + `harness=${result.harness} session=${result.session} `
120
- + `permission=${result.permissionMode.fleetMode} `
121
- + `native=${result.permissionMode.nativeMode}`);
113
+ + `harness=${result.harness} session=${result.session}`);
122
114
  return result;
123
115
  }
124
116
  /**
@@ -148,10 +140,6 @@ export function recordMonitorOwner(dir, owner) {
148
140
  export function buildPaneCommand(launch, roleEnv, exitStatusPath, paneArgv = launch.argv) {
149
141
  const env = {
150
142
  PATH: process.env.PATH ?? '', COLORTERM: 'truecolor', ...launch.env, ...(roleEnv ?? {}),
151
- // Tmux roles have the same daemon-client boundary as ACP roles. Keep this
152
- // last so neither harness preparation nor a role env block can take over
153
- // the shared daemon lifecycle.
154
- OURS_AUTOSTART: FLEET_OURS_AUTOSTART,
155
143
  };
156
144
  // Interactive panes should advertise colour even when the supervisor itself
157
145
  // was launched with NO_COLOR. A role may still deliberately opt back in to
@@ -542,10 +530,6 @@ export async function runOnce(name, opts = {}, partialDeps = {}) {
542
530
  permissions: perms,
543
531
  modeId: adapter.acpPermissionModeId?.(role),
544
532
  permissionMode: effectivePermissionMode(role),
545
- // Provenance travels with the exact ACP launch. Keeping it out of a
546
- // role-only adapter hook prevents a PATH fallback or resolver skew from
547
- // claiming metadata trust for an argv it did not authenticate.
548
- permissionMetadataSource: launch.permissionMetadataSource,
549
533
  log: deps.log,
550
534
  });
551
535
  pid = acpSession.pid;
@@ -993,16 +977,8 @@ export async function runSupervised(name, opts = {}, partialDeps = {}, attempt =
993
977
  continue;
994
978
  }
995
979
  const fastFailSecs = fastFailSecsFor(name, opts.configPath);
996
- // The fast-fail boundary starts a recovery episode; it must not also be
997
- // the boundary that declares recovery successful. Otherwise alternating
998
- // 19s and 20s deaths erase one another forever. Require the configured
999
- // number of fast-fail windows to survive before closing an active streak.
1000
- // This hysteresis stays adapter-relative (100s for the current 20s/5-attempt
1001
- // policy) and still lets a genuinely sustained session reset the breaker.
1002
- const stableRecoverySecs = fastFailSecs * RESTART_FAIL_THRESHOLD;
1003
- const recoveryFailed = result.elapsedSecs < fastFailSecs
1004
- || (ledger.consecutiveImmediateFailures > 0 && result.elapsedSecs < stableRecoverySecs);
1005
- if (!recoveryFailed) {
980
+ const immediate = result.elapsedSecs < fastFailSecs;
981
+ if (!immediate) {
1006
982
  // A session that ran for a while is not a restart loop, whatever ended it.
1007
983
  writeRestartLedger(dir, {
1008
984
  ...emptyLedger(),
@@ -5,21 +5,6 @@ import type { ConversationSnapshot, PromptOrigin, PromptReceipt, SubmitPromptCom
5
5
  import type { ConversationHandlePage, ExitRecord, InterruptOutcome, QueuedPrompt, SessionEvent, RuntimeSelectorMetadata, SessionHandle, SessionSnapshot, SubmitPromptOptions, TurnCancellationSource, TurnOutcome, TurnResult } from './types.js';
6
6
  /** Bound safe-boundary waiting without turning a hung tool into cancellation. */
7
7
  export declare const AFTER_TOOL_BOUNDARY_TIMEOUT_MS = 120000;
8
- /**
9
- * How long a steering-started turn is presumed to still own the adapter after
10
- * its last update. Such a turn has no prompt id, so it never reports a
11
- * stopReason and there is no exact end to observe — silence is the only signal
12
- * available, and this is the bound that turns it into a decision.
13
- *
14
- * Sized from the fleet's own scheduled-run history: across 1513 completed
15
- * scheduled runs the longest silence WITHIN a working turn was 120.2 s (p99
16
- * 41.0 s; 5 runs above 60 s). A shorter grace would release the lease while the
17
- * adapter is still working and re-admit a prompt into a busy turn, which is the
18
- * FLEET-003 failure itself. The costs are deliberately asymmetric: holding too
19
- * long skips one best-effort maintenance tick, releasing too early SIGTERMs a
20
- * live role.
21
- */
22
- export declare const STEERING_OCCUPANCY_IDLE_MS = 150000;
23
8
  /** Server-generated typed provenance followed by the exact human-authored body. */
24
9
  export declare function promptContentBlocks(text: string, origin?: PromptOrigin): acp.ContentBlock[];
25
10
  export declare function runtimeSelector(options: acp.SessionConfigOption[] | null | undefined, category: string): RuntimeSelectorMetadata | undefined;
@@ -35,8 +20,6 @@ export interface AcpSessionOptions {
35
20
  modeId?: string;
36
21
  /** Adapter-resolved live permission policy; separate from ACP agent-specific session modes. */
37
22
  permissionMode?: NonNullable<SessionSnapshot['permissionMode']>;
38
- /** Adapter-authenticated request-metadata vocabulary; never inferred from ACP `_meta`. */
39
- permissionMetadataSource?: 'codex-acp';
40
23
  log(line: string): void;
41
24
  /** Test seam for the cancel-escalation grace period; production uses the default. */
42
25
  cancelGraceMs?: number;
@@ -48,8 +31,6 @@ export interface AcpSessionOptions {
48
31
  controllerGraceMs?: number;
49
32
  /** Test seam; production uses AFTER_TOOL_BOUNDARY_TIMEOUT_MS. */
50
33
  afterToolBoundaryTimeoutMs?: number;
51
- /** Test seam; production uses STEERING_OCCUPANCY_IDLE_MS. */
52
- steeringOccupancyIdleMs?: number;
53
34
  }
54
35
  /**
55
36
  * Classify an ACP `stopReason` into a terminal outcome. A refusal and a
@@ -77,12 +58,6 @@ export declare class AcpSession implements SessionHandle {
77
58
  private connection;
78
59
  private sessionId?;
79
60
  private readiness;
80
- /**
81
- * Last non-replayed session update from the agent. `readiness` cannot answer
82
- * "is this agent working" for a steered turn (FLEET-002), and this is the
83
- * evidence that can.
84
- */
85
- private lastUpdateAt?;
86
61
  private lastError?;
87
62
  private promptTail;
88
63
  private queueDepth;
@@ -98,13 +73,6 @@ export declare class AcpSession implements SessionHandle {
98
73
  private cancelEscalation?;
99
74
  private cancelForceKill?;
100
75
  private cancelRecoveryReason?;
101
- /**
102
- * Held while a steering-started turn is believed to own the adapter. It is a
103
- * lease, not a latch: `steeringRelease` always fires, so the role can never be
104
- * stranded busy by a wake whose turn ended without telling anyone.
105
- */
106
- private steeringOccupied;
107
- private steeringRelease?;
108
76
  /**
109
77
  * Rejects the moment the adapter process is gone. Every in-flight ACP request
110
78
  * races it, so a dead adapter can never leave a turn — and therefore a
@@ -126,20 +94,6 @@ export declare class AcpSession implements SessionHandle {
126
94
  */
127
95
  private recoverOpenPrompts;
128
96
  isAlive(): boolean;
129
- /**
130
- * Take the occupancy lease for a turn the adapter started on its own behalf.
131
- * Refreshed by every adapter update, so it tracks work actually happening
132
- * rather than a fixed guess at how long a wake takes.
133
- */
134
- private holdSteeringOccupancy;
135
- private refreshSteeringOccupancy;
136
- /**
137
- * Every exit from occupancy comes through here, including the ones that are
138
- * not the timer: a real turn boundary, close, and adapter exit. A lease that
139
- * can leak is worse than the bug it fixes — it would leave the role reporting
140
- * `running` forever and starve scheduled admission permanently.
141
- */
142
- private releaseSteeringOccupancy;
143
97
  snapshot(): SessionSnapshot;
144
98
  private toolCall;
145
99
  private reserveTool;
@@ -231,15 +185,6 @@ export declare class AcpSession implements SessionHandle {
231
185
  */
232
186
  private settleAutomatically;
233
187
  private withinAutomaticBoundary;
234
- /**
235
- * Codex ACP 1.1.7 marks its protected MCP elicitation bridge on a locationless
236
- * execute request. The marker is meaningful only together with the runner's
237
- * independently supplied, adapter-authenticated metadata vocabulary and effective
238
- * mode: an arbitrary ACP process cannot gain this path by copying `_meta` alone.
239
- * Exact option ids/kinds bind recognition to the protected-MCP shape and keep
240
- * malformed requests on the ordinary fail-closed path.
241
- */
242
- private isEffectiveCodexProtectedMcpApproval;
243
188
  private recordUpdate;
244
189
  /**
245
190
  * Codex ACP's phase extension is the only currently supported visibility
@@ -16,21 +16,6 @@ const PERMISSION_TIMEOUT_MS = 10 * 60_000;
16
16
  const CONTROLLER_GRACE_MS = 12_000;
17
17
  /** Bound safe-boundary waiting without turning a hung tool into cancellation. */
18
18
  export const AFTER_TOOL_BOUNDARY_TIMEOUT_MS = 120_000;
19
- /**
20
- * How long a steering-started turn is presumed to still own the adapter after
21
- * its last update. Such a turn has no prompt id, so it never reports a
22
- * stopReason and there is no exact end to observe — silence is the only signal
23
- * available, and this is the bound that turns it into a decision.
24
- *
25
- * Sized from the fleet's own scheduled-run history: across 1513 completed
26
- * scheduled runs the longest silence WITHIN a working turn was 120.2 s (p99
27
- * 41.0 s; 5 runs above 60 s). A shorter grace would release the lease while the
28
- * adapter is still working and re-admit a prompt into a busy turn, which is the
29
- * FLEET-003 failure itself. The costs are deliberately asymmetric: holding too
30
- * long skips one best-effort maintenance tick, releasing too early SIGTERMs a
31
- * live role.
32
- */
33
- export const STEERING_OCCUPANCY_IDLE_MS = 150_000;
34
19
  const TERMINAL_TOOL_STATUSES = new Set(['completed', 'failed']);
35
20
  const SCHEDULED_LOOP_REDACTION = '[scheduled-loop content redacted]';
36
21
  const OWNER_COMMENTARY_REDACTION = '[assistant commentary redacted]';
@@ -217,12 +202,6 @@ export class AcpSession {
217
202
  connection;
218
203
  sessionId;
219
204
  readiness = 'starting';
220
- /**
221
- * Last non-replayed session update from the agent. `readiness` cannot answer
222
- * "is this agent working" for a steered turn (FLEET-002), and this is the
223
- * evidence that can.
224
- */
225
- lastUpdateAt;
226
205
  lastError;
227
206
  promptTail = Promise.resolve();
228
207
  queueDepth = 0;
@@ -238,13 +217,6 @@ export class AcpSession {
238
217
  cancelEscalation;
239
218
  cancelForceKill;
240
219
  cancelRecoveryReason;
241
- /**
242
- * Held while a steering-started turn is believed to own the adapter. It is a
243
- * lease, not a latch: `steeringRelease` always fires, so the role can never be
244
- * stranded busy by a wake whose turn ended without telling anyone.
245
- */
246
- steeringOccupied = false;
247
- steeringRelease;
248
220
  /**
249
221
  * Rejects the moment the adapter process is gone. Every in-flight ACP request
250
222
  * races it, so a dead adapter can never leave a turn — and therefore a
@@ -275,7 +247,6 @@ export class AcpSession {
275
247
  if (this.cancelForceKill)
276
248
  clearTimeout(this.cancelForceKill);
277
249
  this.cancelForceKill = undefined;
278
- this.releaseSteeringOccupancy('adapter exited');
279
250
  // Record the child's real exit code/signal. The tmux path can only see a
280
251
  // shell's `$?`; here the truth is available, so keep it.
281
252
  const classified = classifyChildExit(code, signal);
@@ -376,60 +347,17 @@ export class AcpSession {
376
347
  // terminal fact (signal exits deliberately leave exitCode null).
377
348
  return this.child.exitCode === null && (this.child.signalCode ?? null) === null;
378
349
  }
379
- /**
380
- * Take the occupancy lease for a turn the adapter started on its own behalf.
381
- * Refreshed by every adapter update, so it tracks work actually happening
382
- * rather than a fixed guess at how long a wake takes.
383
- */
384
- holdSteeringOccupancy() {
385
- if (this.closing || !this.isAlive())
386
- return;
387
- this.steeringOccupied = true;
388
- this.refreshSteeringOccupancy();
389
- }
390
- refreshSteeringOccupancy() {
391
- if (!this.steeringOccupied)
392
- return;
393
- if (this.steeringRelease)
394
- clearTimeout(this.steeringRelease);
395
- this.steeringRelease = setTimeout(() => this.releaseSteeringOccupancy('adapter silent'), this.options.steeringOccupancyIdleMs ?? STEERING_OCCUPANCY_IDLE_MS);
396
- this.steeringRelease.unref?.();
397
- }
398
- /**
399
- * Every exit from occupancy comes through here, including the ones that are
400
- * not the timer: a real turn boundary, close, and adapter exit. A lease that
401
- * can leak is worse than the bug it fixes — it would leave the role reporting
402
- * `running` forever and starve scheduled admission permanently.
403
- */
404
- releaseSteeringOccupancy(reason) {
405
- if (this.steeringRelease)
406
- clearTimeout(this.steeringRelease);
407
- this.steeringRelease = undefined;
408
- if (!this.steeringOccupied)
409
- return;
410
- this.steeringOccupied = false;
411
- this.options.log(`[${this.options.name}] steering-started turn no longer holds the adapter (${reason})`);
412
- }
413
350
  snapshot() {
414
351
  return {
415
352
  backend: 'acp',
416
353
  alive: this.isAlive(),
417
- // A steering-started turn is real work with no prompt id. Reporting the
418
- // session idle while it runs is what let the arbiter admit a scheduled
419
- // prompt into a busy adapter, whose `session/prompt` then never returned
420
- // a stopReason and ended in a cancellation deadline and a SIGTERM.
421
- readiness: this.readiness === 'idle' && this.steeringOccupied
422
- ? 'running' : this.readiness,
354
+ readiness: this.readiness,
423
355
  sessionId: this.sessionId,
424
356
  lastError: this.lastError,
425
357
  pendingPermissionId: this.pendingPermissions.keys().next().value,
426
358
  runtimeModel: this.runtimeModel,
427
359
  reasoningEffort: this.reasoningEffort,
428
360
  permissionMode: this.options.permissionMode,
429
- activity: {
430
- activeToolCalls: this.activeToolCalls.size,
431
- ...(this.lastUpdateAt ? { lastUpdateAt: this.lastUpdateAt } : {}),
432
- },
433
361
  };
434
362
  }
435
363
  toolCall(toolCallId) {
@@ -927,7 +855,6 @@ export class AcpSession {
927
855
  if (this.controllerGrace)
928
856
  clearTimeout(this.controllerGrace);
929
857
  this.controllerGrace = undefined;
930
- this.releaseSteeringOccupancy('session closed');
931
858
  for (const [permissionId, pending] of [...this.pendingPermissions])
932
859
  this.settlePendingAutomatically(permissionId, pending, 'cancelled', undefined, 'the session closed while this request was pending');
933
860
  this.releaseAllTools();
@@ -1086,10 +1013,6 @@ export class AcpSession {
1086
1013
  }
1087
1014
  finally {
1088
1015
  this.releaseAllTools();
1089
- // A turn this client owned has ended, so the adapter has reported a
1090
- // boundary: whatever a steering call started before it is over too. This
1091
- // is the release path that does not depend on the silence timer.
1092
- this.releaseSteeringOccupancy('turn boundary');
1093
1016
  if (this.activeTurn?.id === turnId) {
1094
1017
  this.activeTurn.settle();
1095
1018
  if (this.cancelEscalation)
@@ -1112,12 +1035,6 @@ export class AcpSession {
1112
1035
  ]);
1113
1036
  if (response.outcome === 'failed')
1114
1037
  return turnResult(false, 'failed', 'ACP steering failed');
1115
- // `injected` joined a turn this client already owns and will settle.
1116
- // `startedNewTurn` created one nobody owns: the adapter is working and
1117
- // will never answer for it, so admission has to learn about it here or
1118
- // not at all.
1119
- if (response.outcome === 'startedNewTurn')
1120
- this.holdSteeringOccupancy();
1121
1038
  return turnResult(true, 'inconclusive', response.outcome);
1122
1039
  }
1123
1040
  catch (error) {
@@ -1145,17 +1062,6 @@ export class AcpSession {
1145
1062
  // Permission is part of the tool lifecycle. Reserve before any policy or
1146
1063
  // human decision so a monitor wake cannot slip between request and answer.
1147
1064
  this.reservePermission(toolCallId, permissionId);
1148
- if (this.isEffectiveCodexProtectedMcpApproval(params)) {
1149
- // Protected MCP approval is already the tool's narrow gate. Never turn
1150
- // this one decision into an adapter-wide standing grant.
1151
- const option = choose(['allow_once']);
1152
- const response = this.settleAutomatically(params, option, 'allowed', 'permissionMode.fleetMode=allow', 'the trusted Codex adapter authenticated a protected MCP approval request');
1153
- if (option)
1154
- this.allowPermission(toolCallId, permissionId);
1155
- else
1156
- this.releasePermission(toolCallId, permissionId);
1157
- return Promise.resolve(response);
1158
- }
1159
1065
  if (this.options.permissions.approval === 'allow' && this.withinAutomaticBoundary(params)) {
1160
1066
  const option = choose(['allow_always', 'allow_once']);
1161
1067
  const response = this.settleAutomatically(params, option, 'allowed', 'permissions.approval=allow', `the request is inside the ${this.options.permissions.filesystem} boundary`);
@@ -1282,34 +1188,7 @@ export class AcpSession {
1282
1188
  const cwd = resolve(this.options.cwd);
1283
1189
  return canonicallyWithin(cwd, locations.map(location => resolve(location.path)));
1284
1190
  }
1285
- /**
1286
- * Codex ACP 1.1.7 marks its protected MCP elicitation bridge on a locationless
1287
- * execute request. The marker is meaningful only together with the runner's
1288
- * independently supplied, adapter-authenticated metadata vocabulary and effective
1289
- * mode: an arbitrary ACP process cannot gain this path by copying `_meta` alone.
1290
- * Exact option ids/kinds bind recognition to the protected-MCP shape and keep
1291
- * malformed requests on the ordinary fail-closed path.
1292
- */
1293
- isEffectiveCodexProtectedMcpApproval(params) {
1294
- const locations = params.toolCall.locations ?? [];
1295
- return this.options.permissionMetadataSource === 'codex-acp'
1296
- && this.options.permissionMode?.fleetMode === 'allow'
1297
- && params.toolCall.kind === 'execute'
1298
- && params.toolCall.status === 'pending'
1299
- && locations.length === 0
1300
- && params._meta?.is_mcp_tool_approval === true
1301
- && params.options.some(option => option.optionId === 'allow_once' && option.kind === 'allow_once')
1302
- && params.options.some(option => option.optionId === 'decline' && option.kind === 'reject_once');
1303
- }
1304
1191
  recordUpdate(update) {
1305
- // Replayed history is not current activity: `session/load` would otherwise
1306
- // make a cold session look like it had just been working. The same reason
1307
- // keeps it from extending the steering lease, which is evidence the adapter
1308
- // is working right now — for a steering-started turn, the only evidence.
1309
- if (!this.replaying) {
1310
- this.lastUpdateAt = new Date().toISOString();
1311
- this.refreshSteeringOccupancy();
1312
- }
1313
1192
  const scheduled = this.activeTurn?.origin?.kind === 'scheduled-loop';
1314
1193
  const messagePhase = update.sessionUpdate === 'agent_message_chunk'
1315
1194
  ? this.codexMessagePhase(update) : undefined;
@@ -1,16 +1,5 @@
1
1
  import type { SessionBackendId } from '../config.js';
2
2
  import type { ConversationEventV1, ConversationSnapshot, PromptReceipt, SubmitPromptCommand } from './conversation-types.js';
3
- /**
4
- * TURN OCCUPANCY, and nothing else: `idle` means no fleet-tracked turn is in
5
- * flight, which is exactly the question `arbiter.tryScheduled` asks before it
6
- * admits a prompt. It is NOT a claim that the agent is doing nothing — a wake
7
- * delivered through the `_session/steering` extension answers `startedNewTurn`
8
- * and runs a whole turn that fleet never gets a `session/prompt` response for
9
- * (ACP has no turn-end session update), so `readiness` stays `idle` for its
10
- * entire duration. Anything reporting activity or liveness to a human must
11
- * corroborate with `SessionSnapshot.activity` instead of reading `idle` here as
12
- * "not working".
13
- */
14
3
  export type SessionReadiness = 'starting' | 'idle' | 'running' | 'awaiting_permission' | 'failed';
15
4
  export type TurnOutcome = 'completed' | 'refused' | 'cancelled' | 'failed' | 'inconclusive';
16
5
  export type TurnCancellationSource = 'owner' | 'local-console' | 'fleet-monitor' | 'scheduled-loop' | 'shutdown';
@@ -181,19 +170,6 @@ export interface SessionSnapshot {
181
170
  /** Exact harness-native approval/permission mode used by this runner. */
182
171
  nativeMode: string;
183
172
  };
184
- /**
185
- * Observed agent activity, independent of turn occupancy: the evidence a
186
- * human-facing surface needs before calling a role idle. Absent on backends
187
- * that cannot observe the agent at all (tmux), which is itself honest — no
188
- * evidence is not evidence of inactivity.
189
- */
190
- activity?: SessionActivity;
191
- }
192
- export interface SessionActivity {
193
- /** ACP tool calls currently reserved (lifecycle open or permission pending). */
194
- activeToolCalls: number;
195
- /** When the agent last sent ANY session update, replay excluded. */
196
- lastUpdateAt?: string;
197
173
  }
198
174
  export type SessionEventKind = 'state' | 'agent_text' | 'thought' | 'tool_call' | 'tool_update' | 'permission' | 'monitor_delivery' | 'turn_stop' | 'error';
199
175
  /** What a settled permission request resolved to. */
@@ -69,6 +69,33 @@ export function makeSystemdBackend(exec = realExec) {
69
69
  dirname(process.execPath),
70
70
  ...(process.env.PATH ?? '').split(delimiter),
71
71
  ].filter(Boolean))].join(delimiter);
72
+ // Same reason as PATH, for the other thing a lingering unit cannot inherit:
73
+ // WHICH ours daemon this fleet was set up against.
74
+ //
75
+ // ours-fleet resolves its daemon from OURS_CONFIG / OURS_PORT / OURS_STATE_DIR
76
+ // and otherwise falls back to ~/.ours and port 3050 (src/monitor.ts). A host
77
+ // set up against a non-default daemon — the installer's multi-daemon profiles
78
+ // do exactly this — passes that selection to `ours-fleet init` in the
79
+ // environment, and it died there: nothing persisted it, so every runner
80
+ // systemd started at boot resolved the default daemon again, and on a host
81
+ // where only the non-default daemon exists that is a daemon that is not there.
82
+ //
83
+ // ONLY OURS_CONFIG is baked, deliberately. It names which daemon, and leaves
84
+ // that daemon's own config file authoritative for port and state directory, so
85
+ // a later edit to it still wins. Baking OURS_PORT/OURS_STATE_DIR would freeze
86
+ // those into a unit file that outranks the config for ever after — the failure
87
+ // mode @ours.network/cli avoids for the same reason (packages/cli/src/
88
+ // service.ts: "The port is deliberately NOT baked").
89
+ //
90
+ // Absent from init's environment ⇒ no line, and the unit is byte-for-byte what
91
+ // it has always been. This teaches fleet nothing about installer profiles; it
92
+ // persists the selection fleet was initialised with.
93
+ const unitEnv = ['PATH=' + servicePath];
94
+ if (process.env.OURS_CONFIG)
95
+ unitEnv.push('OURS_CONFIG=' + process.env.OURS_CONFIG);
96
+ const environmentLines = unitEnv
97
+ .map(value => `Environment="${value.replace(/\\/g, '\\\\').replace(/"/g, '\\"').replace(/%/g, '%%')}"`)
98
+ .join('\n');
72
99
  mkdirSync(unitDir, { recursive: true });
73
100
  writeFileSync(join(unitDir, UNIT_TEMPLATE), `[Unit]
74
101
  Description=ours-fleet agent %i
@@ -76,7 +103,7 @@ After=default.target
76
103
 
77
104
  [Service]
78
105
  Type=simple
79
- Environment="PATH=${servicePath.replace(/\\/g, '\\\\').replace(/"/g, '\\"').replace(/%/g, '%%')}"
106
+ ${environmentLines}
80
107
  ExecStart=${unitArg(process.execPath)} ${unitArg(binPath)} _run %i
81
108
  # The RUNNER owns the child-session restart loop, with a counted, backed-off
82
109
  # circuit breaker (3.2). systemd must only recover the runner PROCESS crashing —
@@ -80,13 +80,6 @@ export function generateWatchdogBriefing(opts) {
80
80
  L.push('- `healthy` — alive, on-briefing, recent progress.');
81
81
  L.push('- `idle` — alive, nothing assigned or nothing to do. Not an anomaly.');
82
82
  L.push('- `stale` = no worklog append and no console progress for ≥ 3 intervals.');
83
- L.push('');
84
- L.push('`session.readiness` from `ours-fleet status` is TURN OCCUPANCY, not activity: a mail');
85
- L.push('wake delivered by ACP steering runs an entire turn while readiness stays `idle`. Never');
86
- L.push('report `idle` or `stale` from `readiness=idle` alone — corroborate with the');
87
- L.push('`activity:` line of the same `status` output (`active` means the agent is working),');
88
- L.push('the worklog, or `ours-fleet peek`. `activity: unobservable` is missing evidence, not');
89
- L.push('an idle agent.');
90
83
  L.push('- `blocked` = waiting on a permission/prompt/modal longer than one interval.');
91
84
  L.push('- `off_briefing` — activity contradicts the briefing (wrong repo, out-of-scope work,');
92
85
  L.push(' ignored routine).');
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@ours.network/fleet",
3
- "version": "0.17.7",
3
+ "version": "0.18.0-nightly.2",
4
4
  "description": "Harness-agnostic fleet of persistent, identity-bound AI agents. Declarative fleet.yaml, tmux or ACP sessions, supervision, and ours.network messaging.",
5
5
  "type": "module",
6
6
  "license": "FSL-1.1-Apache-2.0",
@@ -38,6 +38,7 @@
38
38
  "@agentclientprotocol/sdk": "^1.3.0",
39
39
  "@fastify/static": "^10.1.2",
40
40
  "@fastify/websocket": "^11.2.0",
41
+ "@ours.network/sdk": "1.3.1",
41
42
  "@xterm/addon-fit": "0.10.0",
42
43
  "@xterm/addon-serialize": "0.13.0",
43
44
  "@xterm/headless": "5.5.0",
@@ -1,24 +0,0 @@
1
- export declare class OursMcpError extends Error {
2
- }
3
- export interface OursToolClient {
4
- start(): Promise<void>;
5
- callTool(name: string, args?: Record<string, unknown>): Promise<unknown>;
6
- close(): Promise<void>;
7
- }
8
- /** Minimal MCP stdio client. One instance owns exactly one ours identity binding. */
9
- export declare class OursMcpClient implements OursToolClient {
10
- private readonly command;
11
- private readonly env;
12
- private readonly log;
13
- private child?;
14
- private nextId;
15
- private tail;
16
- constructor(command?: string, env?: Record<string, string>, log?: (line: string) => void);
17
- start(): Promise<void>;
18
- callTool(name: string, args?: Record<string, unknown>): Promise<unknown>;
19
- close(): Promise<void>;
20
- private request;
21
- private requestNow;
22
- private notify;
23
- private write;
24
- }
@@ -1,145 +0,0 @@
1
- import { spawn } from 'node:child_process';
2
- import { randomUUID } from 'node:crypto';
3
- import { createInterface } from 'node:readline';
4
- export class OursMcpError extends Error {
5
- }
6
- /** Minimal MCP stdio client. One instance owns exactly one ours identity binding. */
7
- export class OursMcpClient {
8
- command;
9
- env;
10
- log;
11
- child;
12
- nextId = 0;
13
- tail = Promise.resolve();
14
- constructor(command = 'ours-mcp', env = {}, log = () => undefined) {
15
- this.command = command;
16
- this.env = env;
17
- this.log = log;
18
- }
19
- async start() {
20
- if (this.child && this.child.exitCode === null)
21
- return;
22
- // ours-mcp normally records the long-lived client PID so an identity lease
23
- // survives connector churn. An owner-channel connector has the opposite
24
- // lifecycle: each supervised attempt owns a fresh connector and must make
25
- // its lease reclaimable when that connector exits, even though the fleet
26
- // supervisor itself remains alive. POSIX exec preserves the shell PID as
27
- // the proxy PID, giving the daemon an exact process-lifetime fence without
28
- // interpolating the command path into shell text.
29
- const child = spawn('/bin/sh', [
30
- '-c', 'OURS_CLIENT_PID=$$; export OURS_CLIENT_PID; exec "$1" "$2"',
31
- 'ours-fleet-owner-proxy', this.command, 'proxy',
32
- ], {
33
- env: {
34
- ...process.env,
35
- ...this.env,
36
- // Bindings are keyed by this value. Sharing it would silently rebind a
37
- // role's normal mailbox or another owner channel.
38
- CLAUDE_CODE_SESSION_ID: `ours-fleet-owner-${process.pid}-${randomUUID()}`,
39
- },
40
- stdio: ['pipe', 'pipe', 'pipe'],
41
- });
42
- await new Promise((resolve, reject) => {
43
- child.once('spawn', resolve);
44
- child.once('error', reject);
45
- });
46
- this.child = child;
47
- child.once('exit', (code, signal) => {
48
- this.log(`ours-mcp proxy launcher exited (${code ?? signal ?? 'unknown'})`);
49
- });
50
- child.stdin.on('error', error => this.log(`ours-mcp stdin: ${error.message}`));
51
- createInterface({ input: child.stderr }).on('line', line => this.log(`ours-mcp: ${line}`));
52
- try {
53
- await this.request('initialize', {
54
- protocolVersion: '2025-03-26', capabilities: {},
55
- clientInfo: { name: 'ours-fleet-owner-channel', version: '1' },
56
- });
57
- await this.notify('notifications/initialized', {});
58
- }
59
- catch (error) {
60
- await this.close();
61
- throw error;
62
- }
63
- }
64
- async callTool(name, args = {}) {
65
- const result = await this.request('tools/call', { name, arguments: args });
66
- const text = (result.content ?? [])
67
- .filter(item => item.type === 'text').map(item => item.text ?? '').join('\n').trim();
68
- if (result.isError)
69
- throw new OursMcpError(text || `ours tool ${name} failed`);
70
- if (result.structuredContent !== undefined)
71
- return result.structuredContent;
72
- if (!text)
73
- return {};
74
- try {
75
- return JSON.parse(text);
76
- }
77
- catch {
78
- return text;
79
- }
80
- }
81
- async close() {
82
- const child = this.child;
83
- this.child = undefined;
84
- if (!child || child.exitCode !== null)
85
- return;
86
- // EOF asks the proxy to close normally. Once this exact process exits, the
87
- // daemon can reclaim its lease even while the supervisor stays alive.
88
- child.stdin.end();
89
- const exited = await new Promise(resolve => {
90
- const timer = setTimeout(() => resolve(false), 1_000);
91
- child.once('exit', () => { clearTimeout(timer); resolve(true); });
92
- });
93
- if (exited || child.exitCode !== null)
94
- return;
95
- child.kill('SIGTERM');
96
- await new Promise(resolve => {
97
- const timer = setTimeout(() => { child.kill('SIGKILL'); resolve(); }, 5_000);
98
- child.once('exit', () => { clearTimeout(timer); resolve(); });
99
- });
100
- }
101
- request(method, params) {
102
- const run = this.tail.then(() => this.requestNow(method, params));
103
- this.tail = run.then(() => undefined, () => undefined);
104
- return run;
105
- }
106
- async requestNow(method, params) {
107
- const child = this.child;
108
- if (!child || child.exitCode !== null)
109
- throw new OursMcpError('ours-mcp proxy is not running');
110
- const id = ++this.nextId;
111
- await this.write(child, { jsonrpc: '2.0', id, method, params });
112
- const lines = createInterface({ input: child.stdout, crlfDelay: Infinity });
113
- try {
114
- for await (const line of lines) {
115
- let response;
116
- try {
117
- response = JSON.parse(line);
118
- }
119
- catch {
120
- continue;
121
- }
122
- if (response.id !== id)
123
- continue;
124
- if (response.error !== undefined)
125
- throw new OursMcpError(JSON.stringify(response.error));
126
- return response.result ?? {};
127
- }
128
- throw new OursMcpError('ours-mcp proxy closed its output');
129
- }
130
- finally {
131
- lines.close();
132
- }
133
- }
134
- async notify(method, params) {
135
- const child = this.child;
136
- if (!child || child.exitCode !== null)
137
- throw new OursMcpError('ours-mcp proxy is not running');
138
- await this.write(child, { jsonrpc: '2.0', method, params });
139
- }
140
- write(child, value) {
141
- return new Promise((resolve, reject) => {
142
- child.stdin.write(JSON.stringify(value) + '\n', error => error ? reject(error) : resolve());
143
- });
144
- }
145
- }