@loucompanion/forge-bridge 0.1.3-dev.fbe15023cdfe → 0.1.4-dev.bfa11ebe98c6

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.
Files changed (160) hide show
  1. package/README.md +88 -2
  2. package/dist/bridge/bridge-chat.types.d.ts +111 -0
  3. package/dist/bridge/bridge-chat.types.js +401 -0
  4. package/dist/bridge/bridge.base.d.ts +4 -1
  5. package/dist/bridge/bridge.service.d.ts +72 -3
  6. package/dist/bridge/bridge.service.js +1562 -152
  7. package/dist/bridge/bridge.types.d.ts +46 -1
  8. package/dist/bridge/dispatch-result-outbox.d.ts +25 -0
  9. package/dist/bridge/dispatch-result-outbox.js +62 -0
  10. package/dist/bridge/dispatch-result-worker.d.ts +2 -0
  11. package/dist/bridge/dispatch-result-worker.js +122 -0
  12. package/dist/bridge/internals.d.ts +2 -0
  13. package/dist/bridge/internals.js +1 -0
  14. package/dist/bridge/readiness-result-outbox.d.ts +21 -0
  15. package/dist/bridge/readiness-result-outbox.js +54 -0
  16. package/dist/bridge/readiness-result-worker.d.ts +2 -0
  17. package/dist/bridge/readiness-result-worker.js +121 -0
  18. package/dist/bridge/runner-host-capabilities.d.ts +15 -0
  19. package/dist/bridge/runner-host-capabilities.js +93 -0
  20. package/dist/cleanup/cleanup.service.d.ts +1 -0
  21. package/dist/cleanup/cleanup.service.js +70 -9
  22. package/dist/cli/cli.help.js +2 -2
  23. package/dist/cli/cli.service.d.ts +14 -6
  24. package/dist/cli/cli.service.js +75 -16
  25. package/dist/cli/local-config.d.ts +1 -0
  26. package/dist/cli/local-config.js +4 -0
  27. package/dist/runtime/aspnet-runtime-prebuild.d.ts +36 -0
  28. package/dist/runtime/aspnet-runtime-prebuild.js +341 -0
  29. package/dist/runtime/attempt-node-runtime-launcher.d.ts +2 -1
  30. package/dist/runtime/attempt-node-runtime-launcher.js +57 -18
  31. package/dist/runtime/attempt-path-operation.d.ts +21 -0
  32. package/dist/runtime/attempt-path-operation.js +94 -0
  33. package/dist/runtime/attempt-private-directory.d.ts +6 -0
  34. package/dist/runtime/attempt-private-directory.js +54 -0
  35. package/dist/runtime/attempt-process-supervisor.d.ts +11 -0
  36. package/dist/runtime/attempt-process-supervisor.js +67 -20
  37. package/dist/runtime/canonical-source-file.d.ts +3 -0
  38. package/dist/runtime/canonical-source-file.js +54 -0
  39. package/dist/runtime/docker-engine-probe.d.ts +90 -0
  40. package/dist/runtime/docker-engine-probe.js +333 -0
  41. package/dist/runtime/generated-output-reset.d.ts +10 -0
  42. package/dist/runtime/generated-output-reset.js +62 -8
  43. package/dist/runtime/host-database-capability-boundary.d.ts +16 -0
  44. package/dist/runtime/host-database-capability-boundary.js +62 -0
  45. package/dist/runtime/host-diagnostics.d.ts +11 -0
  46. package/dist/runtime/host-diagnostics.js +191 -0
  47. package/dist/runtime/host-execution-root.d.ts +4 -0
  48. package/dist/runtime/host-execution-root.js +39 -0
  49. package/dist/runtime/host-listener-inspector.d.ts +8 -1
  50. package/dist/runtime/host-listener-inspector.js +121 -60
  51. package/dist/runtime/host-runtime-contract.d.ts +12 -1
  52. package/dist/runtime/host-runtime-contract.js +107 -53
  53. package/dist/runtime/host-runtime-lease.d.ts +90 -8
  54. package/dist/runtime/host-runtime-lease.js +1097 -108
  55. package/dist/runtime/host-runtime-promotion.d.ts +12 -3
  56. package/dist/runtime/host-runtime-promotion.js +59 -12
  57. package/dist/runtime/memory-capacity.d.ts +70 -0
  58. package/dist/runtime/memory-capacity.js +123 -0
  59. package/dist/runtime/native-test-postgres-fallback.d.ts +25 -0
  60. package/dist/runtime/native-test-postgres-fallback.js +91 -0
  61. package/dist/runtime/node-host-runtime-adapter.d.ts +3 -0
  62. package/dist/runtime/node-host-runtime-adapter.js +28 -11
  63. package/dist/runtime/pinned-directory-process.d.ts +9 -0
  64. package/dist/runtime/pinned-directory-process.js +61 -0
  65. package/dist/runtime/pinned-source-directory.d.ts +17 -0
  66. package/dist/runtime/pinned-source-directory.js +94 -0
  67. package/dist/runtime/playwright-host-runtime-adapter.d.ts +1 -0
  68. package/dist/runtime/playwright-host-runtime-adapter.js +53 -5
  69. package/dist/runtime/prepared-runtime-identity.d.ts +6 -0
  70. package/dist/runtime/prepared-runtime-identity.js +218 -0
  71. package/dist/runtime/prepared-runtime-readiness.d.ts +30 -0
  72. package/dist/runtime/prepared-runtime-readiness.js +285 -0
  73. package/dist/runtime/runtime-isolation-lease.js +4 -1
  74. package/dist/runtime/workspace-resource-estimator.d.ts +17 -0
  75. package/dist/runtime/workspace-resource-estimator.js +174 -0
  76. package/dist/sandbox/preflight.d.ts +31 -1
  77. package/dist/sandbox/preflight.js +237 -51
  78. package/dist/sandbox/repository-lock.d.ts +2 -1
  79. package/dist/sandbox/repository-lock.js +17 -8
  80. package/dist/sandbox/runtime.d.ts +6 -1
  81. package/dist/sandbox/runtime.js +132 -103
  82. package/dist/sandbox/sandbox.types.d.ts +7 -0
  83. package/dist/sandbox/workspace.service.js +2 -0
  84. package/dist/session/managed-attempt-tools.d.ts +17 -1
  85. package/dist/session/managed-attempt-tools.js +481 -31
  86. package/dist/session/managed-build-coordinator.d.ts +5 -0
  87. package/dist/session/managed-build-coordinator.js +105 -0
  88. package/dist/session/managed-build-receipt.d.ts +21 -0
  89. package/dist/session/managed-build-receipt.js +337 -0
  90. package/dist/session/managed-verification-timeouts.d.ts +8 -0
  91. package/dist/session/managed-verification-timeouts.js +17 -0
  92. package/dist/session/phase-completion.artifacts.d.ts +17 -0
  93. package/dist/session/phase-completion.artifacts.js +224 -22
  94. package/dist/session/phase-completion.d.ts +11 -1
  95. package/dist/session/phase-completion.helpers.d.ts +5 -0
  96. package/dist/session/phase-completion.helpers.js +30 -7
  97. package/dist/session/phase-completion.js +14 -3
  98. package/dist/session/phase-completion.prompts.js +17 -18
  99. package/dist/session/phase-completion.runtime.d.ts +3 -1
  100. package/dist/session/phase-completion.runtime.js +118 -16
  101. package/dist/session/phase-completion.schemas.js +49 -2
  102. package/dist/session/phase-completion.validation.d.ts +1 -0
  103. package/dist/session/phase-completion.validation.js +111 -21
  104. package/dist/session/pre-publication-validation.d.ts +75 -0
  105. package/dist/session/pre-publication-validation.js +657 -0
  106. package/dist/session/provider-cli/bridge-cli.conformance-fixtures.d.ts +72 -0
  107. package/dist/session/provider-cli/bridge-cli.conformance-fixtures.js +462 -0
  108. package/dist/session/provider-cli/ci-repair-readiness.d.ts +8 -0
  109. package/dist/session/provider-cli/ci-repair-readiness.js +23 -3
  110. package/dist/session/provider-cli/provider-cli.artifacts.d.ts +4 -2
  111. package/dist/session/provider-cli/provider-cli.artifacts.js +47 -13
  112. package/dist/session/provider-cli/provider-cli.deterministic.js +34 -2
  113. package/dist/session/provider-cli/provider-cli.process.d.ts +51 -0
  114. package/dist/session/provider-cli/provider-cli.process.js +422 -22
  115. package/dist/session/provider-cli/provider-cli.prompts.d.ts +5 -0
  116. package/dist/session/provider-cli/provider-cli.prompts.js +136 -16
  117. package/dist/session/provider-cli/provider-cli.repair-epoch.d.ts +6 -0
  118. package/dist/session/provider-cli/provider-cli.repair-epoch.js +11 -0
  119. package/dist/session/provider-cli/provider-cli.runtime.d.ts +8 -1
  120. package/dist/session/provider-cli/provider-cli.runtime.js +40 -3
  121. package/dist/session/provider-cli/provider-cli.service.d.ts +30 -5
  122. package/dist/session/provider-cli/provider-cli.service.js +771 -68
  123. package/dist/session/provider-cli/provider-cli.types.d.ts +2 -0
  124. package/dist/session/provider-cli/provider-output-state.d.ts +10 -0
  125. package/dist/session/provider-cli/provider-output-state.js +361 -17
  126. package/dist/session/provider-cli/provider-usage.d.ts +7 -0
  127. package/dist/session/provider-cli/provider-usage.js +28 -5
  128. package/dist/session/session-redaction.js +434 -9
  129. package/dist/session/session.service.js +145 -23
  130. package/dist/session/session.types.d.ts +147 -2
  131. package/dist/session/session.types.js +587 -7
  132. package/dist/session/unreferenced-exports.d.ts +12 -0
  133. package/dist/session/unreferenced-exports.js +59 -0
  134. package/dist/shared/format/format-bytes.d.ts +1 -0
  135. package/dist/shared/format/format-bytes.js +9 -0
  136. package/dist/shared/git/git.js +2 -1
  137. package/dist/shared/path/path.d.ts +1 -0
  138. package/dist/shared/path/path.js +18 -0
  139. package/dist/shared/process/dotnet-toolchain.d.ts +17 -0
  140. package/dist/shared/process/dotnet-toolchain.js +307 -0
  141. package/dist/shared/process/host-process-supervisor.d.ts +2 -0
  142. package/dist/shared/process/host-process-supervisor.js +57 -9
  143. package/dist/shared/process/process.d.ts +2 -0
  144. package/dist/shared/process/process.js +12 -7
  145. package/dist/shared/provider-capabilities.d.ts +3 -0
  146. package/dist/shared/provider-capabilities.js +85 -0
  147. package/dist/shared/redaction/redaction.d.ts +1 -0
  148. package/dist/shared/redaction/redaction.js +39 -3
  149. package/dist/version.d.ts +1 -1
  150. package/dist/version.js +1 -1
  151. package/dist/worktree/dependency-warmer.d.ts +41 -9
  152. package/dist/worktree/dependency-warmer.js +497 -22
  153. package/dist/worktree/index.d.ts +1 -1
  154. package/dist/worktree/story-qa-reference-hydrator.d.ts +4 -0
  155. package/dist/worktree/story-qa-reference-hydrator.js +253 -0
  156. package/dist/worktree/worktree.service.d.ts +7 -0
  157. package/dist/worktree/worktree.service.js +253 -21
  158. package/dist/worktree/worktree.types.d.ts +7 -1
  159. package/package.json +7 -5
  160. package/runtime/windows-job-supervisor.ps1 +380 -14
@@ -4,25 +4,54 @@ import path from "node:path";
4
4
  import { randomUUID } from "node:crypto";
5
5
  import os from "node:os";
6
6
  import { defaultForgeBridgeConfig } from "../config.js";
7
- import { supportsDraftImageAttachmentsFor } from "../shared/provider-capabilities.js";
7
+ import { bridgeChatCapabilityFor, supportsDraftImageAttachmentsFor } from "../shared/provider-capabilities.js";
8
8
  import { redactText } from "../shared/redaction/redaction.js";
9
9
  import { BridgeSessionService } from "../session/session.service.js";
10
10
  import { ProcessBackedProviderCliService, RoutingProviderCliService, StreamingProviderProcessRunner, } from "../session/provider-cli/provider-cli.service.js";
11
+ import { codexSandboxModeForPlatform, formatProviderCliCommandForCheckpoint } from "../session/provider-cli/provider-cli.prompts.js";
12
+ import { clip, createProviderOutputState } from "../session/provider-cli/provider-output-state.js";
11
13
  import { redactRunnerDispatchEvent, redactRunOnceResultPacket } from "../session/session-redaction.js";
12
- import { HostRuntimeLeaseManager } from "../runtime/host-runtime-lease.js";
14
+ import { PrePublicationValidator } from "../session/pre-publication-validation.js";
15
+ import { HostRuntimeLeaseManager, } from "../runtime/host-runtime-lease.js";
16
+ import { assertHostDatabaseCapabilityBoundary, establishHostDatabaseCapabilityBoundary, } from "../runtime/host-database-capability-boundary.js";
17
+ import { defaultHostExecutionRoot } from "../runtime/host-execution-root.js";
18
+ import { buildMemoryAdmissionDiagnostics } from "../runtime/memory-capacity.js";
19
+ import { DockerEngineProbe } from "../runtime/docker-engine-probe.js";
20
+ import { readNativeTestPostgresFallback } from "../runtime/native-test-postgres-fallback.js";
21
+ import { estimateWorkspaceBytesFromRepo } from "../runtime/workspace-resource-estimator.js";
13
22
  import { GitWorktreeService } from "../worktree/worktree.service.js";
14
23
  import { RepositoryCoordinatorLock } from "../sandbox/repository-lock.js";
15
24
  import { AttemptCoordinatedProviderProcessRunner, AttemptProcessSupervisorRegistry, } from "../runtime/attempt-process-supervisor-registry.js";
16
- import { failureResult, parseRunOnceDispatchPacket } from "../session/session.types.js";
25
+ import { failureFactsForCode, failureFactsFromError, failureResult, parseRunOnceDispatchPacket } from "../session/session.types.js";
17
26
  import { forgeBridgeVersion } from "../version.js";
18
27
  import { canRunDockerMaintenance, DockerPreflight } from "../sandbox/preflight.js";
19
28
  import { validateSandboxContract } from "../sandbox/contract.js";
20
29
  import { runSandboxProcess } from "../sandbox/process.js";
21
30
  import { DockerSandboxRuntime, selectSandboxConfigRefs } from "../sandbox/runtime.js";
22
31
  import { codexSandboxProviderAuthAdapter, sandboxProviderAuthAdapterFor, sandboxProviderAuthUnavailableMessage } from "../sandbox/provider-auth/codex.js";
32
+ import { assertBridgeChatCallbackPayload, bridgeChatMaxProviderThreadIdChars, bridgeChatProtocolVersion, isTerminalChatTurnCallbackKind, parseChatTurnDispatchPacket, } from "./bridge-chat.types.js";
33
+ import { formatBytes } from "../shared/format/format-bytes.js";
34
+ import { emptyRunnerHostCapabilitySnapshot, ProcessRunnerHostCapabilityProbe, } from "./runner-host-capabilities.js";
35
+ import { FileReadinessResultOutbox } from "./readiness-result-outbox.js";
36
+ import { FileDispatchResultOutbox } from "./dispatch-result-outbox.js";
37
+ /**
38
+ * Protocol-v2 events are acknowledged one at a time before runOnce returns.
39
+ * Sending the complete transcript again in DispatchResult made long QA and CI
40
+ * sessions exceed SignalR's receive limit, disconnecting the bridge after the
41
+ * provider had already completed. The API combines its acknowledged streamed
42
+ * events with the terminal result, so the callback only needs the outcome and
43
+ * artifacts. Protocol v1 retains the transcript for legacy servers.
44
+ */
45
+ export function compactTerminalResult(dispatch, result) {
46
+ return dispatch.schemaVersion >= 2 && result.events.length > 0
47
+ ? { ...result, events: [] }
48
+ : result;
49
+ }
23
50
  export class ForgeBridgeService {
24
51
  config;
25
52
  connections;
53
+ hostCapabilityProbe;
54
+ chatProcessRunner;
26
55
  connection = null;
27
56
  heartbeatTimer = null;
28
57
  heartbeatInFlight = false;
@@ -32,18 +61,40 @@ export class ForgeBridgeService {
32
61
  sandboxMaintenanceTimer = null;
33
62
  sandboxMaintenanceRunning = false;
34
63
  closed = null;
64
+ transportRecovery = null;
65
+ transportRecoveryGeneration = 0;
35
66
  stopped = false;
36
67
  activeSessions = new Map();
37
68
  reservations = new Map();
38
69
  activeReadinessProbes = new Map();
70
+ readinessResultOutbox;
71
+ readinessResultFlush = null;
72
+ dispatchResultOutbox;
73
+ dispatchResultFlush = null;
74
+ dispatchResultGeneration = 0;
75
+ dispatchResultFlushedGeneration = -1;
76
+ dispatchResultHandoffs = new Map();
39
77
  preflight;
78
+ dockerEngineProbe;
40
79
  sandboxRuntime;
41
80
  sandboxResourceResolver;
42
81
  sessions;
43
- constructor(config, sessions = undefined, connections = new SignalRBridgeConnectionFactory(), preflight, sandboxRuntime, sandboxResourceResolver) {
82
+ enforceHostDatabaseCapabilityBoundary;
83
+ hostCapabilities = emptyRunnerHostCapabilitySnapshot();
84
+ // Reaping deletes the stale ledger, so its stopped-process proof is held here
85
+ // until a Hello carrying it is accepted. A bridge restart forgets it, which
86
+ // leaves those attempts on Forge's conservative unknown-outcome path.
87
+ recoveredHostAttempts = new Map();
88
+ hostCapabilitiesDetected = false;
89
+ constructor(config, sessions = undefined, connections = new SignalRBridgeConnectionFactory(), preflight, sandboxRuntime, sandboxResourceResolver, dockerEngineProbe, hostCapabilityProbe = new ProcessRunnerHostCapabilityProbe(), readinessResultOutbox, chatProcessRunner = new StreamingProviderProcessRunner(), dispatchResultOutbox) {
44
90
  this.config = config;
45
91
  this.connections = connections;
46
- this.sessions = sessions ?? createDefaultSessionService(config.sandboxStateRoot);
92
+ this.hostCapabilityProbe = hostCapabilityProbe;
93
+ this.chatProcessRunner = chatProcessRunner;
94
+ this.enforceHostDatabaseCapabilityBoundary = sessions === undefined;
95
+ if (this.enforceHostDatabaseCapabilityBoundary) {
96
+ establishHostDatabaseCapabilityBoundary();
97
+ }
47
98
  this.preflight = preflight ?? new DockerPreflight({
48
99
  dockerExecutable: config.dockerExecutable,
49
100
  dockerContext: config.dockerContext,
@@ -59,17 +110,36 @@ export class ForgeBridgeService {
59
110
  runtimeOverheadBytes: config.sandboxRuntimeOverheadBytes,
60
111
  requireRootless: config.requireRootlessDocker
61
112
  });
113
+ this.dockerEngineProbe = dockerEngineProbe ?? new DockerEngineProbe({
114
+ dockerExecutable: config.dockerExecutable,
115
+ dockerContext: config.dockerContext,
116
+ recovery: {
117
+ enabled: config.dockerDesktopRecoveryEnabled,
118
+ failureThreshold: config.dockerRecoveryFailureThreshold,
119
+ cooldownMs: config.dockerRecoveryCooldownMs,
120
+ startupTimeoutMs: config.dockerRecoveryStartupTimeoutMs,
121
+ canRecover: () => this.activeSessions.size === 0 && this.reservations.size === 0,
122
+ },
123
+ onStatus: logBridgeStatus,
124
+ });
125
+ this.sessions = sessions ?? createDefaultSessionService(config.sandboxStateRoot, config.sandboxExecutionRoot, this.dockerEngineProbe);
62
126
  this.sandboxRuntime = sandboxRuntime ?? new DockerSandboxRuntime(config, (request, signal) => this.redeemPublishGrant(request, signal));
63
127
  this.sandboxResourceResolver = sandboxResourceResolver
64
128
  ?? (dispatch => resolveSandboxResourcesFromRepo(dispatch, this.config.sandboxRuntimeOverheadBytes));
129
+ this.readinessResultOutbox = readinessResultOutbox ?? new FileReadinessResultOutbox(config.sandboxStateRoot);
130
+ this.dispatchResultOutbox = dispatchResultOutbox ?? new FileDispatchResultOutbox(config.sandboxStateRoot);
65
131
  }
66
132
  async connectOnce() {
67
133
  await this.stopConnection();
68
134
  this.stopped = false;
135
+ if (this.enforceHostDatabaseCapabilityBoundary) {
136
+ assertHostDatabaseCapabilityBoundary();
137
+ }
69
138
  const hostReap = await this.sessions.reapHostRuntime?.({
70
139
  runnerId: this.config.runnerId,
71
140
  runnerInstanceId: this.config.runnerInstanceId,
72
141
  });
142
+ this.rememberRecoveredHostAttempts(hostReap);
73
143
  if (hostReap && (hostReap.reaped > 0 || hostReap.degraded > 0)) {
74
144
  logBridgeStatus(`Managed-host recovery reaped ${hostReap.reaped} stale attempt(s), protected ${hostReap.reusedPidsProtected} reused PID(s), and left ${hostReap.degraded} degraded attempt(s) for reconciliation.`);
75
145
  }
@@ -80,17 +150,54 @@ export class ForgeBridgeService {
80
150
  await this.preflight.refresh();
81
151
  }
82
152
  }
153
+ await this.dockerEngineProbe.refresh(true);
154
+ if (!this.hostCapabilitiesDetected) {
155
+ this.hostCapabilities = await this.hostCapabilityProbe.detect(this.config);
156
+ this.hostCapabilitiesDetected = true;
157
+ }
83
158
  const connection = this.connections.create(this.config);
84
159
  this.connection = connection;
160
+ this.transportRecovery = null;
85
161
  this.closed = new Promise((resolve) => {
86
- connection.onClose(() => {
162
+ connection.onClose((error) => {
87
163
  this.stopHeartbeat();
164
+ this.completeTransportRecovery(error ?? new Error("bridge connection closed"));
88
165
  this.abortActiveSessions("bridge connection closed");
89
166
  this.abortActiveReadinessProbes("bridge connection closed");
90
167
  resolve();
91
168
  });
92
169
  });
170
+ connection.onReconnecting?.((error) => {
171
+ if (connection !== this.connection || this.stopped)
172
+ return;
173
+ this.stopHeartbeat();
174
+ this.beginTransportRecovery();
175
+ logBridgeStatus(`Forge connection interrupted${error ? `: ${errorMessage(error)}` : ""}. Reconnecting in place; active dispatches remain supervised...`);
176
+ });
177
+ connection.onReconnected?.(async () => {
178
+ if (connection !== this.connection || this.stopped)
179
+ return;
180
+ try {
181
+ await this.registerConnection(connection);
182
+ this.completeTransportRecovery();
183
+ this.startHeartbeat();
184
+ await this.flushDispatchResults(true).catch(error => {
185
+ logBridgeStatus(`Dispatch result replay deferred after reconnect: ${errorMessage(error)}`);
186
+ });
187
+ await this.flushReadinessResults().catch(error => {
188
+ logBridgeStatus(`Readiness result replay deferred after reconnect: ${errorMessage(error)}`);
189
+ });
190
+ await this.flushCleanupReports().catch(() => undefined);
191
+ logBridgeStatus("Forge connection restored; active dispatches remain attached to this runner process.");
192
+ }
193
+ catch (error) {
194
+ this.completeTransportRecovery(error);
195
+ logBridgeStatus(`Forge reconnect registration failed: ${errorMessage(error)}. Closing the connection for a clean retry...`);
196
+ await connection.stop().catch(() => undefined);
197
+ }
198
+ });
93
199
  connection.onDispatch((dispatch) => this.handleDispatch(dispatch));
200
+ connection.onChatTurn?.((dispatch) => this.handleChatTurn(dispatch));
94
201
  connection.onOffer?.((dispatch) => this.handleOffer(dispatch));
95
202
  connection.onStart?.((dispatch) => this.handleStart(dispatch));
96
203
  connection.onLeaseUpdate?.((request) => this.handleLeaseUpdate(request));
@@ -98,18 +205,67 @@ export class ForgeBridgeService {
98
205
  connection.onReap?.((request) => this.handleReap(request));
99
206
  connection.onReadinessProbe?.((request) => this.handleReadinessProbe(request));
100
207
  await connection.start();
101
- await this.invokeAck("Hello", {
102
- version: this.config.version,
103
- protocolVersion: this.config.protocolVersion,
104
- minServerVersion: this.config.minServerVersion ?? null,
105
- runnerInstanceId: this.config.runnerInstanceId,
106
- capabilities: this.capabilities()
208
+ await this.registerConnection(connection);
209
+ await this.flushDispatchResults(true).catch(error => {
210
+ logBridgeStatus(`Dispatch result replay deferred: ${errorMessage(error)}`);
211
+ });
212
+ await this.flushReadinessResults().catch(error => {
213
+ logBridgeStatus(`Readiness result replay deferred: ${errorMessage(error)}`);
107
214
  });
108
215
  await this.flushCleanupReports();
109
216
  this.startHeartbeat();
110
217
  this.startLeaseWatchdog();
111
218
  this.startSandboxMaintenance();
112
219
  }
220
+ rememberRecoveredHostAttempts(report) {
221
+ for (const attempt of report?.recoveredAttempts ?? []) {
222
+ if (this.recoveredHostAttempts.size >= maxRecoveredHostAttempts)
223
+ break;
224
+ this.recoveredHostAttempts.set(attempt.dispatchAttemptId, attempt);
225
+ }
226
+ }
227
+ async recoveredHostAttemptsForHello() {
228
+ if (this.recoveredHostAttempts.size === 0)
229
+ return [];
230
+ // A durable terminal result outranks the stopped-process proof: it replays
231
+ // after Hello and Forge accepts the provider's real outcome instead.
232
+ const pendingResults = new Set((await this.dispatchResultOutbox.pending())
233
+ .map(entry => entry.payload.dispatchAttemptId)
234
+ .filter((id) => typeof id === "string"));
235
+ return [...this.recoveredHostAttempts.values()]
236
+ .filter(attempt => !pendingResults.has(attempt.dispatchAttemptId));
237
+ }
238
+ async registerConnection(connection) {
239
+ let storageRecoveryAttempt = 0;
240
+ while (true) {
241
+ try {
242
+ const recoveredHostAttempts = await this.recoveredHostAttemptsForHello();
243
+ const ack = await connection.invoke("Hello", {
244
+ version: this.config.version,
245
+ protocolVersion: this.config.protocolVersion,
246
+ minServerVersion: this.config.minServerVersion ?? null,
247
+ runnerInstanceId: this.config.runnerInstanceId,
248
+ capabilities: this.capabilities(),
249
+ recoveredHostAttempts,
250
+ });
251
+ if (!ack.accepted) {
252
+ throw new BridgeCallbackRejectedError(ack.code ?? null, ack.message ?? ack.code ?? "Hello callback was rejected.");
253
+ }
254
+ this.recoveredHostAttempts.clear();
255
+ logBridgeStatus(`Registered runner ${this.config.runnerId}, instance ${this.config.runnerInstanceId}, build ${this.config.version}, protocol ${this.config.protocolVersion}.`);
256
+ return;
257
+ }
258
+ catch (error) {
259
+ if (!isProviderStorageUnavailable(error) || storageRecoveryAttempt >= this.config.reconnectBackoffMs.length) {
260
+ throw error;
261
+ }
262
+ const backoff = this.config.reconnectBackoffMs[storageRecoveryAttempt] ?? 5000;
263
+ storageRecoveryAttempt += 1;
264
+ logBridgeStatus(`Hello reached Forge while storage was recovering. Retrying registration ${storageRecoveryAttempt} in ${formatDelay(backoff)}...`);
265
+ await delay(backoff);
266
+ }
267
+ }
268
+ }
113
269
  async run(signal) {
114
270
  let attempt = 0;
115
271
  try {
@@ -151,24 +307,351 @@ export class ForgeBridgeService {
151
307
  await this.stopConnection();
152
308
  await this.preflight.release();
153
309
  }
154
- async handleDispatch(rawDispatch) {
155
- const dispatch = this.parseDispatchOrReport(rawDispatch);
310
+ async handleChatTurn(rawDispatch) {
311
+ const packet = this.parseChatTurnOrReport(rawDispatch);
312
+ if (!packet) {
313
+ return;
314
+ }
315
+ if (packet.runnerInstanceId !== this.config.runnerInstanceId) {
316
+ logBridgeStatus(`Rejected Bridge Chat turn ${packet.turnId}: stale runner instance.`);
317
+ return;
318
+ }
319
+ const capabilityFailure = this.requiredChatCapabilityFailure(packet);
320
+ if (capabilityFailure) {
321
+ await this.sendStandaloneChatTerminal(packet, "failed", {
322
+ code: "bridge_chat.runner_unavailable",
323
+ message: capabilityFailure,
324
+ }).catch(() => undefined);
325
+ return;
326
+ }
327
+ const existing = this.activeSessions.get(packet.leaseId);
328
+ if (existing) {
329
+ if (existing.kind === "chat"
330
+ && existing.fencingToken === packet.fencingToken
331
+ && existing.chatRunId === packet.chatRunId
332
+ && existing.conversationId === packet.conversationId
333
+ && existing.turnId === packet.turnId) {
334
+ return;
335
+ }
336
+ await this.sendStandaloneChatTerminal(packet, "failed", {
337
+ code: "bridge_chat.stale_fencing_token",
338
+ message: "Bridge Chat turn fence conflicts with an active bridge session.",
339
+ }).catch(() => undefined);
340
+ return;
341
+ }
342
+ const controller = new AbortController();
343
+ const active = {
344
+ leaseId: packet.leaseId,
345
+ kind: "chat",
346
+ fencingToken: packet.fencingToken,
347
+ dispatchAttemptId: null,
348
+ runnerInstanceId: packet.runnerInstanceId,
349
+ leaseExpiresAt: chatTurnLeaseExpiresAt(packet),
350
+ sandboxed: false,
351
+ resources: zeroSandboxResources(),
352
+ controller,
353
+ chatRunId: packet.chatRunId,
354
+ conversationId: packet.conversationId,
355
+ turnId: packet.turnId,
356
+ providerThreadId: packet.providerThreadId ?? null,
357
+ chatSequence: 0,
358
+ terminalCallbackAttempted: false,
359
+ };
360
+ this.activeSessions.set(packet.leaseId, active);
361
+ let workspacePath = null;
362
+ let cleanupAttempted = false;
363
+ let cleanupOutcome = null;
364
+ let cleanupReported = false;
365
+ const cleanup = async () => {
366
+ if (!workspacePath || cleanupAttempted)
367
+ return null;
368
+ cleanupAttempted = true;
369
+ cleanupOutcome = await cleanupChatTurnWorkspace(workspacePath, chatRedactionSecrets(packet));
370
+ return cleanupOutcome;
371
+ };
372
+ const reportCleanup = async (outcome) => {
373
+ if (!outcome || cleanupReported)
374
+ return;
375
+ cleanupReported = true;
376
+ await this.sendChatCallback(active, "output_delta", outcome).catch(() => undefined);
377
+ };
378
+ try {
379
+ workspacePath = await this.createChatTurnWorkspace(packet);
380
+ await materializeChatTurnMetadata(packet, workspacePath);
381
+ const run = await this.runChatProviderProcess(packet, active, workspacePath);
382
+ cleanupOutcome = await cleanup();
383
+ await reportCleanup(cleanupOutcome);
384
+ if (controller.signal.aborted) {
385
+ await this.sendChatAbortTerminal(packet, active, cleanupOutcome);
386
+ return;
387
+ }
388
+ if (cleanupOutcome?.status === "failed") {
389
+ throw new Error(`chat_turn_cleanup_failed: ${cleanupOutcome.message}`);
390
+ }
391
+ if (run.result.exitCode !== 0) {
392
+ throw chatProviderExitFailure(run.command, run.result, chatRedactionSecrets(packet));
393
+ }
394
+ await this.sendChatCallback(active, "output_completed", {
395
+ code: "output_completed",
396
+ text: clip(redactText(run.output.finalMessage ?? "", chatRedactionSecrets(packet)), maxChatTerminalTextChars),
397
+ provider: packet.providerKind,
398
+ model: packet.model,
399
+ cleanup: cleanupOutcome,
400
+ });
401
+ }
402
+ catch (error) {
403
+ cleanupOutcome = cleanupOutcome ?? await cleanup();
404
+ if (active.terminalCallbackAttempted) {
405
+ return;
406
+ }
407
+ await reportCleanup(cleanupOutcome);
408
+ const payload = controller.signal.aborted
409
+ ? chatAbortPayload(controller.signal.reason, cleanupOutcome, packet)
410
+ : {
411
+ code: "provider_failed",
412
+ message: clip(redactText(errorMessage(error), chatRedactionSecrets(packet)), 2000),
413
+ cleanup: cleanupOutcome,
414
+ };
415
+ const kind = controller.signal.aborted ? chatAbortKind(controller.signal.reason) : "failed";
416
+ await this.sendChatCallback(active, kind, payload).catch(callbackError => {
417
+ logBridgeStatus(`Bridge Chat terminal callback for turn ${packet.turnId} was not accepted: ${errorMessage(callbackError)}`);
418
+ });
419
+ }
420
+ finally {
421
+ await this.releaseActiveSessionAndRefreshCapacity(packet.leaseId);
422
+ }
423
+ }
424
+ async createChatTurnWorkspace(packet) {
425
+ const root = path.join(this.config.sandboxStateRoot, "chat-turns");
426
+ await fs.promises.mkdir(root, { recursive: true });
427
+ return await fs.promises.mkdtemp(path.join(root, `${safePathComponent(packet.turnId)}-`));
428
+ }
429
+ async runChatProviderProcess(packet, active, cwd) {
430
+ const output = createProviderOutputState(packet.providerKind);
431
+ const plan = buildChatProviderCommand(packet, cwd);
432
+ const redactionSecrets = chatRedactionSecrets(packet);
433
+ let started = false;
434
+ const result = await this.chatProcessRunner.run({
435
+ command: plan.command.command,
436
+ args: plan.command.args,
437
+ cwd,
438
+ env: {
439
+ FORGE_BRIDGE_CHAT_RUN_ID: packet.chatRunId,
440
+ FORGE_BRIDGE_CHAT_TURN_ID: packet.turnId,
441
+ },
442
+ stdin: plan.stdin,
443
+ }, {
444
+ onStarted: async () => {
445
+ started = true;
446
+ await this.sendChatCallback(active, "turn_started", {
447
+ status: "running",
448
+ provider: packet.providerKind,
449
+ model: packet.model,
450
+ command: redactText(formatProviderCliCommandForCheckpoint(plan.command), redactionSecrets),
451
+ });
452
+ },
453
+ onStdoutLine: async (line) => {
454
+ for (const event of output.consume(line)) {
455
+ await this.sendChatProviderEvent(packet, active, event, output.providerThreadId);
456
+ }
457
+ },
458
+ }, active.controller.signal);
459
+ if (started && !active.controller.signal.aborted) {
460
+ await this.sendChatCallback(active, "output_delta", {
461
+ eventType: "provider_cli_completed",
462
+ role: "runner",
463
+ kind: "checkpoint",
464
+ content: "Bridge Chat provider process exited.",
465
+ }).catch(() => undefined);
466
+ }
467
+ return { command: plan.command, output, result };
468
+ }
469
+ async sendChatProviderEvent(packet, active, event, providerThreadId) {
470
+ if (providerThreadId) {
471
+ active.providerThreadId = providerThreadId;
472
+ }
473
+ else if (event.eventType === "provider_thread_started") {
474
+ active.providerThreadId = providerThreadIdFromEvent(event) ?? active.providerThreadId;
475
+ }
476
+ if (event.eventType === "provider_tool_heartbeat") {
477
+ return;
478
+ }
479
+ const redactionSecrets = chatRedactionSecrets(packet);
480
+ const payload = {
481
+ role: event.role,
482
+ kind: event.kind,
483
+ eventType: event.eventType ?? "provider_event",
484
+ content: clip(redactText(event.content, redactionSecrets), maxChatEventContentChars),
485
+ data: event.data ? clip(redactText(event.data, redactionSecrets), maxChatEventContentChars) : undefined,
486
+ };
487
+ await this.sendChatCallback(active, "output_delta", payload);
488
+ }
489
+ async sendChatAbortTerminal(packet, active, cleanupOutcome) {
490
+ await this.sendChatCallback(active, chatAbortKind(active.controller.signal.reason), chatAbortPayload(active.controller.signal.reason, cleanupOutcome, packet));
491
+ }
492
+ async sendStandaloneChatTerminal(packet, kind, payload) {
493
+ const active = {
494
+ leaseId: packet.leaseId,
495
+ kind: "chat",
496
+ fencingToken: packet.fencingToken,
497
+ dispatchAttemptId: null,
498
+ runnerInstanceId: packet.runnerInstanceId,
499
+ leaseExpiresAt: new Date(Date.now() + defaultChatTurnLeaseMs).toISOString(),
500
+ sandboxed: false,
501
+ resources: zeroSandboxResources(),
502
+ controller: new AbortController(),
503
+ chatRunId: packet.chatRunId,
504
+ conversationId: packet.conversationId,
505
+ turnId: packet.turnId,
506
+ providerThreadId: packet.providerThreadId ?? null,
507
+ chatSequence: 0,
508
+ terminalCallbackAttempted: false,
509
+ };
510
+ await this.sendChatCallback(active, kind, payload);
511
+ }
512
+ async sendChatCallback(active, kind, payload = null) {
513
+ if (active.kind !== "chat" || !active.chatRunId || !active.conversationId || !active.turnId) {
514
+ throw new Error("Bridge Chat callback requires an active chat turn.");
515
+ }
516
+ const terminal = isTerminalChatTurnCallbackKind(kind);
517
+ if (terminal && active.terminalCallbackAttempted) {
518
+ return;
519
+ }
520
+ assertBridgeChatCallbackPayload(payload);
521
+ const sequence = (active.chatSequence ?? 0) + 1;
522
+ const envelope = {
523
+ callbackId: randomUUID(),
524
+ chatRunId: active.chatRunId,
525
+ conversationId: active.conversationId,
526
+ turnId: active.turnId,
527
+ leaseId: active.leaseId,
528
+ fencingToken: active.fencingToken,
529
+ runnerId: this.config.runnerId,
530
+ runnerInstanceId: active.runnerInstanceId ?? this.config.runnerInstanceId,
531
+ sequence,
532
+ eventType: kind,
533
+ redactionStatus: "redacted",
534
+ providerThreadId: active.providerThreadId ?? null,
535
+ payload,
536
+ occurredAt: new Date().toISOString(),
537
+ protocolVersion: bridgeChatProtocolVersion,
538
+ };
539
+ if (terminal) {
540
+ active.terminalCallbackAttempted = true;
541
+ }
542
+ await this.invokeChatCallbackWithRetry(envelope);
543
+ active.chatSequence = sequence;
544
+ }
545
+ async invokeChatCallbackWithRetry(envelope) {
546
+ const maxAttempts = 3;
547
+ for (let attempt = 0; attempt < maxAttempts; attempt += 1) {
548
+ try {
549
+ await this.invokeAck("ChatTurnCallback", envelope);
550
+ return;
551
+ }
552
+ catch (error) {
553
+ if (error instanceof BridgeCallbackRejectedError || attempt === maxAttempts - 1) {
554
+ throw error;
555
+ }
556
+ await delay(200 * (attempt + 1));
557
+ }
558
+ }
559
+ }
560
+ requiredChatCapabilityFailure(packet) {
561
+ const capability = this.bridgeChatCapability();
562
+ if (capability.availability === "unavailable" || !capability.supportsTextStreaming) {
563
+ return capability.unavailableReason ?? "Bridge Chat text capability is unavailable.";
564
+ }
565
+ if (capability.supportedProviders.length > 0
566
+ && !containsIgnoreCase(capability.supportedProviders, packet.providerKind)) {
567
+ return `Bridge Chat provider '${packet.providerKind}' is not advertised by this bridge.`;
568
+ }
569
+ if (capability.supportedModels.length > 0
570
+ && !containsIgnoreCase(capability.supportedModels, packet.model)) {
571
+ return `Bridge Chat model '${packet.model}' is not advertised by this bridge.`;
572
+ }
573
+ if (packet.requestedCapabilities.includes("cancel") && !capability.supportsCancel) {
574
+ return "Bridge Chat cancellation is not advertised by this bridge.";
575
+ }
576
+ if (packet.requestedCapabilities.includes("resume") && !capability.supportsResume) {
577
+ return "Bridge Chat resume is not advertised by this bridge.";
578
+ }
579
+ if (packet.requestedCapabilities.includes("thread_continuity")
580
+ && !capability.supportsProviderThreadContinuity) {
581
+ return "Bridge Chat provider thread continuity is not advertised by this bridge.";
582
+ }
583
+ if (packet.requestedCapabilities.includes("attachment_handoffs") && !capability.supportsAttachments) {
584
+ return "Bridge Chat attachment handoffs are not advertised by this bridge.";
585
+ }
586
+ if (packet.requestedCapabilities.includes("action_broker_grants") && !capability.supportsActionBrokerGrants) {
587
+ return "Bridge Chat action-broker grants are not advertised by this bridge.";
588
+ }
589
+ return null;
590
+ }
591
+ bridgeChatCapability() {
592
+ return bridgeChatCapabilityFor(this.config.cliProviders, this.config.providerStatuses, this.config.bridgeChat);
593
+ }
594
+ async handleDispatch(rawDispatch, reservationHandoff) {
595
+ const dispatch = await this.parseDispatchOrReport(rawDispatch);
156
596
  if (!dispatch) {
597
+ this.releaseReservedDispatchHandoff(reservationHandoff);
598
+ return;
599
+ }
600
+ if (this.enforceHostDatabaseCapabilityBoundary) {
601
+ try {
602
+ assertHostDatabaseCapabilityBoundary();
603
+ }
604
+ catch (error) {
605
+ const message = errorMessage(error);
606
+ await this.sendResult(dispatch, failureResult(message, [{
607
+ role: "runner",
608
+ kind: "error",
609
+ content: message,
610
+ eventType: "host_database_capability_forbidden",
611
+ }], failureFactsFromError(error, { source: "bridge" }))).catch(() => undefined);
612
+ this.releaseReservedDispatchHandoff(reservationHandoff);
613
+ return;
614
+ }
615
+ }
616
+ let capabilityFailure;
617
+ try {
618
+ capabilityFailure = await this.requiredCapabilityFailure(dispatch);
619
+ }
620
+ catch (error) {
621
+ this.releaseReservedDispatchHandoff(reservationHandoff);
622
+ throw error;
623
+ }
624
+ if (capabilityFailure) {
625
+ await this.sendResult(dispatch, failureResult(capabilityFailure, [{
626
+ role: "runner",
627
+ kind: "error",
628
+ content: capabilityFailure,
629
+ eventType: "docker_engine_unavailable",
630
+ }], failureFactsForCode("docker_engine_unavailable"))).catch(() => undefined);
631
+ this.releaseReservedDispatchHandoff(reservationHandoff);
157
632
  return;
158
633
  }
159
634
  const leaseId = dispatch.lease.id;
160
635
  const fencingToken = dispatch.lease.fencingToken;
161
636
  const redactionSecrets = exactSecrets(dispatch, this.config);
637
+ if (reservationHandoff && !this.reservedDispatchHandoffIsCurrent(reservationHandoff, dispatch)) {
638
+ await this.sendResult(dispatch, failureResult("Start rejected because its reservation is missing or stale.", [], failureFactsForCode("bridge.reservation_stale"))).catch(() => undefined);
639
+ return;
640
+ }
162
641
  const existing = this.activeSessions.get(leaseId);
163
642
  if (existing) {
164
643
  if (existing.fencingToken === fencingToken) {
644
+ this.releaseReservedDispatchHandoff(reservationHandoff);
165
645
  return;
166
646
  }
167
- await this.sendResult(dispatch, failureResult("Dispatch fence conflicts with an active bridge session.")).catch(() => undefined);
647
+ await this.sendResult(dispatch, failureResult("Dispatch fence conflicts with an active bridge session.", [], failureFactsForCode("bridge.dispatch_fence_conflict"))).catch(() => undefined);
648
+ this.releaseReservedDispatchHandoff(reservationHandoff);
168
649
  return;
169
650
  }
170
- const controller = new AbortController();
651
+ const controller = reservationHandoff?.controller ?? new AbortController();
171
652
  this.activeSessions.set(leaseId, {
653
+ leaseId,
654
+ kind: "dispatch",
172
655
  fencingToken,
173
656
  dispatchAttemptId: dispatch.sandbox?.dispatchAttemptId ?? null,
174
657
  runnerInstanceId: dispatch.sandbox?.runnerInstanceId ?? null,
@@ -179,6 +662,18 @@ export class ForgeBridgeService {
179
662
  : zeroSandboxResources(),
180
663
  controller
181
664
  });
665
+ if (reservationHandoff) {
666
+ // Promotion is synchronous with active-session registration. A qualified
667
+ // pre-start Cancel therefore sees either the cancellable reservation or
668
+ // the fully active owner; there is no cancellation-blind handoff gap.
669
+ this.releaseReservedDispatchHandoff(reservationHandoff);
670
+ }
671
+ // Server-requested runtime readiness is background maintenance. A probe may
672
+ // have passed admission during the narrow handoff between prepared-runtime
673
+ // verification and the foreground attempt's final admission. Preempt it as
674
+ // soon as real agent work arrives so the probe cannot consume the memory
675
+ // headroom that the foreground attempt is waiting to acquire.
676
+ this.deferActiveReadinessProbes();
182
677
  try {
183
678
  const normalized = this.normalizeDispatch(dispatch);
184
679
  if (normalized.sandbox) {
@@ -205,7 +700,10 @@ export class ForgeBridgeService {
205
700
  dispatchAttemptId: dispatch.sandbox.dispatchAttemptId,
206
701
  runnerInstanceId: dispatch.sandbox.runnerInstanceId
207
702
  } : {}),
208
- event: redactRunnerDispatchEvent(event, redactionSecrets)
703
+ event: redactRunnerDispatchEvent({ ...event, eventId: event.eventId ?? randomUUID() }, redactionSecrets)
704
+ }, {
705
+ replayCallbackTimeout: true,
706
+ signal: controller.signal,
209
707
  });
210
708
  }
211
709
  catch (error) {
@@ -223,23 +721,86 @@ export class ForgeBridgeService {
223
721
  if (normalized.sandbox?.executionMode === "docker") {
224
722
  await this.flushCleanupReports().catch(() => undefined);
225
723
  }
226
- await this.sendResult(dispatch, redactRunOnceResultPacket(result, redactionSecrets));
724
+ const redactedResult = redactRunOnceResultPacket(result, redactionSecrets);
725
+ await this.sendResult(dispatch, compactTerminalResult(dispatch, redactedResult));
227
726
  }
228
727
  catch (error) {
229
728
  const summary = controller.signal.aborted
230
729
  ? `Dispatch cancelled: ${String(controller.signal.reason ?? "cancelled")}`
231
730
  : redactText(error instanceof Error ? error.message : String(error), redactionSecrets);
232
- const failureEventType = infrastructureFailureEventType(summary);
731
+ const failureFacts = failureFactsFromError(error, {
732
+ cancelled: controller.signal.aborted,
733
+ source: "bridge",
734
+ });
735
+ const failureEventType = failureFacts.category === "unknown" ? null : failureFacts.code;
233
736
  const failureEvents = failureEventType ? [{
234
737
  role: "runner",
235
738
  kind: "error",
236
739
  content: summary,
237
740
  eventType: failureEventType,
238
741
  }] : [];
239
- await this.sendResult(dispatch, failureResult(summary, failureEvents)).catch(() => undefined);
742
+ await this.sendResult(dispatch, failureResult(summary, failureEvents, failureFacts)).catch(() => undefined);
240
743
  }
241
744
  finally {
242
- this.activeSessions.delete(leaseId);
745
+ await this.releaseActiveSessionAndRefreshCapacity(leaseId);
746
+ }
747
+ }
748
+ reservedDispatchHandoffIsCurrent(handoff, dispatch) {
749
+ const sandbox = dispatch.sandbox;
750
+ const reservation = handoff.reservation;
751
+ return sandbox !== null
752
+ && sandbox !== undefined
753
+ && sandbox.dispatchAttemptId === handoff.dispatchAttemptId
754
+ && this.reservations.get(handoff.dispatchAttemptId) === reservation
755
+ && reservation.startClaim === handoff
756
+ && !handoff.controller.signal.aborted
757
+ && reservation.leaseId === dispatch.lease.id
758
+ && reservation.fencingToken === dispatch.lease.fencingToken
759
+ && reservation.runnerInstanceId === sandbox.runnerInstanceId
760
+ && Date.parse(reservation.reservedUntil) > Date.now();
761
+ }
762
+ releaseReservedDispatchHandoff(handoff) {
763
+ if (!handoff
764
+ || this.reservations.get(handoff.dispatchAttemptId) !== handoff.reservation
765
+ || handoff.reservation.startClaim !== handoff)
766
+ return;
767
+ this.reservations.delete(handoff.dispatchAttemptId);
768
+ }
769
+ async releaseActiveSessionAndRefreshCapacity(leaseId, expectedDispatch) {
770
+ if (expectedDispatch) {
771
+ const active = this.activeSessions.get(leaseId);
772
+ // Admission failures may arrive after a newer attempt starts on the same
773
+ // lease. A terminal callback can release only the dispatch it identifies.
774
+ if (!active || active.kind !== "dispatch"
775
+ || active.fencingToken !== expectedDispatch.lease.fencingToken
776
+ || active.dispatchAttemptId !== (expectedDispatch.sandbox?.dispatchAttemptId ?? null)
777
+ || active.runnerInstanceId !== (expectedDispatch.sandbox?.runnerInstanceId ?? null))
778
+ return;
779
+ }
780
+ if (!this.activeSessions.delete(leaseId))
781
+ return;
782
+ // Publish released host/runtime capacity before the terminal result can
783
+ // enqueue the next pipeline phase. Otherwise the server may reject that
784
+ // handoff from the previous periodic heartbeat's active reservation.
785
+ // Publish the updated reservation counts from cached samples only. Probes
786
+ // belong to periodic/admission heartbeats and must not outlive this bounded
787
+ // terminal handoff. Storage recovery must not block durable result delivery.
788
+ const controller = new AbortController();
789
+ let timer;
790
+ try {
791
+ await Promise.race([
792
+ this.sendHeartbeatSnapshot(controller.signal).catch(() => undefined),
793
+ new Promise(resolve => {
794
+ timer = setTimeout(() => {
795
+ controller.abort("terminal capacity refresh deadline");
796
+ resolve();
797
+ }, Math.min(5000, positiveFiniteNumber(this.config.callbackAckTimeoutMs ?? 30_000, 30_000)));
798
+ }),
799
+ ]);
800
+ }
801
+ finally {
802
+ if (timer)
803
+ clearTimeout(timer);
243
804
  }
244
805
  }
245
806
  async handleOffer(rawDispatch) {
@@ -247,20 +808,19 @@ export class ForgeBridgeService {
247
808
  if (!envelope) {
248
809
  return;
249
810
  }
250
- const dispatch = this.parseDispatchOrReport(envelope.dispatch);
811
+ const dispatch = await this.parseDispatchOrReport(envelope.dispatch);
251
812
  if (!dispatch?.sandbox
252
813
  || dispatch.sandbox.dispatchAttemptId !== envelope.dispatchAttemptId
253
814
  || dispatch.lease.id !== envelope.leaseId
254
815
  || dispatch.lease.fencingToken !== envelope.fencingToken)
255
816
  return;
256
817
  const sandbox = dispatch.sandbox;
257
- let existing = this.reservations.get(sandbox.dispatchAttemptId);
258
- if (existing && Date.parse(existing.reservedUntil) <= Date.now()) {
259
- this.reservations.delete(sandbox.dispatchAttemptId);
260
- existing = undefined;
261
- }
818
+ this.expireReservations(Date.now());
819
+ const existing = this.reservations.get(sandbox.dispatchAttemptId);
262
820
  if (existing) {
263
- const sameFence = existing.fencingToken === dispatch.lease.fencingToken;
821
+ const sameFence = existing.leaseId === dispatch.lease.id
822
+ && existing.fencingToken === dispatch.lease.fencingToken
823
+ && existing.runnerInstanceId === sandbox.runnerInstanceId;
264
824
  await this.sendOfferDecision(dispatch, sameFence, sameFence ? null : "stale_fence", sameFence ? null : "Offer fence conflicts with the existing reservation.", sameFence ? existing.reservedUntil : null, sameFence ? existing.resources : null);
265
825
  return;
266
826
  }
@@ -287,7 +847,7 @@ export class ForgeBridgeService {
287
847
  }
288
848
  const providerNetworkFailure = sandboxProviderNetworkFailure(dispatch, this.config);
289
849
  if (providerNetworkFailure) {
290
- await this.sendOfferDecision(dispatch, false, "provider_network_unavailable", providerNetworkFailure);
850
+ await this.sendOfferDecision(dispatch, false, "provider_network_policy_denied", providerNetworkFailure);
291
851
  return;
292
852
  }
293
853
  const authFailure = sandboxAuthFailure(dispatch, this.config);
@@ -296,15 +856,12 @@ export class ForgeBridgeService {
296
856
  return;
297
857
  }
298
858
  }
299
- const runtimeState = this.sandboxRuntime.snapshot();
300
- const usedResources = this.usedSandboxResources(runtimeState.degraded);
301
- const activeResources = this.activeSandboxResources(runtimeState.degraded);
302
- const capacity = this.sandboxResourceCapacity(preflight, activeResources);
303
859
  let requestedResources = sandbox.requestedResources;
304
860
  let sandboxConfigCommitSha = sandbox.sandboxConfigCommitSha ?? null;
861
+ const normalizedDispatch = this.normalizeDispatch(dispatch);
305
862
  if (sandbox.executionMode === "docker") {
306
863
  try {
307
- const profile = await this.sandboxResourceResolver(this.normalizeDispatch(dispatch));
864
+ const profile = await this.sandboxResourceResolver(normalizedDispatch);
308
865
  requestedResources = profile.resources;
309
866
  sandboxConfigCommitSha = profile.sandboxConfigCommitSha;
310
867
  }
@@ -313,73 +870,166 @@ export class ForgeBridgeService {
313
870
  return;
314
871
  }
315
872
  }
316
- if (sandbox.executionMode === "docker"
317
- && !canReserveSandboxResources(usedResources, requestedResources, capacity)) {
318
- await this.sendOfferDecision(dispatch, false, "capacity_unavailable", sandboxCapacityFailureMessage(usedResources, requestedResources, capacity, preflight, this.config));
319
- return;
873
+ else if (normalizedDispatch.repo.serverLocalPath) {
874
+ // Host execution uses the same project-derived disk reservation as Docker.
875
+ // The API supplies a placement-time estimate; the runner can safely raise it
876
+ // from its authoritative checkout before admitting the attempt.
877
+ const measuredWorkspaceBytes = await estimateWorkspaceBytesFromRepo(normalizedDispatch.repo.serverLocalPath, { role: normalizedDispatch.execution.role }).catch(() => null);
878
+ if (measuredWorkspaceBytes !== null) {
879
+ requestedResources = {
880
+ ...requestedResources,
881
+ workspaceBytes: Math.max(requestedResources.workspaceBytes, measuredWorkspaceBytes),
882
+ };
883
+ }
884
+ }
885
+ // Resource profiling can await filesystem and Docker work. Admission needs
886
+ // a dynamic host/daemon sample that began after profiling completed, so an
887
+ // older in-flight heartbeat inspection cannot supply the commit decision.
888
+ const admissionPreflight = sandbox.executionMode === "docker"
889
+ ? await this.preflight.refreshCapacityForAdmission()
890
+ : preflight;
891
+ // Resource profiling and preflight may await slow filesystem or Docker I/O.
892
+ // Re-read every mutable capacity input only after that work, then decide and
893
+ // commit synchronously. JavaScript cannot interleave another Offer handler
894
+ // inside this non-async method, so the ledger check and insertion are one
895
+ // atomic event-loop turn without holding a lock across external I/O.
896
+ const decision = this.commitOfferReservation(dispatch, requestedResources, admissionPreflight);
897
+ await this.sendOfferDecision(dispatch, decision.accepted, decision.code, decision.message, decision.reservedUntil, decision.resources, sandboxConfigCommitSha);
898
+ }
899
+ commitOfferReservation(dispatch, requestedResources, preparedPreflight) {
900
+ const sandbox = dispatch.sandbox;
901
+ const now = Date.now();
902
+ this.expireReservations(now);
903
+ const existing = this.reservations.get(sandbox.dispatchAttemptId);
904
+ if (existing) {
905
+ const sameFence = existing.leaseId === dispatch.lease.id
906
+ && existing.fencingToken === dispatch.lease.fencingToken
907
+ && existing.runnerInstanceId === sandbox.runnerInstanceId;
908
+ return {
909
+ accepted: sameFence,
910
+ code: sameFence ? null : "stale_fence",
911
+ message: sameFence ? null : "Offer fence conflicts with the existing reservation.",
912
+ reservedUntil: sameFence ? existing.reservedUntil : null,
913
+ resources: sameFence ? existing.resources : null,
914
+ };
915
+ }
916
+ if (Date.parse(dispatch.lease.expiresAt) <= now) {
917
+ return rejectedOfferReservation("lease_expired", "Offer lease is already expired.");
320
918
  }
321
- const reservedUntil = new Date(Math.min(Date.parse(dispatch.lease.expiresAt), Date.now() + 30_000)).toISOString();
322
- this.reservations.set(sandbox.dispatchAttemptId, {
919
+ if (sandbox.executionMode === "docker") {
920
+ // A concurrent heartbeat/preflight refresh may have produced a newer
921
+ // conservative dual-domain sample while this offer was being profiled.
922
+ const admissionPreflight = this.preflight.current() ?? preparedPreflight;
923
+ if (!this.config.dockerEnabled || !admissionPreflight?.healthy) {
924
+ return rejectedOfferReservation(admissionPreflight?.unavailableCode ?? "docker_unavailable", admissionPreflight?.unavailableReason ?? "Docker sandbox preflight is unavailable.");
925
+ }
926
+ const runtimeState = this.sandboxRuntime.snapshot();
927
+ const usedResources = this.usedSandboxResources(runtimeState.degraded);
928
+ const activeResources = this.activeSandboxResources(runtimeState.degraded);
929
+ const capacity = this.sandboxResourceCapacity(admissionPreflight, activeResources);
930
+ if (!canReserveSandboxResources(usedResources, requestedResources, capacity)) {
931
+ return rejectedOfferReservation("capacity_unavailable", sandboxCapacityFailureMessage(usedResources, requestedResources, capacity, admissionPreflight, this.config));
932
+ }
933
+ }
934
+ const reservedUntil = new Date(Math.min(Date.parse(dispatch.lease.expiresAt), now + 30_000)).toISOString();
935
+ const reservation = {
323
936
  leaseId: dispatch.lease.id,
324
937
  fencingToken: dispatch.lease.fencingToken,
325
938
  runnerInstanceId: sandbox.runnerInstanceId,
326
939
  expiresAt: dispatch.lease.expiresAt,
327
940
  reservedUntil,
328
941
  sandboxed: sandbox.executionMode === "docker",
329
- resources: requestedResources
330
- });
331
- await this.sendOfferDecision(dispatch, true, null, null, reservedUntil, requestedResources, sandboxConfigCommitSha);
942
+ resources: requestedResources,
943
+ };
944
+ this.reservations.set(sandbox.dispatchAttemptId, reservation);
945
+ return {
946
+ accepted: true,
947
+ code: null,
948
+ message: null,
949
+ reservedUntil,
950
+ resources: requestedResources,
951
+ };
332
952
  }
333
953
  async handleStart(rawDispatch) {
334
954
  const envelope = parseStartEnvelope(rawDispatch);
335
955
  if (!envelope) {
336
956
  return;
337
957
  }
338
- let dispatch = this.parseDispatchOrReport(envelope.dispatch);
958
+ let dispatch = await this.parseDispatchOrReport(envelope.dispatch);
339
959
  if (!dispatch?.sandbox
340
960
  || dispatch.sandbox.dispatchAttemptId !== envelope.dispatchAttemptId
341
961
  || dispatch.lease.id !== envelope.leaseId
342
962
  || dispatch.lease.fencingToken !== envelope.fencingToken)
343
963
  return;
964
+ const active = this.activeSessions.get(dispatch.lease.id);
965
+ if (active?.kind === "dispatch"
966
+ && active.fencingToken === dispatch.lease.fencingToken
967
+ && active.dispatchAttemptId === dispatch.sandbox.dispatchAttemptId
968
+ && active.runnerInstanceId === dispatch.sandbox.runnerInstanceId) {
969
+ return;
970
+ }
344
971
  const reservation = this.reservations.get(dispatch.sandbox.dispatchAttemptId);
345
972
  if (!reservation
346
973
  || reservation.leaseId !== dispatch.lease.id
347
974
  || reservation.fencingToken !== dispatch.lease.fencingToken
348
975
  || reservation.runnerInstanceId !== dispatch.sandbox.runnerInstanceId
349
976
  || Date.parse(reservation.reservedUntil) <= Date.now()) {
350
- await this.sendResult(dispatch, failureResult("Start rejected because its reservation is missing or stale.")).catch(() => undefined);
977
+ await this.sendResult(dispatch, failureResult("Start rejected because its reservation is missing or stale.", [], failureFactsForCode("bridge.reservation_stale"))).catch(() => undefined);
351
978
  return;
352
979
  }
980
+ // Claim the exact reservation before redemption or any other post-validation
981
+ // await. A duplicate Start for the same reservation is an idempotent replay:
982
+ // it must neither redeem a one-time grant nor disturb the original owner.
983
+ if (reservation.startClaim)
984
+ return;
985
+ const startClaim = {
986
+ dispatchAttemptId: envelope.dispatchAttemptId,
987
+ reservation,
988
+ controller: new AbortController(),
989
+ };
990
+ reservation.startClaim = startClaim;
353
991
  if (dispatch.apiKeySecretGrant) {
354
- this.reservations.delete(dispatch.sandbox.dispatchAttemptId);
355
- await this.sendResult(dispatch, failureResult("Start rejected because protocol-v2 packets may not contain a plaintext provider grant.")).catch(() => undefined);
992
+ this.releaseReservedDispatchHandoff(startClaim);
993
+ await this.sendResult(dispatch, failureResult("Start rejected because protocol-v2 packets may not contain a plaintext provider grant.", [], failureFactsForCode("contract_plaintext_secret_rejected"))).catch(() => undefined);
356
994
  return;
357
995
  }
358
996
  if (dispatch.apiKeySecretGrantHandle) {
359
- const response = await this.redeemSecretGrant(dispatch);
997
+ let response;
998
+ try {
999
+ response = await this.redeemSecretGrant(dispatch);
1000
+ }
1001
+ catch (error) {
1002
+ this.releaseReservedDispatchHandoff(startClaim);
1003
+ throw error;
1004
+ }
1005
+ // Redemption may yield while a qualified Cancel releases this reservation.
1006
+ // Never launch from it or remove a successor reservation after that wait.
1007
+ if (!this.reservedDispatchHandoffIsCurrent(startClaim, dispatch)) {
1008
+ await this.sendResult(dispatch, failureResult("Start rejected because its reservation is missing or stale.", [], failureFactsForCode("bridge.reservation_stale"))).catch(() => undefined);
1009
+ return;
1010
+ }
360
1011
  const grant = response.grant;
361
1012
  if (!response.accepted
362
1013
  || !grant
363
1014
  || grant.grantId !== dispatch.apiKeySecretGrantHandle.grantId
364
1015
  || grant.provider !== dispatch.apiKeySecretGrantHandle.provider
365
1016
  || Date.parse(grant.expiresAt) <= Date.now()) {
366
- this.reservations.delete(dispatch.sandbox.dispatchAttemptId);
367
- await this.sendResult(dispatch, failureResult(response.message ?? "Secret grant redemption was rejected.")).catch(() => undefined);
1017
+ this.releaseReservedDispatchHandoff(startClaim);
1018
+ await this.sendResult(dispatch, failureResult(response.message ?? "Secret grant redemption was rejected.", [], failureFactsForCode("bridge.api_key_secret_grant_denied"))).catch(() => undefined);
368
1019
  return;
369
1020
  }
370
1021
  dispatch = { ...dispatch, apiKeySecretGrant: grant, apiKeySecretGrantHandle: null };
371
1022
  }
372
1023
  else if (dispatch.execution.requiresApiKeySecretGrant) {
373
- this.reservations.delete(dispatch.sandbox.dispatchAttemptId);
374
- await this.sendResult(dispatch, failureResult("Start rejected because the provider secret grant handle is missing.")).catch(() => undefined);
1024
+ this.releaseReservedDispatchHandoff(startClaim);
1025
+ await this.sendResult(dispatch, failureResult("Start rejected because the provider secret grant handle is missing.", [], failureFactsForCode("bridge.api_key_secret_grant_missing"))).catch(() => undefined);
375
1026
  return;
376
1027
  }
377
1028
  dispatch = {
378
1029
  ...dispatch,
379
1030
  sandbox: { ...dispatch.sandbox, requestedResources: reservation.resources },
380
1031
  };
381
- this.reservations.delete(envelope.dispatchAttemptId);
382
- await this.handleDispatch(dispatch);
1032
+ await this.handleDispatch(dispatch, startClaim);
383
1033
  }
384
1034
  async redeemSecretGrant(dispatch) {
385
1035
  const sandbox = dispatch.sandbox;
@@ -443,7 +1093,7 @@ export class ForgeBridgeService {
443
1093
  };
444
1094
  await this.invokeAck("OfferDecision", payload);
445
1095
  }
446
- parseDispatchOrReport(rawDispatch) {
1096
+ async parseDispatchOrReport(rawDispatch) {
447
1097
  try {
448
1098
  return parseRunOnceDispatchPacket(rawDispatch);
449
1099
  }
@@ -453,24 +1103,78 @@ export class ForgeBridgeService {
453
1103
  return null;
454
1104
  }
455
1105
  const summary = invalidDispatchSummary(error);
456
- void this.sendLegacyResult(identifiers.leaseId, identifiers.fencingToken, failureResult(summary, [
1106
+ await this.sendLegacyResult(identifiers.leaseId, identifiers.fencingToken, failureResult(summary, [
457
1107
  {
458
1108
  role: "runner",
459
1109
  kind: "error",
460
1110
  content: summary,
461
1111
  eventType: "invalid_dispatch"
462
1112
  }
463
- ])).catch(() => undefined);
1113
+ ], failureFactsForCode("invalid_dispatch"))).catch(() => undefined);
1114
+ return null;
1115
+ }
1116
+ }
1117
+ parseChatTurnOrReport(rawDispatch) {
1118
+ try {
1119
+ return parseChatTurnDispatchPacket(rawDispatch, this.bridgeChatCapability());
1120
+ }
1121
+ catch (error) {
1122
+ logBridgeStatus(invalidChatTurnSummary(error));
464
1123
  return null;
465
1124
  }
466
1125
  }
467
1126
  async handleCancel(request) {
1127
+ if (request.dispatchAttemptId !== undefined && request.dispatchAttemptId !== null) {
1128
+ // Qualified cancellation is reservation-only. In particular, a delayed
1129
+ // pre-start Cancel must not abort an active session if Start won the race.
1130
+ const reservation = this.reservations.get(request.dispatchAttemptId);
1131
+ if (reservation
1132
+ && request.runnerId === this.config.runnerId
1133
+ && request.runnerInstanceId === this.config.runnerInstanceId
1134
+ && reservation.runnerInstanceId === request.runnerInstanceId
1135
+ && reservation.leaseId === request.leaseId
1136
+ && reservation.fencingToken === request.fencingToken) {
1137
+ reservation.startClaim?.controller.abort(request.reason ?? "cancelled by server");
1138
+ this.reservations.delete(request.dispatchAttemptId);
1139
+ }
1140
+ return;
1141
+ }
468
1142
  const active = this.activeSessions.get(request.leaseId);
469
1143
  if (!active || active.fencingToken !== request.fencingToken) {
470
1144
  return;
471
1145
  }
1146
+ if (request.runnerId && request.runnerId !== this.config.runnerId) {
1147
+ return;
1148
+ }
1149
+ if (request.runnerInstanceId && request.runnerInstanceId !== (active.runnerInstanceId ?? this.config.runnerInstanceId)) {
1150
+ return;
1151
+ }
1152
+ if (active.kind === "chat" && !this.chatCancelMatchesActiveTurn(active, request)) {
1153
+ return;
1154
+ }
472
1155
  active.controller.abort(request.reason ?? "cancelled by server");
473
1156
  }
1157
+ chatCancelMatchesActiveTurn(active, request) {
1158
+ if (request.protocolVersion && request.protocolVersion !== bridgeChatProtocolVersion) {
1159
+ return false;
1160
+ }
1161
+ try {
1162
+ assertBridgeChatCallbackPayload(request.payload, "cancel payload");
1163
+ }
1164
+ catch {
1165
+ return false;
1166
+ }
1167
+ if (request.chatRunId && request.chatRunId !== active.chatRunId)
1168
+ return false;
1169
+ if (request.conversationId && request.conversationId !== active.conversationId)
1170
+ return false;
1171
+ if (request.turnId && request.turnId !== active.turnId)
1172
+ return false;
1173
+ if (request.providerThreadId && active.providerThreadId && request.providerThreadId !== active.providerThreadId) {
1174
+ return false;
1175
+ }
1176
+ return true;
1177
+ }
474
1178
  async handleReap(request) {
475
1179
  if (request.runnerId !== this.config.runnerId
476
1180
  || request.runnerInstanceId !== this.config.runnerInstanceId)
@@ -492,32 +1196,43 @@ export class ForgeBridgeService {
492
1196
  timer.unref?.();
493
1197
  let result;
494
1198
  try {
495
- if (!this.sessions.probeHostRuntime) {
496
- throw new Error("host_runtime_probe_unavailable: this bridge does not implement runtime readiness probes.");
497
- }
498
- const probe = await this.sessions.probeHostRuntime({
499
- ...request,
500
- runnerId: this.config.runnerId,
501
- repoRoot: this.config.defaultRepoRoot,
502
- cleanupWorktrees: this.config.cleanupWorktrees,
503
- }, controller.signal);
504
- result = {
1199
+ const accepted = {
505
1200
  schemaVersion: 1,
506
1201
  requestId: request.requestId,
507
1202
  fencingToken: request.fencingToken,
508
1203
  runnerInstanceId: this.config.runnerInstanceId,
509
- ...probe,
1204
+ acceptedAt: new Date().toISOString(),
510
1205
  };
1206
+ await this.invokeAck("ReadinessAccepted", accepted);
1207
+ if (this.activeSessions.size > 0) {
1208
+ result = deferredReadinessResult(request, this.config.runnerInstanceId);
1209
+ }
1210
+ else if (!this.sessions.probeHostRuntime) {
1211
+ throw new Error("host_runtime_probe_unavailable: this bridge does not implement runtime readiness probes.");
1212
+ }
1213
+ else {
1214
+ const probe = await this.sessions.probeHostRuntime(this.normalizeReadinessProbe(request), controller.signal);
1215
+ result = readinessDeferralReason(controller.signal)
1216
+ ? deferredReadinessResult(request, this.config.runnerInstanceId)
1217
+ : {
1218
+ schemaVersion: 1,
1219
+ requestId: request.requestId,
1220
+ fencingToken: request.fencingToken,
1221
+ runnerInstanceId: this.config.runnerInstanceId,
1222
+ ...probe,
1223
+ };
1224
+ }
511
1225
  }
512
1226
  catch (error) {
1227
+ const deferredReason = readinessDeferralReason(controller.signal);
513
1228
  result = {
514
1229
  schemaVersion: 1,
515
1230
  requestId: request.requestId,
516
1231
  fencingToken: request.fencingToken,
517
1232
  runnerInstanceId: this.config.runnerInstanceId,
518
1233
  status: "failed",
519
- reasonCode: "host_runtime_probe_failed",
520
- reasonMessage: errorMessage(error).slice(0, 1000),
1234
+ reasonCode: deferredReason ? "host_runtime_probe_deferred" : "host_runtime_probe_failed",
1235
+ reasonMessage: (deferredReason ?? errorMessage(error)).slice(0, 1000),
521
1236
  checkedAt: new Date().toISOString(),
522
1237
  };
523
1238
  }
@@ -525,9 +1240,69 @@ export class ForgeBridgeService {
525
1240
  clearTimeout(timer);
526
1241
  this.activeReadinessProbes.delete(request.requestId);
527
1242
  }
528
- await this.invokeAck("ReadinessResult", result).catch(() => undefined);
1243
+ try {
1244
+ await this.readinessResultOutbox.persist(result, request.expiresAt);
1245
+ }
1246
+ catch (error) {
1247
+ if (errorMessage(error).includes("runtime_readiness_result_conflict:")) {
1248
+ logBridgeStatus(`Readiness result ${request.requestId} rejected conflicting reuse of a durable request identity.`);
1249
+ await this.sendHeartbeat().catch(() => undefined);
1250
+ return;
1251
+ }
1252
+ logBridgeStatus(`Readiness result ${request.requestId} could not be persisted; attempting direct delivery: ${errorMessage(error)}`);
1253
+ await this.invokeAck("ReadinessResult", result).catch(deliveryError => {
1254
+ logBridgeStatus(`Readiness result ${request.requestId} could not be persisted or delivered: ${errorMessage(deliveryError)}`);
1255
+ });
1256
+ await this.sendHeartbeat().catch(() => undefined);
1257
+ return;
1258
+ }
1259
+ try {
1260
+ await this.flushReadinessResults();
1261
+ }
1262
+ catch (error) {
1263
+ logBridgeStatus(`Readiness result ${request.requestId} remains queued for replay: ${errorMessage(error)}`);
1264
+ }
529
1265
  await this.sendHeartbeat().catch(() => undefined);
530
1266
  }
1267
+ deferActiveReadinessProbes() {
1268
+ for (const controller of this.activeReadinessProbes.values()) {
1269
+ if (!controller.signal.aborted) {
1270
+ controller.abort("host_runtime_probe_deferred: foreground agent work took priority over background runtime readiness.");
1271
+ }
1272
+ }
1273
+ }
1274
+ async flushReadinessResults() {
1275
+ if (this.readinessResultFlush)
1276
+ return await this.readinessResultFlush;
1277
+ const flush = this.flushReadinessResultsCore();
1278
+ this.readinessResultFlush = flush;
1279
+ try {
1280
+ await flush;
1281
+ }
1282
+ finally {
1283
+ if (this.readinessResultFlush === flush)
1284
+ this.readinessResultFlush = null;
1285
+ }
1286
+ }
1287
+ async flushReadinessResultsCore() {
1288
+ for (const pending of await this.readinessResultOutbox.pending()) {
1289
+ if (Date.parse(pending.expiresAt) <= Date.now()) {
1290
+ await this.readinessResultOutbox.acknowledge(pending.result.requestId);
1291
+ continue;
1292
+ }
1293
+ try {
1294
+ await this.invokeAck("ReadinessResult", pending.result);
1295
+ await this.readinessResultOutbox.acknowledge(pending.result.requestId);
1296
+ }
1297
+ catch (error) {
1298
+ if (isReadinessNoLongerPending(error)) {
1299
+ await this.readinessResultOutbox.acknowledge(pending.result.requestId);
1300
+ continue;
1301
+ }
1302
+ throw error;
1303
+ }
1304
+ }
1305
+ }
531
1306
  abortActiveSessions(reason) {
532
1307
  for (const active of this.activeSessions.values()) {
533
1308
  active.controller.abort(reason);
@@ -539,8 +1314,42 @@ export class ForgeBridgeService {
539
1314
  controller.abort(reason);
540
1315
  this.activeReadinessProbes.clear();
541
1316
  }
1317
+ normalizeReadinessProbe(request) {
1318
+ const mapping = findRepoMapping({
1319
+ id: request.repoId,
1320
+ name: request.repoName,
1321
+ remoteUrl: request.remoteUrl,
1322
+ }, repoMappingsFor(this.config));
1323
+ const configuredPath = mapping?.localPath?.trim();
1324
+ if (!configuredPath) {
1325
+ throw new Error("bridge_repo_mapping_missing: readiness requires a matching local repository mapping.");
1326
+ }
1327
+ // Local configuration, not a path supplied by the server, grants checkout
1328
+ // authority. Equality also excludes a different child/sibling of this repo.
1329
+ const suppliedPath = request.runnerRepoLocalPath.trim();
1330
+ if (!path.isAbsolute(suppliedPath)
1331
+ || (process.platform === "win32" && path.parse(suppliedPath).root.length === 1)) {
1332
+ throw new Error("bridge_repo_mapping_mismatch: readiness requires a fully qualified checkout path.");
1333
+ }
1334
+ const localPath = path.resolve(configuredPath);
1335
+ const requestedPath = path.resolve(suppliedPath);
1336
+ const pathsMatch = process.platform === "win32"
1337
+ ? localPath.toLowerCase() === requestedPath.toLowerCase()
1338
+ : localPath === requestedPath;
1339
+ if (!pathsMatch) {
1340
+ throw new Error("bridge_repo_mapping_mismatch: readiness checkout differs from the locally configured repository.");
1341
+ }
1342
+ return {
1343
+ ...request,
1344
+ runnerId: this.config.runnerId,
1345
+ runnerRepoLocalPath: localPath,
1346
+ cloneOnDemand: false,
1347
+ repoRoot: path.dirname(localPath),
1348
+ cleanupWorktrees: this.config.cleanupWorktrees,
1349
+ };
1350
+ }
542
1351
  normalizeDispatch(dispatch) {
543
- const mapping = findRepoMapping(dispatch, repoMappingsFor(this.config));
1352
+ const mapping = findRepoMapping(dispatch.repo, repoMappingsFor(this.config));
544
1353
  const localPath = mapping?.localPath?.trim();
545
1354
  const repoRoot = localPath ? path.dirname(path.resolve(localPath)) : this.config.defaultRepoRoot;
546
1355
  const qaLaunchTarget = normalizeQaAppLaunchTarget(mapping?.qaAppLaunchTarget) ?? dispatch.qaBrowser?.launchTarget ?? null;
@@ -569,24 +1378,105 @@ export class ForgeBridgeService {
569
1378
  fencingToken: dispatch.lease.fencingToken,
570
1379
  dispatchAttemptId: dispatch.sandbox?.dispatchAttemptId ?? null,
571
1380
  runnerInstanceId: dispatch.sandbox?.runnerInstanceId ?? null,
572
- result
1381
+ // Every terminal writer, including early admission failures and exception
1382
+ // paths, crosses the same redaction boundary before outbox persistence.
1383
+ result: redactRunOnceResultPacket(result, exactSecrets(dispatch, this.config))
573
1384
  };
574
- await this.sendResultPayload(payload);
1385
+ await this.sendResultPayload(payload, () => this.releaseActiveSessionAndRefreshCapacity(dispatch.lease.id, dispatch));
575
1386
  }
576
1387
  async sendLegacyResult(leaseId, fencingToken, result) {
577
1388
  await this.sendResultPayload({ leaseId, fencingToken, result });
578
1389
  }
579
- async sendResultPayload(payload) {
1390
+ async sendResultPayload(payload, beforeDelivery) {
1391
+ // Register before persistence: replay can observe a record as soon as the
1392
+ // durable write becomes visible, even before persist() has returned.
1393
+ let releaseHandoff;
1394
+ const priorHandoff = this.dispatchResultHandoffs.get(payload.leaseId);
1395
+ const handoff = new Promise(resolve => { releaseHandoff = resolve; });
1396
+ this.dispatchResultHandoffs.set(payload.leaseId, handoff);
1397
+ let persisted = false;
1398
+ try {
1399
+ try {
1400
+ await this.dispatchResultOutbox.persist(payload);
1401
+ persisted = true;
1402
+ this.dispatchResultGeneration += 1;
1403
+ }
1404
+ catch (error) {
1405
+ logBridgeStatus(`Dispatch result ${payload.leaseId} could not be persisted; attempting direct delivery: ${errorMessage(error)}`);
1406
+ }
1407
+ await priorHandoff;
1408
+ await beforeDelivery?.();
1409
+ }
1410
+ finally {
1411
+ releaseHandoff();
1412
+ if (this.dispatchResultHandoffs.get(payload.leaseId) === handoff)
1413
+ this.dispatchResultHandoffs.delete(payload.leaseId);
1414
+ }
1415
+ if (!persisted) {
1416
+ await this.sendResultPayloadDirect(payload);
1417
+ return;
1418
+ }
1419
+ await this.flushDispatchResults().catch(error => {
1420
+ logBridgeStatus(`Dispatch result ${payload.leaseId} remains queued for replay: ${errorMessage(error)}`);
1421
+ });
1422
+ }
1423
+ async sendResultPayloadDirect(payload) {
580
1424
  try {
581
1425
  await this.invokeAck("DispatchResult", payload);
582
1426
  }
583
1427
  catch (error) {
584
- if (isDispatchNoLongerPending(error)) {
1428
+ if (isDispatchNoLongerPending(error))
585
1429
  return;
586
- }
587
1430
  throw error;
588
1431
  }
589
1432
  }
1433
+ async flushDispatchResults(force = false) {
1434
+ if (!force && this.dispatchResultGeneration === this.dispatchResultFlushedGeneration)
1435
+ return;
1436
+ if (this.dispatchResultFlush) {
1437
+ await this.dispatchResultFlush;
1438
+ if (!force && this.dispatchResultGeneration === this.dispatchResultFlushedGeneration)
1439
+ return;
1440
+ }
1441
+ const flush = this.flushDispatchResultsCore();
1442
+ this.dispatchResultFlush = flush;
1443
+ try {
1444
+ await flush;
1445
+ }
1446
+ finally {
1447
+ if (this.dispatchResultFlush === flush)
1448
+ this.dispatchResultFlush = null;
1449
+ }
1450
+ }
1451
+ async flushDispatchResultsCore() {
1452
+ const startedGeneration = this.dispatchResultGeneration;
1453
+ while (true) {
1454
+ const records = await this.dispatchResultOutbox.pending();
1455
+ for (const pending of records) {
1456
+ // Periodic heartbeats and reconnect replay share the same release
1457
+ // barrier as the terminal writer. Restart replay has no live handoff.
1458
+ while (this.dispatchResultHandoffs.has(pending.payload.leaseId)) {
1459
+ await this.dispatchResultHandoffs.get(pending.payload.leaseId);
1460
+ }
1461
+ try {
1462
+ await this.sendResultPayloadDirect(pending.payload);
1463
+ await this.dispatchResultOutbox.acknowledge(pending.payload.leaseId);
1464
+ }
1465
+ catch (error) {
1466
+ if (isDispatchNoLongerPending(error)) {
1467
+ await this.dispatchResultOutbox.acknowledge(pending.payload.leaseId);
1468
+ continue;
1469
+ }
1470
+ throw error;
1471
+ }
1472
+ }
1473
+ if (records.length < 3)
1474
+ break;
1475
+ }
1476
+ if (this.dispatchResultGeneration === startedGeneration) {
1477
+ this.dispatchResultFlushedGeneration = startedGeneration;
1478
+ }
1479
+ }
590
1480
  startHeartbeat() {
591
1481
  this.stopHeartbeat();
592
1482
  this.consecutiveHeartbeatFailures = 0;
@@ -605,6 +1495,8 @@ export class ForgeBridgeService {
605
1495
  return;
606
1496
  this.heartbeatInFlight = true;
607
1497
  try {
1498
+ await this.flushDispatchResults();
1499
+ await this.flushReadinessResults();
608
1500
  await this.sendHeartbeat();
609
1501
  this.consecutiveHeartbeatFailures = 0;
610
1502
  }
@@ -637,17 +1529,20 @@ export class ForgeBridgeService {
637
1529
  }
638
1530
  expireLeases() {
639
1531
  const now = Date.now();
640
- for (const [attemptId, reservation] of this.reservations) {
641
- if (Date.parse(reservation.expiresAt) <= now || Date.parse(reservation.reservedUntil) <= now) {
642
- this.reservations.delete(attemptId);
643
- }
644
- }
1532
+ this.expireReservations(now);
645
1533
  for (const active of this.activeSessions.values()) {
646
1534
  if (Date.parse(active.leaseExpiresAt) <= now && !active.controller.signal.aborted) {
647
1535
  active.controller.abort("dispatch lease expired");
648
1536
  }
649
1537
  }
650
1538
  }
1539
+ expireReservations(now) {
1540
+ for (const [attemptId, reservation] of this.reservations) {
1541
+ if (Date.parse(reservation.expiresAt) <= now || Date.parse(reservation.reservedUntil) <= now) {
1542
+ this.reservations.delete(attemptId);
1543
+ }
1544
+ }
1545
+ }
651
1546
  stopHeartbeat() {
652
1547
  if (this.heartbeatTimer) {
653
1548
  clearInterval(this.heartbeatTimer);
@@ -676,6 +1571,7 @@ export class ForgeBridgeService {
676
1571
  runnerId: this.config.runnerId,
677
1572
  runnerInstanceId: this.config.runnerInstanceId,
678
1573
  });
1574
+ this.rememberRecoveredHostAttempts(hostReap);
679
1575
  if (hostReap && (hostReap.reaped > 0 || hostReap.degraded > 0)) {
680
1576
  logBridgeStatus(`Managed-host maintenance reaped ${hostReap.reaped} stale attempt(s) and left ${hostReap.degraded} degraded attempt(s) for reconciliation.`);
681
1577
  }
@@ -716,22 +1612,50 @@ export class ForgeBridgeService {
716
1612
  }
717
1613
  }
718
1614
  async sendHeartbeat() {
1615
+ // Heartbeats are the recovery path for capacity-blocked work. Refresh the
1616
+ // dynamic sample before advertising it, while reusing the static Docker
1617
+ // capability snapshot established at startup or by an offer.
1618
+ if (this.config.dockerEnabled)
1619
+ await this.preflight.refreshCapacity();
1620
+ // A due probe can start a Docker Desktop recovery that runs for minutes.
1621
+ // Wait only as long as a normal probe takes, then advertise the last
1622
+ // snapshot so the server doesn't mark the runner offline meanwhile.
1623
+ const probeWait = new AbortController();
1624
+ await Promise.race([
1625
+ this.dockerEngineProbe.refresh().catch(() => undefined),
1626
+ delay(positiveFiniteNumber(this.config.heartbeatDockerProbeWaitMs ?? 15_000, 15_000), probeWait.signal),
1627
+ ]);
1628
+ probeWait.abort();
1629
+ await this.sendHeartbeatSnapshot();
1630
+ }
1631
+ async sendHeartbeatSnapshot(signal) {
719
1632
  await this.invokeAck("Heartbeat", {
720
1633
  status: "online",
721
1634
  capabilities: this.capabilities(),
722
1635
  version: this.config.version,
723
1636
  protocolVersion: this.config.protocolVersion
724
- });
1637
+ }, { signal });
725
1638
  }
726
1639
  capabilities() {
727
1640
  const preflight = this.preflight.current();
1641
+ const dockerEngine = this.dockerEngineProbe.current();
728
1642
  const runtimeState = this.sandboxRuntime.snapshot();
729
1643
  const reserved = [...this.reservations.values()].filter(reservation => reservation.sandboxed).length;
730
1644
  const used = this.usedSandboxResources(runtimeState.degraded);
731
1645
  const activeResources = this.activeSandboxResources(runtimeState.degraded);
732
1646
  const capacity = this.sandboxResourceCapacity(preflight, activeResources);
1647
+ const memoryAdmission = preflight ? buildMemoryAdmissionDiagnostics({
1648
+ sampledAt: preflight.sampledAt,
1649
+ totalMemoryBytes: preflight.totalMemoryBytes,
1650
+ rawAvailableMemoryBytes: preflight.rawAvailableMemoryBytes,
1651
+ stabilizedAvailableMemoryBytes: preflight.availableMemoryBytes,
1652
+ controlPlaneMemoryBytes: process.memoryUsage().rss,
1653
+ reserveMemoryBytes: preflight.hostReserveBytes,
1654
+ committedMemoryBytes: used.memoryBytes,
1655
+ pendingCommitmentMemoryBytes: Math.max(0, used.memoryBytes - activeResources.memoryBytes),
1656
+ }) : null;
733
1657
  return {
734
- schemaVersion: 3,
1658
+ schemaVersion: 6,
735
1659
  activeSessionCount: this.activeSessions.size,
736
1660
  cliProviders: this.config.cliProviders,
737
1661
  providerStatuses: this.config.providerStatuses,
@@ -741,6 +1665,19 @@ export class ForgeBridgeService {
741
1665
  sandboxProfiles: this.config.sandboxProfiles,
742
1666
  runnerInstanceId: this.config.runnerInstanceId,
743
1667
  trustDomainId: this.config.trustDomainId,
1668
+ executionFacilities: dockerEngine?.available ? ["docker_engine"] : [],
1669
+ dockerEngine,
1670
+ toolchains: {
1671
+ ...this.hostCapabilities.toolchains,
1672
+ docker: {
1673
+ available: dockerEngine?.available === true,
1674
+ executable: this.config.dockerExecutable,
1675
+ version: dockerEngine?.runtimeVersion ?? null,
1676
+ },
1677
+ },
1678
+ playwrightProjects: this.hostCapabilities.playwrightProjects,
1679
+ browserLaunch: this.hostCapabilities.browserLaunch,
1680
+ providerConnectivity: this.hostCapabilities.providerConnectivity,
744
1681
  maxConcurrentSandboxes: capacity.sandboxSlots,
745
1682
  reservedSandboxCount: reserved,
746
1683
  activeSandboxCount: runtimeState.active,
@@ -752,7 +1689,7 @@ export class ForgeBridgeService {
752
1689
  allowsOpenPublicSandbox: this.config.allowOpenPublicSandbox,
753
1690
  sandboxPublicationModes: [this.config.sandboxPublicationMode ?? "shared_filesystem"],
754
1691
  sandboxAuthModes: sandboxAuthModesFor(this.config),
755
- availableMemoryBytes: preflight ? Math.max(0, capacity.memoryBytes - used.memoryBytes) : null,
1692
+ availableMemoryBytes: memoryAdmission?.availableForAdmissionBytes ?? null,
756
1693
  availableCpuMillis: preflight ? Math.max(0, capacity.cpuMillis - used.cpuMillis) : null,
757
1694
  availableWorkspaceBytes: preflight ? Math.max(0, capacity.workspaceBytes - used.workspaceBytes) : null,
758
1695
  diskLow: preflight?.diskLow ?? null,
@@ -766,8 +1703,11 @@ export class ForgeBridgeService {
766
1703
  hostReserveBytes: preflight.hostReserveBytes,
767
1704
  activeReservationBytes: used.memoryBytes,
768
1705
  daemonTotalBytes: preflight.totalMemoryBytes,
769
- daemonAvailableBytes: preflight.availableMemoryBytes,
770
- availableForAdmissionBytes: Math.max(0, capacity.memoryBytes - used.memoryBytes),
1706
+ daemonAvailableBytes: preflight.rawDaemonAvailableMemoryBytes ?? 0,
1707
+ hostAvailableBytes: preflight.rawHostAvailableMemoryBytes,
1708
+ conservativeAvailableBytes: preflight.rawAvailableMemoryBytes,
1709
+ availableForAdmissionBytes: memoryAdmission.availableForAdmissionBytes,
1710
+ memoryAdmission: memoryAdmission,
771
1711
  } : null,
772
1712
  hostRuntime: this.sessions.runtimeCapacity?.() ?? null,
773
1713
  supportsApiSecretGrant: this.config.supportsApiSecretGrant,
@@ -776,9 +1716,20 @@ export class ForgeBridgeService {
776
1716
  supportsRunnerAuthoredPhaseArtifacts: this.config.supportsRunnerAuthoredPhaseArtifacts,
777
1717
  supportsDraftRunOnce: this.config.supportsDraftRunOnce,
778
1718
  supportsDraftAttachments: this.config.supportsDraftAttachments,
779
- supportsDraftImageAttachments: this.config.supportsDraftImageAttachments
1719
+ supportsDraftImageAttachments: this.config.supportsDraftImageAttachments,
1720
+ bridgeChat: bridgeChatCapabilityFor(this.config.cliProviders, this.config.providerStatuses, this.config.bridgeChat)
780
1721
  };
781
1722
  }
1723
+ async requiredCapabilityFailure(dispatch) {
1724
+ const requiredCapabilities = dispatch.qaScope?.required_capabilities ?? [];
1725
+ if (!requiredCapabilities.includes("docker_engine"))
1726
+ return null;
1727
+ const executionMode = dispatch.sandbox?.executionMode;
1728
+ if (executionMode === "docker") {
1729
+ return dockerEngineRequirementFailure(requiredCapabilities, executionMode, null);
1730
+ }
1731
+ return dockerEngineRequirementFailure(requiredCapabilities, executionMode, await this.dockerEngineProbe.refresh(true));
1732
+ }
782
1733
  usedSandboxResources(degradedReservations) {
783
1734
  const resources = [
784
1735
  ...[...this.reservations.values()]
@@ -811,9 +1762,7 @@ export class ForgeBridgeService {
811
1762
  workerMemoryLimitBytes: 0,
812
1763
  serviceMemoryLimitBytes: 0,
813
1764
  runtimeOverheadBytes: 0,
814
- memoryBytes: Math.max(0, preflight.availableMemoryBytes
815
- - (this.config.sandboxHostMemoryReserveBytes ?? 2 * 1024 ** 3)
816
- + activeResources.memoryBytes),
1765
+ memoryBytes: sandboxMemoryCommitmentCapacity(preflight, activeResources.memoryBytes),
817
1766
  workspaceBytes: Math.max(0, preflight.availableWorkspaceBytes
818
1767
  - this.config.minimumWorkspaceBytes
819
1768
  + activeResources.workspaceBytes),
@@ -821,22 +1770,96 @@ export class ForgeBridgeService {
821
1770
  buildSlots: this.config.maxConcurrentSandboxBuilds
822
1771
  };
823
1772
  }
824
- async invokeAck(methodName, payload) {
825
- const connection = this.connection;
826
- if (!connection) {
827
- throw new Error("Bridge is not connected.");
1773
+ async invokeAck(methodName, payload, options = {}) {
1774
+ let storageRecoveryAttempt = 0;
1775
+ let callbackTimeoutAttempt = 0;
1776
+ while (true) {
1777
+ if (options.signal?.aborted) {
1778
+ throw new Error(String(options.signal.reason ?? `${methodName} callback was cancelled.`));
1779
+ }
1780
+ const connection = this.connection;
1781
+ if (!connection) {
1782
+ throw new Error("Bridge is not connected.");
1783
+ }
1784
+ await this.waitForTransportRecovery(connection, this.transportRecovery, options.signal);
1785
+ const recoveryGeneration = this.transportRecoveryGeneration;
1786
+ try {
1787
+ const ack = await waitForCallbackAck(connection.invoke(methodName, payload), methodName, positiveFiniteNumber(this.config.callbackAckTimeoutMs ?? 30_000, 30_000), options.signal);
1788
+ if (!ack.accepted) {
1789
+ throw new BridgeCallbackRejectedError(ack.code ?? null, ack.message ?? ack.code ?? `${methodName} callback was rejected.`);
1790
+ }
1791
+ return ack;
1792
+ }
1793
+ catch (error) {
1794
+ if (options.signal?.aborted)
1795
+ throw error;
1796
+ if (connection !== this.connection || this.stopped)
1797
+ throw error;
1798
+ const recovery = this.transportRecovery;
1799
+ if (recovery && recovery.generation > recoveryGeneration) {
1800
+ await this.waitForTransportRecovery(connection, recovery, options.signal);
1801
+ continue;
1802
+ }
1803
+ if (isProviderStorageUnavailable(error)) {
1804
+ const backoff = this.config.reconnectBackoffMs[Math.min(storageRecoveryAttempt, this.config.reconnectBackoffMs.length - 1)] ?? 5000;
1805
+ storageRecoveryAttempt += 1;
1806
+ logBridgeStatus(`${methodName} reached Forge while storage was recovering. Retrying callback ${storageRecoveryAttempt} in ${formatDelay(backoff)}...`);
1807
+ await delay(backoff, options.signal);
1808
+ continue;
1809
+ }
1810
+ if (options.replayCallbackTimeout && isBridgeCallbackTimeout(error)) {
1811
+ const backoff = this.config.reconnectBackoffMs[Math.min(callbackTimeoutAttempt, this.config.reconnectBackoffMs.length - 1)] ?? 5000;
1812
+ callbackTimeoutAttempt += 1;
1813
+ logBridgeStatus(`${methodName} acknowledgement timed out. Replaying the idempotent callback ${callbackTimeoutAttempt} in ${formatDelay(backoff)}...`);
1814
+ await delay(backoff, options.signal);
1815
+ continue;
1816
+ }
1817
+ throw error;
1818
+ }
828
1819
  }
829
- const ack = await connection.invoke(methodName, payload);
830
- if (!ack.accepted) {
831
- throw new BridgeCallbackRejectedError(ack.code ?? null, ack.message ?? ack.code ?? `${methodName} callback was rejected.`);
1820
+ }
1821
+ beginTransportRecovery() {
1822
+ if (this.transportRecovery && !this.transportRecovery.completed)
1823
+ return;
1824
+ let resolve = () => undefined;
1825
+ const promise = new Promise((complete) => { resolve = complete; });
1826
+ this.transportRecoveryGeneration += 1;
1827
+ this.transportRecovery = {
1828
+ generation: this.transportRecoveryGeneration,
1829
+ promise,
1830
+ resolve,
1831
+ completed: false,
1832
+ error: null,
1833
+ };
1834
+ }
1835
+ completeTransportRecovery(error = null) {
1836
+ const recovery = this.transportRecovery;
1837
+ if (!recovery || recovery.completed)
1838
+ return;
1839
+ recovery.completed = true;
1840
+ recovery.error = error;
1841
+ recovery.resolve();
1842
+ }
1843
+ async waitForTransportRecovery(connection, recovery, signal) {
1844
+ if (!recovery)
1845
+ return;
1846
+ if (recovery.completed) {
1847
+ if (recovery.error)
1848
+ throw recovery.error;
1849
+ return;
832
1850
  }
833
- return ack;
1851
+ await waitForPromiseOrAbort(recovery.promise, signal);
1852
+ signal?.throwIfAborted();
1853
+ if (connection !== this.connection)
1854
+ throw new Error("Bridge connection changed while recovering.");
1855
+ if (recovery.error)
1856
+ throw recovery.error;
834
1857
  }
835
1858
  async waitForClose(signal) {
836
1859
  if (!this.closed) {
837
1860
  return;
838
1861
  }
839
- await Promise.race([this.closed, abortPromise(signal)]);
1862
+ await waitForPromiseOrAbort(this.closed, signal);
840
1863
  }
841
1864
  async stopConnection() {
842
1865
  this.stopHeartbeat();
@@ -862,11 +1885,23 @@ function isDispatchNoLongerPending(error) {
862
1885
  }
863
1886
  return error instanceof Error && error.message === "Dispatch is no longer pending.";
864
1887
  }
865
- function createDefaultSessionService(sandboxStateRoot) {
1888
+ function isReadinessNoLongerPending(error) {
1889
+ return error instanceof BridgeCallbackRejectedError
1890
+ && (error.code === "bridge.stale_readiness_probe" || error.code === "bridge.stale_runner_instance");
1891
+ }
1892
+ function isProviderStorageUnavailable(error) {
1893
+ const message = errorMessage(error).toLowerCase();
1894
+ return message.includes("provider_storage_unavailable")
1895
+ || (message.includes("storage is recovering") && message.includes("503"));
1896
+ }
1897
+ function isBridgeCallbackTimeout(error) {
1898
+ return errorMessage(error).startsWith("bridge_callback_timeout:");
1899
+ }
1900
+ function createDefaultSessionService(sandboxStateRoot, sandboxExecutionRoot, dockerEngineProbe) {
866
1901
  const rawProcessRunner = new StreamingProviderProcessRunner();
867
1902
  const processSupervisors = new AttemptProcessSupervisorRegistry();
868
1903
  const coordinatedProcessRunner = new AttemptCoordinatedProviderProcessRunner(processSupervisors, rawProcessRunner);
869
- return new BridgeSessionService(new GitWorktreeService(undefined, new RepositoryCoordinatorLock(sandboxStateRoot)), new RoutingProviderCliService(new ProcessBackedProviderCliService(coordinatedProcessRunner)), new HostRuntimeLeaseManager(sandboxStateRoot, undefined, undefined, processSupervisors, rawProcessRunner));
1904
+ return new BridgeSessionService(new GitWorktreeService(undefined, new RepositoryCoordinatorLock(sandboxStateRoot)), new RoutingProviderCliService(new ProcessBackedProviderCliService(coordinatedProcessRunner, undefined, undefined, undefined, new PrePublicationValidator(undefined, dockerEngineProbe, readNativeTestPostgresFallback))), new HostRuntimeLeaseManager(sandboxStateRoot, undefined, undefined, processSupervisors, rawProcessRunner, undefined, undefined, undefined, undefined, undefined, sandboxExecutionRoot));
870
1905
  }
871
1906
  export class SignalRBridgeConnectionFactory {
872
1907
  create(config) {
@@ -890,10 +1925,11 @@ export class SignalRBridgeConnection {
890
1925
  Authorization: `Bearer ${config.deviceToken}`
891
1926
  }
892
1927
  })
1928
+ .withAutomaticReconnect([...config.reconnectBackoffMs])
893
1929
  .configureLogging(LogLevel.Warning)
894
1930
  .build();
895
1931
  this.connection.on("Superseded", () => {
896
- void this.connection.stop();
1932
+ void this.connection.stop().catch(() => undefined);
897
1933
  });
898
1934
  }
899
1935
  start() {
@@ -905,11 +1941,24 @@ export class SignalRBridgeConnection {
905
1941
  onClose(handler) {
906
1942
  this.connection.onclose(handler);
907
1943
  }
1944
+ onReconnecting(handler) {
1945
+ this.connection.onreconnecting(handler);
1946
+ }
1947
+ onReconnected(handler) {
1948
+ this.connection.onreconnected((connectionId) => {
1949
+ void Promise.resolve(handler(connectionId)).catch(() => undefined);
1950
+ });
1951
+ }
908
1952
  onDispatch(handler) {
909
1953
  this.connection.on("Dispatch", (raw) => {
910
1954
  void handler(raw).catch(() => undefined);
911
1955
  });
912
1956
  }
1957
+ onChatTurn(handler) {
1958
+ this.connection.on("ChatTurn", (raw) => {
1959
+ void handler(raw).catch(() => undefined);
1960
+ });
1961
+ }
913
1962
  onOffer(handler) {
914
1963
  this.connection.on("Offer", (raw) => {
915
1964
  void handler(raw).catch(() => undefined);
@@ -947,9 +1996,7 @@ export class SignalRBridgeConnection {
947
1996
  export function defaultBridgeServiceConfig(overrides = {}) {
948
1997
  const defaults = defaultForgeBridgeConfig();
949
1998
  const cliProviders = splitCsv(process.env.FORGE_CLI_PROVIDERS ?? "codex");
950
- const sandboxHostMemoryReserveDefault = process.platform === "linux"
951
- ? 2 * 1024 ** 3
952
- : 1024 ** 3;
1999
+ const configuredSandboxHostMemoryReserveBytes = optionalPositiveFiniteNumber(process.env.FORGE_SANDBOX_HOST_MEMORY_RESERVE_BYTES);
953
2000
  const sandboxPublicationMode = process.env.FORGE_SANDBOX_PUBLICATION_MODE === "bridge_push" ? "bridge_push" : "shared_filesystem";
954
2001
  const config = {
955
2002
  serverUrl: process.env.FORGE_URL ?? "http://localhost:5208",
@@ -964,12 +2011,19 @@ export function defaultBridgeServiceConfig(overrides = {}) {
964
2011
  sandboxPublicationMode,
965
2012
  dockerExecutable: process.env.FORGE_DOCKER_EXECUTABLE ?? "docker",
966
2013
  dockerContext: process.env.FORGE_DOCKER_CONTEXT ?? "",
2014
+ dockerDesktopRecoveryEnabled: process.env.FORGE_DOCKER_DESKTOP_AUTO_RECOVERY === "true",
2015
+ dockerRecoveryFailureThreshold: positiveFiniteNumber(Number(process.env.FORGE_DOCKER_RECOVERY_FAILURE_THRESHOLD ?? 3), 3),
2016
+ dockerRecoveryCooldownMs: positiveFiniteNumber(Number(process.env.FORGE_DOCKER_RECOVERY_COOLDOWN_MS ?? 10 * 60_000), 10 * 60_000),
2017
+ dockerRecoveryStartupTimeoutMs: positiveFiniteNumber(Number(process.env.FORGE_DOCKER_RECOVERY_STARTUP_TIMEOUT_MS ?? 2 * 60_000), 2 * 60_000),
967
2018
  sandboxStateRoot: process.env.FORGE_SANDBOX_STATE_ROOT ?? path.join(os.homedir(), ".forge", "bridge", "sandboxes"),
2019
+ sandboxExecutionRoot: defaultHostExecutionRoot(),
968
2020
  requireRootlessDocker: process.env.FORGE_REQUIRE_ROOTLESS_DOCKER === "true" || process.platform === "linux",
969
2021
  diskHighWatermarkPercent: positiveFiniteNumber(Number(process.env.FORGE_DISK_HIGH_WATERMARK_PERCENT ?? 85), 85),
970
2022
  diskLowWatermarkPercent: positiveFiniteNumber(Number(process.env.FORGE_DISK_LOW_WATERMARK_PERCENT ?? 75), 75),
971
2023
  minimumWorkspaceBytes: positiveFiniteNumber(Number(process.env.FORGE_MINIMUM_WORKSPACE_BYTES ?? 10 * 1024 ** 3), 10 * 1024 ** 3),
972
- sandboxHostMemoryReserveBytes: positiveFiniteNumber(Number(process.env.FORGE_SANDBOX_HOST_MEMORY_RESERVE_BYTES ?? sandboxHostMemoryReserveDefault), sandboxHostMemoryReserveDefault),
2024
+ // Undefined means preflight derives the bounded reserve from the actual
2025
+ // Docker/host capacity domain after sampling it.
2026
+ sandboxHostMemoryReserveBytes: configuredSandboxHostMemoryReserveBytes,
973
2027
  defaultSandboxWorkerMemoryLimitBytes: positiveFiniteNumber(Number(process.env.FORGE_SANDBOX_DEFAULT_WORKER_MEMORY_LIMIT_BYTES ?? 2 * 1024 ** 3), 2 * 1024 ** 3),
974
2028
  defaultSandboxServiceMemoryLimitBytes: nonNegativeFiniteNumber(Number(process.env.FORGE_SANDBOX_DEFAULT_SERVICE_MEMORY_LIMIT_BYTES ?? 0), 0),
975
2029
  sandboxRuntimeOverheadBytes: Math.max(128 * 1024 ** 2, nonNegativeFiniteNumber(Number(process.env.FORGE_SANDBOX_RUNTIME_OVERHEAD_BYTES ?? 128 * 1024 ** 2), 128 * 1024 ** 2)),
@@ -995,10 +2049,12 @@ export function defaultBridgeServiceConfig(overrides = {}) {
995
2049
  supportsDraftRunOnce: true,
996
2050
  supportsDraftAttachments: true,
997
2051
  supportsDraftImageAttachments: supportsDraftImageAttachmentsFor(cliProviders),
2052
+ bridgeChat: bridgeChatCapabilityFor(cliProviders, providerStatusesFor(cliProviders)),
998
2053
  version: process.env.FORGE_BRIDGE_VERSION ?? forgeBridgeVersion,
999
- protocolVersion: 2,
2054
+ protocolVersion: 4,
1000
2055
  minServerVersion: null,
1001
2056
  heartbeatIntervalMs: positiveFiniteNumber(Number(process.env.FORGE_HEARTBEAT_INTERVAL_MS ?? 30_000), 30_000),
2057
+ callbackAckTimeoutMs: positiveFiniteNumber(Number(process.env.FORGE_CALLBACK_ACK_TIMEOUT_MS ?? 30_000), 30_000),
1002
2058
  reconnectBackoffMs: [1000, 2000, 5000, 10000, 30000],
1003
2059
  ...overrides
1004
2060
  };
@@ -1007,7 +2063,9 @@ export function defaultBridgeServiceConfig(overrides = {}) {
1007
2063
  providerStatuses: overrides.providerStatuses ?? providerStatusesFor(config.cliProviders),
1008
2064
  sandboxProfiles: config.sandboxProfiles ?? defaultSandboxProfiles(config.cliProviders),
1009
2065
  supportsDraftImageAttachments: (overrides.supportsDraftImageAttachments ?? true) && supportsDraftImageAttachmentsFor(config.cliProviders),
1010
- heartbeatIntervalMs: positiveFiniteNumber(config.heartbeatIntervalMs, 30_000)
2066
+ bridgeChat: overrides.bridgeChat ?? bridgeChatCapabilityFor(config.cliProviders, overrides.providerStatuses ?? providerStatusesFor(config.cliProviders), config.bridgeChat),
2067
+ heartbeatIntervalMs: positiveFiniteNumber(config.heartbeatIntervalMs, 30_000),
2068
+ callbackAckTimeoutMs: positiveFiniteNumber(config.callbackAckTimeoutMs ?? 30_000, 30_000),
1011
2069
  };
1012
2070
  }
1013
2071
  export function resolveBridgeRunnerHubUrl(serverUrl) {
@@ -1022,6 +2080,12 @@ function normalizeServerUrl(serverUrl) {
1022
2080
  function positiveFiniteNumber(value, fallback) {
1023
2081
  return Number.isFinite(value) && value > 0 ? value : fallback;
1024
2082
  }
2083
+ function optionalPositiveFiniteNumber(value) {
2084
+ if (value === undefined || value.trim().length === 0)
2085
+ return undefined;
2086
+ const parsed = Number(value);
2087
+ return Number.isFinite(parsed) && parsed > 0 ? parsed : undefined;
2088
+ }
1025
2089
  function nonNegativeFiniteNumber(value, fallback) {
1026
2090
  return Number.isFinite(value) && value >= 0 ? value : fallback;
1027
2091
  }
@@ -1126,15 +2190,32 @@ function mappingKeys(mapping) {
1126
2190
  mapping.localPath ? `path:${path.resolve(mapping.localPath).toLowerCase()}` : null
1127
2191
  ].filter((key) => Boolean(key));
1128
2192
  }
1129
- function findRepoMapping(dispatch, mappings) {
1130
- return mappings.find((mapping) => {
1131
- if (!mapping.localPath?.trim()) {
1132
- return false;
2193
+ function findRepoMapping(repo, mappings) {
2194
+ const usable = mappings.filter(mapping => Boolean(mapping.localPath?.trim()));
2195
+ const selectUnique = (candidates, identity) => {
2196
+ if (candidates.length > 1) {
2197
+ throw new Error(`bridge_repo_mapping_ambiguous: multiple local checkouts match ${identity}.`);
1133
2198
  }
1134
- return (equalsIgnoreCase(mapping.repoId, dispatch.repo.id) ||
1135
- equalsIgnoreCase(mapping.name, dispatch.repo.name) ||
1136
- (!!mapping.remoteUrl?.trim() && equalsIgnoreCase(mapping.remoteUrl, dispatch.repo.remoteUrl)));
1137
- });
2199
+ return candidates[0];
2200
+ };
2201
+ // Repository identity is authoritative. A same-name mapping for a different
2202
+ // repository must never win merely because it appears first in configuration.
2203
+ const byId = usable.filter(mapping => equalsIgnoreCase(mapping.repoId, repo.id));
2204
+ if (byId.length > 0)
2205
+ return selectUnique(byId, `repo id '${repo.id}'`);
2206
+ const byRemote = repo.remoteUrl?.trim()
2207
+ ? usable.filter(mapping => !mapping.repoId?.trim()
2208
+ && Boolean(mapping.remoteUrl?.trim())
2209
+ && equalsIgnoreCase(mapping.remoteUrl, repo.remoteUrl))
2210
+ : [];
2211
+ if (byRemote.length > 0)
2212
+ return selectUnique(byRemote, `remote '${repo.remoteUrl}'`);
2213
+ const byName = usable.filter(mapping => !mapping.repoId?.trim()
2214
+ && !mapping.remoteUrl?.trim()
2215
+ && equalsIgnoreCase(mapping.name, repo.name));
2216
+ return byName.length > 0
2217
+ ? selectUnique(byName, `repo name '${repo.name}'`)
2218
+ : undefined;
1138
2219
  }
1139
2220
  function normalizeQaAppLaunchTarget(target) {
1140
2221
  if (!target) {
@@ -1183,6 +2264,10 @@ function invalidDispatchSummary(error) {
1183
2264
  const detail = error instanceof Error ? error.message : String(error);
1184
2265
  return `Invalid bridge dispatch packet: ${detail}`;
1185
2266
  }
2267
+ function invalidChatTurnSummary(error) {
2268
+ const detail = error instanceof Error ? error.message : String(error);
2269
+ return `Invalid Bridge Chat turn packet: ${detail}`;
2270
+ }
1186
2271
  function errorMessage(error) {
1187
2272
  return error instanceof Error ? error.message : String(error);
1188
2273
  }
@@ -1192,29 +2277,328 @@ function formatDelay(ms) {
1192
2277
  }
1193
2278
  return `${ms}ms`;
1194
2279
  }
2280
+ async function waitForCallbackAck(callback, methodName, timeoutMs, signal) {
2281
+ let timeout = null;
2282
+ let rejectAbort = null;
2283
+ const expired = new Promise((_resolve, reject) => {
2284
+ timeout = setTimeout(() => {
2285
+ reject(new Error(`bridge_callback_timeout: ${methodName} callback was not acknowledged within ${formatDelay(timeoutMs)}.`));
2286
+ }, timeoutMs);
2287
+ });
2288
+ const aborted = new Promise((_resolve, reject) => {
2289
+ rejectAbort = reject;
2290
+ });
2291
+ const onAbort = () => rejectAbort?.(new Error(String(signal?.reason ?? `${methodName} callback was cancelled.`)));
2292
+ signal?.addEventListener("abort", onAbort, { once: true });
2293
+ if (signal?.aborted)
2294
+ onAbort();
2295
+ try {
2296
+ return await Promise.race(signal ? [callback, expired, aborted] : [callback, expired]);
2297
+ }
2298
+ finally {
2299
+ if (timeout)
2300
+ clearTimeout(timeout);
2301
+ signal?.removeEventListener("abort", onAbort);
2302
+ }
2303
+ }
1195
2304
  function logBridgeStatus(message) {
1196
2305
  process.stderr.write(`[${new Date().toISOString()}] ${message}\n`);
1197
2306
  }
1198
- function delay(ms, signal) {
2307
+ export function delay(ms, signal) {
1199
2308
  if (signal?.aborted) {
1200
2309
  return Promise.resolve();
1201
2310
  }
1202
2311
  return new Promise((resolve) => {
1203
- const timeout = setTimeout(resolve, ms);
1204
- signal?.addEventListener("abort", () => {
2312
+ const complete = () => {
1205
2313
  clearTimeout(timeout);
2314
+ signal?.removeEventListener("abort", complete);
1206
2315
  resolve();
1207
- }, { once: true });
2316
+ };
2317
+ const timeout = setTimeout(complete, ms);
2318
+ signal?.addEventListener("abort", complete, { once: true });
2319
+ if (signal?.aborted)
2320
+ complete();
1208
2321
  });
1209
2322
  }
1210
- function abortPromise(signal) {
2323
+ export async function waitForPromiseOrAbort(promise, signal) {
1211
2324
  if (!signal) {
1212
- return new Promise(() => undefined);
2325
+ await promise;
2326
+ return;
1213
2327
  }
1214
2328
  if (signal.aborted) {
1215
- return Promise.resolve();
2329
+ return;
2330
+ }
2331
+ let resolveAbort;
2332
+ const abort = new Promise(resolve => { resolveAbort = resolve; });
2333
+ signal.addEventListener("abort", resolveAbort, { once: true });
2334
+ try {
2335
+ await Promise.race([promise, abort]);
2336
+ }
2337
+ finally {
2338
+ signal.removeEventListener("abort", resolveAbort);
2339
+ }
2340
+ }
2341
+ // Matches the API Hello bound on settled recovered host attempts.
2342
+ const maxRecoveredHostAttempts = 128;
2343
+ const defaultChatTurnLeaseMs = 60 * 60 * 1000;
2344
+ const maxChatEventContentChars = 4000;
2345
+ const maxChatTerminalTextChars = 120_000;
2346
+ const maxChatExactSecretChars = 8 * 1024;
2347
+ const chatTurnMetadataFileName = "forge-chat-metadata.json";
2348
+ async function materializeChatTurnMetadata(packet, workspacePath) {
2349
+ if (packet.attachmentHandoffs.length === 0 && packet.actionBrokerGrants.length === 0) {
2350
+ return;
2351
+ }
2352
+ await fs.promises.writeFile(path.join(workspacePath, chatTurnMetadataFileName), `${JSON.stringify({
2353
+ schemaVersion: 1,
2354
+ attachmentHandoffs: packet.attachmentHandoffs.map((handoff) => ({
2355
+ turnId: handoff.turnId,
2356
+ attachmentId: handoff.attachmentId,
2357
+ tempFileToken: handoff.tempFileToken,
2358
+ mimeType: handoff.mimeType,
2359
+ sizeBytes: handoff.sizeBytes,
2360
+ sha256: handoff.sha256,
2361
+ fileNameDisplay: handoff.fileNameDisplay,
2362
+ expiresAt: handoff.expiresAt,
2363
+ status: handoff.status,
2364
+ })),
2365
+ actionBrokerGrants: packet.actionBrokerGrants.map((grant) => ({
2366
+ grantId: grant.grantId,
2367
+ turnId: grant.turnId,
2368
+ leaseId: grant.leaseId,
2369
+ fencingToken: grant.fencingToken,
2370
+ allowedScopes: grant.allowedScopes,
2371
+ expiresAt: grant.expiresAt,
2372
+ redactionStatus: grant.redactionStatus,
2373
+ nonceHeldByBridge: true,
2374
+ })),
2375
+ }, null, 2)}\n`, "utf8");
2376
+ }
2377
+ function chatTurnLeaseExpiresAt(packet, nowMs = Date.now()) {
2378
+ const defaultExpiresAtMs = nowMs + defaultChatTurnLeaseMs;
2379
+ const metadataExpiresAtMs = [
2380
+ ...packet.attachmentHandoffs.map(handoff => Date.parse(handoff.expiresAt)),
2381
+ ...packet.actionBrokerGrants.map(grant => Date.parse(grant.expiresAt)),
2382
+ ].filter(Number.isFinite);
2383
+ return new Date(Math.min(defaultExpiresAtMs, ...metadataExpiresAtMs)).toISOString();
2384
+ }
2385
+ async function cleanupChatTurnWorkspace(workspacePath, redactionSecrets) {
2386
+ try {
2387
+ await fs.promises.rm(workspacePath, { recursive: true, force: true, maxRetries: 3 });
2388
+ return {
2389
+ status: "completed",
2390
+ eventType: "chat_turn_cleanup_completed",
2391
+ role: "runner",
2392
+ kind: "checkpoint",
2393
+ content: "Bridge Chat workspace cleanup completed.",
2394
+ };
2395
+ }
2396
+ catch (error) {
2397
+ return {
2398
+ status: "failed",
2399
+ eventType: "chat_turn_cleanup_failed",
2400
+ role: "runner",
2401
+ kind: "error",
2402
+ content: "Bridge Chat workspace cleanup failed.",
2403
+ message: clip(redactText(errorMessage(error), redactionSecrets), 1000),
2404
+ };
2405
+ }
2406
+ }
2407
+ function buildChatProviderCommand(packet, cwd) {
2408
+ const provider = packet.providerKind.trim().toLowerCase();
2409
+ const model = packet.model.trim();
2410
+ const prompt = buildChatPrompt(packet);
2411
+ const resumeThreadId = packet.providerThreadId?.trim() || null;
2412
+ if (provider === "codex") {
2413
+ const args = ["-s", codexSandboxModeForPlatform(process.platform, "host"), "-a", "never", "exec"];
2414
+ if (resumeThreadId) {
2415
+ args.push("--cd", cwd, "resume");
2416
+ }
2417
+ args.push("--model", model, "--json");
2418
+ if (!resumeThreadId) {
2419
+ args.push("--cd", cwd);
2420
+ }
2421
+ args.push("--skip-git-repo-check");
2422
+ if (resumeThreadId) {
2423
+ args.push(resumeThreadId);
2424
+ }
2425
+ args.push("-");
2426
+ return { command: { command: "codex", args }, stdin: prompt };
2427
+ }
2428
+ if (provider === "cursor") {
2429
+ const args = ["--print", "--output-format", "stream-json", "--model", model, "--workspace", cwd];
2430
+ if (resumeThreadId) {
2431
+ args.push("--resume", resumeThreadId);
2432
+ }
2433
+ args.push("--yolo", "--approve-mcps", "--trust", prompt);
2434
+ return { command: { command: "cursor-agent", args } };
2435
+ }
2436
+ if (provider === "claude") {
2437
+ const args = ["-p", "--output-format", "stream-json", "--verbose", "--model", model, "--permission-mode", "bypassPermissions"];
2438
+ if (resumeThreadId) {
2439
+ args.push("--resume", resumeThreadId);
2440
+ }
2441
+ return { command: { command: "claude", args }, stdin: prompt };
2442
+ }
2443
+ return {
2444
+ command: {
2445
+ command: provider,
2446
+ args: ["--model", model, prompt],
2447
+ },
2448
+ };
2449
+ }
2450
+ function buildChatPrompt(packet) {
2451
+ const messages = packet.messages
2452
+ .map((message, index) => {
2453
+ const role = message.role.trim() || `message-${index + 1}`;
2454
+ return `[${role}]\n${message.content}`;
2455
+ })
2456
+ .join("\n\n");
2457
+ const contextRefs = packet.contextRefs.length === 0
2458
+ ? ""
2459
+ : `\n\n# Context References\n\n${packet.contextRefs.map((ref, index) => {
2460
+ const title = ref.title?.trim() ? `\nTitle: ${ref.title.trim()}` : "";
2461
+ const summary = ref.summary?.trim() ? `\nSummary: ${ref.summary.trim()}` : "";
2462
+ return `${index + 1}. ${ref.kind}:${ref.refId}${title}${summary}`;
2463
+ }).join("\n\n")}`;
2464
+ const continuation = packet.providerThreadId
2465
+ ? "\n\nContinue the existing provider thread identified by the bridge; do not restart the conversation context."
2466
+ : "";
2467
+ const boundedMetadata = buildChatBoundedMetadataPrompt(packet);
2468
+ return `You are handling a Forge Bridge Chat turn.
2469
+
2470
+ Conversation ID: ${packet.conversationId}
2471
+ Turn ID: ${packet.turnId}
2472
+ Provider: ${packet.providerKind}
2473
+ Model: ${packet.model}${continuation}${boundedMetadata}
2474
+
2475
+ # Conversation
2476
+
2477
+ ${messages}${contextRefs}
2478
+ `;
2479
+ }
2480
+ function buildChatBoundedMetadataPrompt(packet) {
2481
+ if (packet.attachmentHandoffs.length === 0 && packet.actionBrokerGrants.length === 0) {
2482
+ return "";
2483
+ }
2484
+ const sections = [
2485
+ `\n\n# Bounded Handoff Metadata\n\nMetadata file: ${chatTurnMetadataFileName}\n\nThe metadata file is turn-scoped and cleaned up by the bridge when the turn ends. It contains opaque handles only; attachment content, raw paths, and grant nonces are not included.`,
2486
+ ];
2487
+ if (packet.attachmentHandoffs.length > 0) {
2488
+ sections.push(`Attachment handoffs:\n${packet.attachmentHandoffs.map((handoff, index) => [
2489
+ `${index + 1}. turnId: ${handoff.turnId}`,
2490
+ ` attachmentId: ${handoff.attachmentId}`,
2491
+ ` tempFileToken: ${handoff.tempFileToken}`,
2492
+ ` mimeType: ${handoff.mimeType}`,
2493
+ ` sizeBytes: ${handoff.sizeBytes}`,
2494
+ ` sha256: ${handoff.sha256}`,
2495
+ ` fileNameDisplay: ${handoff.fileNameDisplay}`,
2496
+ ` expiresAt: ${handoff.expiresAt}`,
2497
+ ` status: ${handoff.status}`,
2498
+ ].join("\n")).join("\n\n")}`);
2499
+ }
2500
+ if (packet.actionBrokerGrants.length > 0) {
2501
+ sections.push(`Action-broker grants:\n${packet.actionBrokerGrants.map((grant, index) => [
2502
+ `${index + 1}. grantId: ${grant.grantId}`,
2503
+ ` turnId: ${grant.turnId}`,
2504
+ ` leaseId: ${grant.leaseId}`,
2505
+ ` fencingToken: ${grant.fencingToken}`,
2506
+ ` allowedScopes: ${grant.allowedScopes.join(", ")}`,
2507
+ ` expiresAt: ${grant.expiresAt}`,
2508
+ ` redactionStatus: ${grant.redactionStatus}`,
2509
+ " nonce: held by bridge",
2510
+ ].join("\n")).join("\n\n")}`);
2511
+ }
2512
+ return sections.join("\n\n");
2513
+ }
2514
+ function chatProviderExitFailure(command, result, redactionSecrets) {
2515
+ const commandText = redactText(formatProviderCliCommandForCheckpoint(command), redactionSecrets);
2516
+ const stderr = clip(redactText(result.stderr.trim(), redactionSecrets), 1000);
2517
+ const stdout = clip(redactText(result.stdout.trim(), redactionSecrets), 1000);
2518
+ const diagnostics = [
2519
+ stderr ? `stderr: ${stderr}` : null,
2520
+ stdout ? `stdout: ${stdout}` : null,
2521
+ ].filter(Boolean).join(" ");
2522
+ return new Error(`provider_failed: ${commandText} exited with code ${result.exitCode}.${diagnostics ? ` ${diagnostics}` : ""}`);
2523
+ }
2524
+ function chatAbortKind(reason) {
2525
+ const message = typeof reason === "string" ? reason.toLowerCase() : "";
2526
+ return message.includes("expired") ? "expired" : "cancelled";
2527
+ }
2528
+ function chatAbortPayload(reason, cleanupOutcome, packet) {
2529
+ const kind = chatAbortKind(reason);
2530
+ const message = typeof reason === "string" && reason.trim()
2531
+ ? reason.trim()
2532
+ : (kind === "expired" ? "Bridge Chat turn expired." : "Bridge Chat turn cancelled.");
2533
+ return {
2534
+ code: kind === "expired" ? "chat_turn_expired" : "chat_turn_cancelled",
2535
+ message: clip(redactText(message, packet ? chatRedactionSecrets(packet) : []), 1000),
2536
+ cleanup: cleanupOutcome,
2537
+ };
2538
+ }
2539
+ function providerThreadIdFromEvent(event) {
2540
+ if (!event.data) {
2541
+ return null;
2542
+ }
2543
+ try {
2544
+ const data = JSON.parse(event.data);
2545
+ for (const key of ["threadId", "thread_id", "sessionId", "session_id", "providerThreadId"]) {
2546
+ const value = data[key];
2547
+ if (typeof value === "string") {
2548
+ const clean = value.trim();
2549
+ if (clean.length > 0 && clean.length <= bridgeChatMaxProviderThreadIdChars) {
2550
+ return clean;
2551
+ }
2552
+ }
2553
+ }
2554
+ }
2555
+ catch {
2556
+ return null;
1216
2557
  }
1217
- return new Promise((resolve) => signal.addEventListener("abort", () => resolve(), { once: true }));
2558
+ return null;
2559
+ }
2560
+ function chatRedactionSecrets(packet) {
2561
+ return [
2562
+ packet.chatRunId,
2563
+ packet.conversationId,
2564
+ packet.turnId,
2565
+ packet.leaseId,
2566
+ packet.fencingToken,
2567
+ packet.runnerInstanceId,
2568
+ packet.providerThreadId,
2569
+ ...packet.attachmentHandoffs.flatMap((handoff) => [
2570
+ handoff.attachmentId,
2571
+ handoff.tempFileToken,
2572
+ ]),
2573
+ ...packet.actionBrokerGrants.flatMap((grant) => [
2574
+ grant.grantId,
2575
+ grant.nonce,
2576
+ ]),
2577
+ ...packet.messages.flatMap((message) => boundedChatSecretFragments(message.content)),
2578
+ ...packet.contextRefs.flatMap((ref) => [
2579
+ ...boundedChatSecretFragments(ref.title ?? ""),
2580
+ ...boundedChatSecretFragments(ref.summary ?? ""),
2581
+ ]),
2582
+ ].filter((value) => typeof value === "string" && value.length > 0);
2583
+ }
2584
+ function boundedChatSecretFragments(value) {
2585
+ const clean = value.trim();
2586
+ if (!clean) {
2587
+ return [];
2588
+ }
2589
+ if (clean.length <= maxChatExactSecretChars) {
2590
+ return [clean];
2591
+ }
2592
+ const half = Math.floor(maxChatExactSecretChars / 2);
2593
+ return [clean.slice(0, half), clean.slice(clean.length - half)];
2594
+ }
2595
+ function safePathComponent(value) {
2596
+ const clean = value.trim().replace(/[^a-zA-Z0-9_.-]+/gu, "-").replace(/^-+|-+$/gu, "");
2597
+ return clean || "chat-turn";
2598
+ }
2599
+ function containsIgnoreCase(values, value) {
2600
+ const normalized = value.trim().toLowerCase();
2601
+ return values.some((candidate) => candidate.trim().toLowerCase() === normalized);
1218
2602
  }
1219
2603
  export function canReapManagedSandboxes(preflight) {
1220
2604
  return canRunDockerMaintenance(preflight);
@@ -1234,24 +2618,15 @@ function sandboxNetworkPolicyFailure(dispatch, config) {
1234
2618
  }
1235
2619
  return null;
1236
2620
  }
1237
- function infrastructureFailureEventType(summary) {
1238
- if (summary.startsWith("worktree_prepare_failed:"))
1239
- return "worktree_prepare_failed";
1240
- if (summary.startsWith("provider_start_unavailable:"))
1241
- return "provider_start_unavailable";
1242
- if (summary.startsWith("host_port_collision:"))
1243
- return "host_port_collision";
1244
- if (summary.startsWith("host_runtime_process_exited:"))
1245
- return "host_runtime_process_exited";
1246
- if (summary.startsWith("host_runtime_listener_timeout:"))
1247
- return "host_runtime_listener_timeout";
1248
- if (summary.startsWith("host_runtime_http_readiness_timeout:"))
1249
- return "host_runtime_http_readiness_timeout";
1250
- if (summary.startsWith("attempt_process_supervisor_start_failed:"))
1251
- return "attempt_process_supervisor_start_failed";
1252
- if (summary.startsWith("attempt_process_supervisor_start_timeout:"))
1253
- return "attempt_process_supervisor_start_timeout";
1254
- return null;
2621
+ export function dockerEngineRequirementFailure(requiredCapabilities, executionMode, capability) {
2622
+ if (!requiredCapabilities.includes("docker_engine"))
2623
+ return null;
2624
+ if (executionMode === "docker") {
2625
+ return "docker_engine_unavailable: docker_engine is not exposed inside Forge agent sandboxes; use a managed host runner with Docker-daemon access.";
2626
+ }
2627
+ return capability?.available
2628
+ ? null
2629
+ : `docker_engine_unavailable: ${capability?.unavailableReason ?? "Docker daemon is unreachable from the bridge process."}`;
1255
2630
  }
1256
2631
  function parseReadinessProbe(value) {
1257
2632
  if (!value || typeof value !== "object" || Array.isArray(value))
@@ -1279,6 +2654,24 @@ function parseReadinessProbe(value) {
1279
2654
  }
1280
2655
  return request;
1281
2656
  }
2657
+ function deferredReadinessResult(request, runnerInstanceId) {
2658
+ return {
2659
+ schemaVersion: 1,
2660
+ requestId: request.requestId,
2661
+ fencingToken: request.fencingToken,
2662
+ runnerInstanceId,
2663
+ status: "failed",
2664
+ reasonCode: "host_runtime_probe_deferred",
2665
+ reasonMessage: "Foreground agent work is active; Forge deferred background runtime readiness so it cannot consume the agent's host capacity.",
2666
+ checkedAt: new Date().toISOString(),
2667
+ };
2668
+ }
2669
+ function readinessDeferralReason(signal) {
2670
+ if (!signal.aborted)
2671
+ return null;
2672
+ const reason = errorMessage(signal.reason);
2673
+ return reason.startsWith("host_runtime_probe_deferred:") ? reason : null;
2674
+ }
1282
2675
  function sandboxProviderNetworkFailure(dispatch, config) {
1283
2676
  const sandbox = dispatch.sandbox;
1284
2677
  if (!sandbox || sandbox.networkPolicy.mode !== "offline")
@@ -1308,6 +2701,15 @@ function sandboxAuthFailure(dispatch, config) {
1308
2701
  const doctor = adapter.doctor();
1309
2702
  return doctor.available ? null : doctor.message;
1310
2703
  }
2704
+ function rejectedOfferReservation(code, message) {
2705
+ return {
2706
+ accepted: false,
2707
+ code,
2708
+ message,
2709
+ reservedUntil: null,
2710
+ resources: null,
2711
+ };
2712
+ }
1311
2713
  export function canReserveSandboxResources(used, requested, capacity) {
1312
2714
  return used.sandboxSlots + requested.sandboxSlots <= capacity.sandboxSlots
1313
2715
  && used.cpuMillis + requested.cpuMillis <= capacity.cpuMillis
@@ -1316,6 +2718,14 @@ export function canReserveSandboxResources(used, requested, capacity) {
1316
2718
  && used.pids + requested.pids <= capacity.pids
1317
2719
  && used.buildSlots + requested.buildSlots <= capacity.buildSlots;
1318
2720
  }
2721
+ export function sandboxMemoryCommitmentCapacity(preflight, activeMemoryBytes) {
2722
+ const ambientCapacity = preflight.availableMemoryBytes
2723
+ - preflight.hostReserveBytes
2724
+ + activeMemoryBytes;
2725
+ const commitmentCapacity = preflight.totalMemoryBytes
2726
+ - preflight.hostReserveBytes;
2727
+ return Math.max(0, Math.min(ambientCapacity, commitmentCapacity));
2728
+ }
1319
2729
  export function sandboxCapacityFailureMessage(used, requested, capacity, preflight, config) {
1320
2730
  const availableSlots = Math.max(0, capacity.sandboxSlots - used.sandboxSlots);
1321
2731
  if (requested.sandboxSlots > availableSlots) {
@@ -1346,14 +2756,6 @@ export function sandboxCapacityFailureMessage(used, requested, capacity, preflig
1346
2756
  const availableBuildSlots = Math.max(0, capacity.buildSlots - used.buildSlots);
1347
2757
  return `Sandbox build capacity is insufficient: requested ${requested.buildSlots}, available ${availableBuildSlots}, ${used.buildSlots} already reserved.`;
1348
2758
  }
1349
- function formatBytes(bytes) {
1350
- if (!Number.isFinite(bytes) || bytes <= 0)
1351
- return "0 B";
1352
- const units = ["B", "KiB", "MiB", "GiB", "TiB"];
1353
- const power = Math.min(units.length - 1, Math.floor(Math.log(bytes) / Math.log(1024)));
1354
- const value = bytes / 1024 ** power;
1355
- return `${value >= 10 || power === 0 ? value.toFixed(0) : value.toFixed(2)} ${units[power]}`;
1356
- }
1357
2759
  function sumSandboxResources(resources) {
1358
2760
  return resources.reduce((total, resource) => ({
1359
2761
  sandboxSlots: total.sandboxSlots + resource.sandboxSlots,
@@ -1425,8 +2827,16 @@ export async function resolveSandboxResourcesFromRepo(dispatch, baseRuntimeOverh
1425
2827
  throw new Error("The approved immutable sandbox contract is unavailable in the mapped repository.");
1426
2828
  }
1427
2829
  const contract = validateSandboxContract(contractResult.stdout);
2830
+ const workspaceBytes = await estimateWorkspaceBytesFromRepo(repoPath, {
2831
+ role: dispatch.execution.role,
2832
+ serviceCount: contract.contract.services.length,
2833
+ browser: contract.contract.browser,
2834
+ });
1428
2835
  return {
1429
- resources: resolveDynamicSandboxResources(sandbox.requestedResources, contract.contract.services.length, sandbox.networkPolicy.mode, baseRuntimeOverheadBytes),
2836
+ resources: resolveDynamicSandboxResources({
2837
+ ...sandbox.requestedResources,
2838
+ workspaceBytes: Math.max(sandbox.requestedResources.workspaceBytes, workspaceBytes),
2839
+ }, contract.contract.services.length, sandbox.networkPolicy.mode, baseRuntimeOverheadBytes),
1430
2840
  sandboxConfigCommitSha: configCommitSha,
1431
2841
  };
1432
2842
  }