@fitlab-ai/agent-infra 0.9.5 → 0.9.7

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 (172) hide show
  1. package/dist/bin/internal-cli.js +8 -3
  2. package/dist/lib/agent-clients/adapters/codex-lifecycle/app-server.js +171 -20
  3. package/dist/lib/agent-clients/adapters/codex-lifecycle/build-identity.js +162 -0
  4. package/dist/lib/agent-clients/adapters/codex-lifecycle/capability-store.js +244 -0
  5. package/dist/lib/agent-clients/adapters/codex-lifecycle/evidence.js +14 -2
  6. package/dist/lib/agent-clients/adapters/codex-lifecycle/sandbox-controller.js +370 -0
  7. package/dist/lib/agent-clients/adapters/codex-lifecycle/store.js +68 -8
  8. package/dist/lib/agent-clients/adapters/codex.js +1 -1
  9. package/dist/lib/defaults.json +10 -2
  10. package/dist/lib/git/workflow.js +30 -4
  11. package/dist/lib/internal/codex-lifecycle.js +169 -3
  12. package/dist/lib/internal/codex-sandbox-controller.js +106 -0
  13. package/dist/lib/internal/release-workflow.js +108 -11
  14. package/dist/lib/internal/sandbox-control.js +43 -6
  15. package/dist/lib/internal/task-create.js +12 -4
  16. package/dist/lib/internal/task-orchestration.js +40 -7
  17. package/dist/lib/platform/context.js +6 -4
  18. package/dist/lib/platform/github-client.js +55 -2
  19. package/dist/lib/platform/issue-comments.js +2 -1
  20. package/dist/lib/process-data/checkpoint.js +319 -0
  21. package/dist/lib/process-data/index.js +128 -60
  22. package/dist/lib/process-data/materialize.js +84 -0
  23. package/dist/lib/process-data/normalize.js +34 -1
  24. package/dist/lib/process-data/reconcile.js +55 -1
  25. package/dist/lib/process-data/sources.js +422 -1
  26. package/dist/lib/process-data/store.js +341 -11
  27. package/dist/lib/release/workflow.js +29 -21
  28. package/dist/lib/sandbox/commands/create.js +111 -60
  29. package/dist/lib/sandbox/commands/prune.js +40 -16
  30. package/dist/lib/sandbox/commands/rm.js +372 -94
  31. package/dist/lib/sandbox/commands/start.js +4 -3
  32. package/dist/lib/sandbox/control/client.js +138 -22
  33. package/dist/lib/sandbox/control/container-identity.js +64 -0
  34. package/dist/lib/sandbox/control/executor.js +142 -0
  35. package/dist/lib/sandbox/control/lifecycle.js +557 -0
  36. package/dist/lib/sandbox/control/protocol.js +69 -26
  37. package/dist/lib/sandbox/control/server.js +395 -99
  38. package/dist/lib/sandbox/control/state.js +191 -0
  39. package/dist/lib/sandbox/control/timing.js +8 -0
  40. package/dist/lib/sandbox/index.js +2 -2
  41. package/dist/lib/sandbox/managed-fs.js +14 -2
  42. package/dist/lib/sandbox/mounts.js +22 -19
  43. package/dist/lib/sandbox/recovery.js +284 -50
  44. package/dist/lib/sandbox/workspace-view.js +90 -10
  45. package/dist/lib/sandbox/worktree-safety.js +196 -0
  46. package/dist/lib/server/daemon.js +1 -1
  47. package/dist/lib/server/process-control.js +17 -1
  48. package/dist/lib/server/process-state.js +57 -20
  49. package/dist/lib/task/artifact-lifecycle.js +87 -15
  50. package/dist/lib/task/artifact-receipts.js +128 -0
  51. package/dist/lib/task/codex-orchestration.js +236 -0
  52. package/dist/lib/task/commands/status.js +3 -0
  53. package/dist/lib/task/commands/validate.js +289 -0
  54. package/dist/lib/task/commit-finalization.js +36 -9
  55. package/dist/lib/task/create.js +10 -2
  56. package/dist/lib/task/delegation-receipts.js +109 -2
  57. package/dist/lib/task/events.js +149 -6
  58. package/dist/lib/task/finalization-artifacts.js +2 -1
  59. package/dist/lib/task/implementation-inputs.js +1 -7
  60. package/dist/lib/task/index.js +9 -0
  61. package/dist/lib/task/ledger.js +13 -1
  62. package/dist/lib/task/orchestration.js +365 -15
  63. package/dist/lib/task/sections.js +5 -1
  64. package/dist/lib/task/task-execution-lock.js +5 -4
  65. package/dist/lib/task/verification-engine.js +78 -14
  66. package/dist/lib/task/verification.js +1 -0
  67. package/dist/lifecycle-build-manifest.json +6 -0
  68. package/lib/agent-clients/adapters/codex-lifecycle/app-server.ts +202 -18
  69. package/lib/agent-clients/adapters/codex-lifecycle/build-identity.ts +227 -0
  70. package/lib/agent-clients/adapters/codex-lifecycle/capability-store.ts +321 -0
  71. package/lib/agent-clients/adapters/codex-lifecycle/evidence.ts +19 -3
  72. package/lib/agent-clients/adapters/codex-lifecycle/manifest-files.json +26 -0
  73. package/lib/agent-clients/adapters/codex-lifecycle/sandbox-controller.ts +453 -0
  74. package/lib/agent-clients/adapters/codex-lifecycle/store.ts +83 -6
  75. package/lib/agent-clients/adapters/codex.ts +1 -1
  76. package/lib/defaults.json +10 -2
  77. package/lib/git/workflow.ts +25 -3
  78. package/lib/internal/codex-lifecycle.ts +175 -4
  79. package/lib/internal/codex-sandbox-controller.ts +109 -0
  80. package/lib/internal/release-workflow.ts +107 -11
  81. package/lib/internal/sandbox-control.ts +38 -6
  82. package/lib/internal/task-create.ts +16 -4
  83. package/lib/internal/task-orchestration.ts +34 -7
  84. package/lib/platform/context.ts +8 -5
  85. package/lib/platform/github-client.ts +61 -3
  86. package/lib/platform/issue-comments.ts +2 -1
  87. package/lib/process-data/checkpoint.ts +335 -0
  88. package/lib/process-data/index.ts +112 -60
  89. package/lib/process-data/materialize.ts +83 -0
  90. package/lib/process-data/normalize.ts +29 -1
  91. package/lib/process-data/reconcile.ts +64 -2
  92. package/lib/process-data/sources.ts +449 -2
  93. package/lib/process-data/store.ts +304 -11
  94. package/lib/process-data/types.ts +124 -2
  95. package/lib/release/workflow.ts +69 -20
  96. package/lib/sandbox/commands/create.ts +70 -25
  97. package/lib/sandbox/commands/prune.ts +53 -15
  98. package/lib/sandbox/commands/rm.ts +423 -108
  99. package/lib/sandbox/commands/start.ts +4 -3
  100. package/lib/sandbox/control/client.ts +152 -28
  101. package/lib/sandbox/control/container-identity.ts +93 -0
  102. package/lib/sandbox/control/executor.ts +167 -0
  103. package/lib/sandbox/control/lifecycle.ts +573 -0
  104. package/lib/sandbox/control/protocol.ts +121 -68
  105. package/lib/sandbox/control/server.ts +398 -109
  106. package/lib/sandbox/control/state.ts +213 -0
  107. package/lib/sandbox/control/timing.ts +12 -0
  108. package/lib/sandbox/index.ts +2 -2
  109. package/lib/sandbox/managed-fs.ts +17 -2
  110. package/lib/sandbox/mounts.ts +16 -8
  111. package/lib/sandbox/recovery.ts +335 -65
  112. package/lib/sandbox/workspace-view.ts +102 -11
  113. package/lib/sandbox/worktree-safety.ts +234 -0
  114. package/lib/server/daemon.ts +1 -1
  115. package/lib/server/process-control.ts +24 -3
  116. package/lib/server/process-state.ts +82 -27
  117. package/lib/task/artifact-lifecycle.ts +78 -14
  118. package/lib/task/artifact-receipts.ts +161 -0
  119. package/lib/task/codex-orchestration.ts +319 -0
  120. package/lib/task/commands/status.ts +2 -0
  121. package/lib/task/commands/validate.ts +282 -0
  122. package/lib/task/commit-finalization.ts +52 -9
  123. package/lib/task/create.ts +11 -2
  124. package/lib/task/delegation-receipts.ts +208 -7
  125. package/lib/task/events.ts +131 -5
  126. package/lib/task/finalization-artifacts.ts +2 -1
  127. package/lib/task/implementation-inputs.ts +1 -5
  128. package/lib/task/index.ts +8 -0
  129. package/lib/task/ledger.ts +16 -1
  130. package/lib/task/orchestration.ts +552 -18
  131. package/lib/task/sections.ts +6 -2
  132. package/lib/task/task-execution-lock.ts +11 -9
  133. package/lib/task/verification-engine.ts +93 -24
  134. package/lib/task/verification.ts +2 -1
  135. package/package.json +1 -1
  136. package/templates/.agents/README.en.md +7 -0
  137. package/templates/.agents/README.zh-CN.md +5 -0
  138. package/templates/.agents/hooks/lifecycle-delegation.js +72 -12
  139. package/templates/.agents/rules/lifecycle-orchestration.en.md +8 -2
  140. package/templates/.agents/rules/lifecycle-orchestration.zh-CN.md +8 -2
  141. package/templates/.agents/rules/no-mid-flow-questions.en.md +1 -0
  142. package/templates/.agents/rules/no-mid-flow-questions.zh-CN.md +1 -0
  143. package/templates/.agents/rules/pr-checks-commands.github.en.md +1 -1
  144. package/templates/.agents/rules/pr-checks-commands.github.zh-CN.md +1 -1
  145. package/templates/.agents/rules/review-handshake.en.md +3 -1
  146. package/templates/.agents/rules/review-handshake.zh-CN.md +3 -1
  147. package/templates/.agents/skills/code-task/reference/dual-mode.en.md +3 -1
  148. package/templates/.agents/skills/code-task/reference/dual-mode.zh-CN.md +3 -1
  149. package/templates/.agents/skills/complete-task/SKILL.en.md +4 -0
  150. package/templates/.agents/skills/complete-task/SKILL.zh-CN.md +4 -0
  151. package/templates/.agents/skills/post-release/SKILL.en.md +32 -6
  152. package/templates/.agents/skills/post-release/SKILL.zh-CN.md +32 -6
  153. package/templates/.agents/skills/run-manual-validation/SKILL.en.md +42 -0
  154. package/templates/.agents/skills/run-manual-validation/SKILL.zh-CN.md +42 -0
  155. package/templates/.agents/skills/run-manual-validation/config/verify.en.json +25 -0
  156. package/templates/.agents/skills/run-manual-validation/config/verify.zh-CN.json +25 -0
  157. package/templates/.agents/skills/run-manual-validation/reference/discovery-and-execution.en.md +67 -0
  158. package/templates/.agents/skills/run-manual-validation/reference/discovery-and-execution.zh-CN.md +67 -0
  159. package/templates/.agents/skills/run-manual-validation/reference/report-template.en.md +69 -0
  160. package/templates/.agents/skills/run-manual-validation/reference/report-template.zh-CN.md +69 -0
  161. package/templates/.agents/skills/run-task/SKILL.en.md +9 -7
  162. package/templates/.agents/skills/run-task/SKILL.zh-CN.md +9 -7
  163. package/templates/.agents/skills/run-task/reference/host-validation.en.md +5 -3
  164. package/templates/.agents/skills/run-task/reference/host-validation.zh-CN.md +5 -3
  165. package/templates/.agents/skills/update-agent-infra/scripts/sync-templates.js +75 -3
  166. package/templates/.claude/commands/run-manual-validation.en.md +9 -0
  167. package/templates/.claude/commands/run-manual-validation.zh-CN.md +9 -0
  168. package/templates/.codex/hooks.json +8 -8
  169. package/templates/.opencode/commands/run-manual-validation.en.md +11 -0
  170. package/templates/.opencode/commands/run-manual-validation.zh-CN.md +11 -0
  171. package/templates/.agents/workspace/README.en.md +0 -26
  172. package/templates/.agents/workspace/README.zh-CN.md +0 -26
@@ -1,32 +1,81 @@
1
- type ReleasePhase = 'unprepared' | 'prepared' | 'partially-published' | 'published' | 'post-pending' | 'complete';
2
- type ReleaseFacts = { localTag: boolean; localTagAncestor?: boolean; localTagConflict?: boolean; remoteBranch: boolean; remoteTag: boolean; githubRelease: boolean | null; npm: boolean | null; homebrew: boolean | null; smoke: 'success' | 'pending' | 'failed' | null; postCommit: boolean };
1
+ import crypto from 'node:crypto';
2
+
3
+ type ReleasePhase = 'unprepared' | 'prepared' | 'partially-published' | 'published' | 'post-pending' | 'post-prepared' | 'complete';
4
+ type PostReleaseFacts = {
5
+ commit: string | null;
6
+ isHead: boolean;
7
+ published: boolean;
8
+ branch: string;
9
+ upstream: string | null;
10
+ remoteHead: string | null;
11
+ newVersion: string | null;
12
+ changedPaths: string[];
13
+ demoInputSha256: string | null;
14
+ worktree: string[];
15
+ staged: string[];
16
+ };
17
+ type ReleaseFacts = {
18
+ localTag: boolean;
19
+ localTagAncestor?: boolean;
20
+ localTagConflict?: boolean;
21
+ remoteBranch: boolean;
22
+ remoteTag: boolean;
23
+ githubRelease: boolean | null;
24
+ npm: boolean | null;
25
+ homebrew: boolean | null;
26
+ smoke: 'success' | 'pending' | 'failed' | null;
27
+ post: PostReleaseFacts;
28
+ };
29
+ type PostConfirmation = {
30
+ version: string;
31
+ commit: string;
32
+ branch: string;
33
+ upstream: string | null;
34
+ remoteHead: string | null;
35
+ newVersion: string | null;
36
+ changedPaths: string[];
37
+ demoInputSha256: string | null;
38
+ worktree: string[];
39
+ staged: string[];
40
+ sha256: string;
41
+ };
42
+
43
+ function byteSort(values: readonly string[]): string[] {
44
+ return [...values].sort((left, right) => Buffer.from(left).compare(Buffer.from(right)));
45
+ }
3
46
 
4
47
  function deriveReleasePhase(facts: ReleaseFacts): ReleasePhase {
5
48
  if (!facts.localTag && !facts.localTagAncestor) return 'unprepared';
49
+ if (facts.post.commit) return facts.post.published ? 'complete' : 'post-prepared';
6
50
  if (!facts.remoteBranch && !facts.remoteTag) return 'prepared';
7
51
  if (!facts.remoteBranch || !facts.remoteTag) return 'partially-published';
8
52
  if (facts.githubRelease !== true || facts.npm !== true || facts.homebrew !== true) return 'published';
9
- if (facts.smoke !== 'success' || !facts.postCommit) return 'post-pending';
10
- return 'complete';
53
+ return 'post-pending';
11
54
  }
12
55
 
13
- function releaseSnapshot(version: string, facts: ReleaseFacts) {
14
- return { version, tag: `v${version.replace(/^v/, '')}`, phase: deriveReleasePhase(facts), facts };
56
+ function createPostConfirmation(version: string, post: PostReleaseFacts): PostConfirmation | null {
57
+ if (!post.commit || !post.isHead || post.worktree.length || post.staged.length) return null;
58
+ const canonical = {
59
+ version,
60
+ commit: post.commit,
61
+ branch: post.branch,
62
+ upstream: post.upstream,
63
+ remoteHead: post.remoteHead,
64
+ newVersion: post.newVersion,
65
+ changedPaths: byteSort(post.changedPaths),
66
+ demoInputSha256: post.demoInputSha256,
67
+ worktree: byteSort(post.worktree),
68
+ staged: byteSort(post.staged)
69
+ };
70
+ const sha256 = `sha256:${crypto.createHash('sha256').update(JSON.stringify(canonical), 'utf8').digest('hex')}`;
71
+ return { ...canonical, sha256 };
15
72
  }
16
73
 
17
- async function runReleaseAction(input: { version: string; action: 'inspect' | 'prepare' | 'publish' | 'post'; inspect: () => Promise<ReleaseFacts>; prepare?: () => Promise<void>; publish?: () => Promise<void>; post?: () => Promise<void> }) {
18
- const before = releaseSnapshot(input.version, await input.inspect());
19
- if (input.action === 'inspect') return { status: 'no-op' as const, changed: false, before, snapshot: before, error: null };
20
- if (before.facts.localTagConflict) return { status: 'failed' as const, changed: false, before, snapshot: before, error: { code: 'GIT_TAG_CONFLICT', message: `Tag ${before.tag} is not reachable from HEAD` } };
21
- const allowed = input.action === 'prepare' ? before.phase === 'unprepared'
22
- : input.action === 'publish' ? before.facts.localTag && ['prepared', 'partially-published'].includes(before.phase)
23
- : ['published', 'post-pending'].includes(before.phase);
24
- if (!allowed) return { status: before.phase === 'complete' ? 'no-op' as const : 'failed' as const, changed: false, before, snapshot: before, error: before.phase === 'complete' ? null : { code: 'RELEASE_PHASE_INVALID', message: `Cannot ${input.action} from ${before.phase}` } };
25
- await input[input.action]?.();
26
- const snapshot = releaseSnapshot(input.version, await input.inspect());
27
- const progressed = snapshot.phase !== before.phase;
28
- return { status: progressed ? 'applied' as const : 'blocked' as const, changed: progressed, before, snapshot, error: progressed ? null : { code: 'RELEASE_PROGRESS_PENDING', message: `${input.action} has not reached the next observable phase` } };
74
+ function releaseSnapshot(version: string, facts: ReleaseFacts) {
75
+ const snapshot = { version, tag: `v${version.replace(/^v/, '')}`, phase: deriveReleasePhase(facts), facts };
76
+ const postConfirmation = createPostConfirmation(version, facts.post);
77
+ return postConfirmation ? { ...snapshot, postConfirmation } : snapshot;
29
78
  }
30
79
 
31
- export { deriveReleasePhase, releaseSnapshot, runReleaseAction };
32
- export type { ReleaseFacts, ReleasePhase };
80
+ export { createPostConfirmation, deriveReleasePhase, releaseSnapshot };
81
+ export type { PostConfirmation, PostReleaseFacts, ReleaseFacts, ReleasePhase };
@@ -2,10 +2,9 @@ import fs from 'node:fs';
2
2
  import os from 'node:os';
3
3
  import path from 'node:path';
4
4
  import { createHash } from 'node:crypto';
5
- import { execFileSync, spawn } from 'node:child_process';
5
+ import { execFileSync } from 'node:child_process';
6
6
  import type { ExecFileSyncOptions, StdioOptions } from 'node:child_process';
7
7
  import { parseArgs } from 'node:util';
8
- import { fileURLToPath } from 'node:url';
9
8
  import * as p from '@clack/prompts';
10
9
  import pc from 'picocolors';
11
10
  import { listAgentClientAdapters } from '../../agent-clients/registry.ts';
@@ -67,14 +66,17 @@ import type { SandboxTool, TmpfsSeedEntry } from '../tools.ts';
67
66
  import {
68
67
  assertFreshSandboxReady,
69
68
  hydrateTmpfsSeedEntries,
70
- prepareTmpfsMounts
69
+ prepareTmpfsMounts,
70
+ startSandboxControlBroker
71
71
  } from '../recovery.ts';
72
72
  import { hostJoin, toEnginePath, volumeArg } from '../engines/wsl2-paths.ts';
73
73
  import { sandboxCoreBindMounts } from '../mounts.ts';
74
74
  import {
75
75
  assertSandboxTaskSource,
76
+ finalizeSandboxControlManifest,
76
77
  materializeSandboxControl,
77
- materializeSandboxWorkspaceView
78
+ materializeSandboxWorkspaceView,
79
+ prepareSandboxWorkspaceMountTarget,
78
80
  } from '../workspace-view.ts';
79
81
  import { clipboardHostDir, CONTAINER_CLIPBOARD_MOUNT } from '../clipboard/paths.ts';
80
82
  import { validateSelinuxDisableEnv } from '../engines/selinux.ts';
@@ -157,12 +159,6 @@ type HostShellConfig = {
157
159
  mounts: Array<{ hostPath: string; containerPath: string }>;
158
160
  };
159
161
 
160
- function internalCliPath(): string {
161
- const directory = path.dirname(fileURLToPath(import.meta.url));
162
- const extension = path.extname(fileURLToPath(import.meta.url));
163
- return path.resolve(directory, '..', '..', '..', 'bin', `internal-cli${extension}`);
164
- }
165
-
166
162
  function resolveToolDirs(config: Pick<SandboxCreateConfig, 'project'>, tools: SandboxTool[], branch: string): ResolvedTool[] {
167
163
  return tools.map((tool) => {
168
164
  const candidates = toolConfigDirCandidates(tool, config.project, branch);
@@ -1254,9 +1250,13 @@ export async function create(args: string[]): Promise<void> {
1254
1250
  const requestedDescription = target.workspace.mode === 'task-bound'
1255
1251
  ? `task-bound:${target.workspace.taskId}`
1256
1252
  : target.workspace.mode;
1253
+ const recoveryTarget = target.workspace.mode === 'task-bound'
1254
+ ? target.workspace.taskId
1255
+ : branch;
1257
1256
  throw new Error(
1258
1257
  `SANDBOX_WORKSPACE_IDENTITY_CONFLICT: container '${name}' is ${existingDescription}, `
1259
- + `but this request is ${requestedDescription}. Run 'ai sandbox rm ${branch}' and retry.`
1258
+ + `but this request is ${requestedDescription}. `
1259
+ + `Run 'ai sandbox start --recreate ${recoveryTarget}' on the host, then retry.`
1260
1260
  );
1261
1261
  }
1262
1262
  }
@@ -1342,6 +1342,7 @@ export async function create(args: string[]): Promise<void> {
1342
1342
  container,
1343
1343
  identity: target.workspace
1344
1344
  });
1345
+ prepareSandboxWorkspaceMountTarget(worktree);
1345
1346
  const control = materializeSandboxControl({
1346
1347
  base: effectiveConfig.controlBase,
1347
1348
  repoRoot: effectiveConfig.repoRoot,
@@ -1349,7 +1350,8 @@ export async function create(args: string[]): Promise<void> {
1349
1350
  project: effectiveConfig.project,
1350
1351
  container,
1351
1352
  branch,
1352
- identity: target.workspace
1353
+ identity: target.workspace,
1354
+ engine
1353
1355
  });
1354
1356
  hostShellConfig = prepareHostShellConfig({
1355
1357
  home: effectiveConfig.home,
@@ -1362,6 +1364,7 @@ export async function create(args: string[]): Promise<void> {
1362
1364
  shellConfigHostDir: hostShellConfig.hostDir,
1363
1365
  workspaceViewRoot: workspaceView.root,
1364
1366
  controlDir: control.channelDir,
1367
+ controlStatusDir: control.statusDir,
1365
1368
  ...(target.workspace.mode === 'task-bound'
1366
1369
  ? {
1367
1370
  taskSource: assertSandboxTaskSource(
@@ -1421,9 +1424,12 @@ export async function create(args: string[]): Promise<void> {
1421
1424
  const hostTz = detectHostTimezone();
1422
1425
  const tzFlags = hostTz ? ['-e', `TZ=${hostTz}`] : [];
1423
1426
 
1424
- runEngineTaskCommand(engine, 'docker', [
1427
+ let createdContainerRef: string | null = null;
1428
+ try {
1429
+ const dockerRunId = runEngineTaskCommand(engine, 'docker', [
1425
1430
  'run',
1426
1431
  '-d',
1432
+ '--init',
1427
1433
  '--name',
1428
1434
  container,
1429
1435
  '--hostname',
@@ -1445,7 +1451,11 @@ export async function create(args: string[]): Promise<void> {
1445
1451
  '-e',
1446
1452
  `AGENT_INFRA_CONTROL_TOKEN=${control.token}`,
1447
1453
  '-e',
1454
+ `AGENT_INFRA_CONTROL_GENERATION=${control.generation}`,
1455
+ '-e',
1448
1456
  'AGENT_INFRA_CONTROL_DIR=/run/agent-infra/control',
1457
+ '-e',
1458
+ 'AGENT_INFRA_CONTROL_STATUS_DIR=/run/agent-infra/control-status',
1449
1459
  '--label',
1450
1460
  `${sandboxRuntimeCapabilityLabel(effectiveConfig)}=${capabilityPlan.runtimeSignature}`,
1451
1461
  ...coreVolumes,
@@ -1466,18 +1476,53 @@ export async function create(args: string[]): Promise<void> {
1466
1476
  '-w',
1467
1477
  '/workspace',
1468
1478
  effectiveConfig.imageName
1469
- ]);
1470
- const broker = spawn(
1471
- process.execPath,
1472
- [internalCliPath(), 'sandbox-control', 'serve', '--manifest', control.manifestPath],
1473
- { cwd: effectiveConfig.repoRoot, detached: true, stdio: 'ignore' }
1474
- );
1475
- broker.once('error', () => {
1476
- // Fresh readiness verification below remains the source of truth;
1477
- // consume detached spawn errors so they cannot crash the command.
1478
- });
1479
- broker.unref();
1480
- createdTmpfsSeedPlan = tmpfsSeedPlan;
1479
+ ]);
1480
+ createdContainerRef = dockerRunId.trim();
1481
+ const containerId = runEngineTaskCommand(engine, 'docker', [
1482
+ 'inspect', '--format', '{{.Id}}', dockerRunId.trim()
1483
+ ]).trim();
1484
+ if (!containerId) throw new Error('SANDBOX_CONTROL_CONTAINER_ID_INVALID');
1485
+ const rawContainerLabels = runEngineTaskCommand(engine, 'docker', [
1486
+ 'inspect', '--format', '{{json .Config.Labels}}', containerId
1487
+ ]).trim();
1488
+ let inspectedLabels: unknown;
1489
+ try {
1490
+ inspectedLabels = JSON.parse(rawContainerLabels);
1491
+ } catch {
1492
+ throw new Error('SANDBOX_CONTROL_CONTAINER_LABELS_INVALID');
1493
+ }
1494
+ if (!inspectedLabels || typeof inspectedLabels !== 'object' || Array.isArray(inspectedLabels)) {
1495
+ throw new Error('SANDBOX_CONTROL_CONTAINER_LABELS_INVALID');
1496
+ }
1497
+ const expectedLabels: Record<string, string> = {
1498
+ [sandboxLabel(effectiveConfig)]: '',
1499
+ [sandboxBranchLabel(effectiveConfig)]: branch,
1500
+ [sandboxWorkspaceModeLabel(effectiveConfig)]: target.workspace.mode,
1501
+ [sandboxRuntimeCapabilityLabel(effectiveConfig)]: capabilityPlan.runtimeSignature,
1502
+ ...(target.workspace.mode === 'task-bound'
1503
+ ? { [sandboxTaskIdLabel(effectiveConfig)]: target.workspace.taskId }
1504
+ : {})
1505
+ };
1506
+ const labels = inspectedLabels as Record<string, unknown>;
1507
+ for (const [key, expected] of Object.entries(expectedLabels)) {
1508
+ if (labels[key] !== expected) throw new Error('SANDBOX_CONTROL_CONTAINER_IDENTITY_MISMATCH');
1509
+ }
1510
+ finalizeSandboxControlManifest(control, {
1511
+ engine,
1512
+ id: containerId,
1513
+ labels: expectedLabels
1514
+ });
1515
+ await startSandboxControlBroker(effectiveConfig.repoRoot, control.manifestPath);
1516
+ createdTmpfsSeedPlan = tmpfsSeedPlan;
1517
+ } catch (error) {
1518
+ if (createdContainerRef) {
1519
+ runSafeEngine(engine, 'docker', ['stop', createdContainerRef]);
1520
+ runSafeEngine(engine, 'docker', ['rm', createdContainerRef]);
1521
+ }
1522
+ removeDirRecursive(control.root);
1523
+ removeDirRecursive(workspaceView.root);
1524
+ throw error;
1525
+ }
1481
1526
  } finally {
1482
1527
  envFile.cleanup();
1483
1528
  }
@@ -6,16 +6,27 @@ import pc from 'picocolors';
6
6
  import { loadConfig } from '../config.ts';
7
7
  import type { SandboxConfig } from '../config.ts';
8
8
  import { safeNameCandidates, sandboxBranchLabel, sandboxLabel } from '../constants.ts';
9
- import { detectEngine } from '../engine.ts';
9
+ import { detectEngine, ENGINES } from '../engine.ts';
10
10
  import { hostJoin } from '../engines/wsl2-paths.ts';
11
11
  import { removeManagedDir, removeWorktreeDir } from '../managed-fs.ts';
12
12
  import { parseLabels } from './ls.ts';
13
13
  import { runEngine, runSafe } from '../shell.ts';
14
14
  import { createSandboxCapabilityPlan } from '../agent-client-reconciler.ts';
15
15
  import type { SandboxTool } from '../tools.ts';
16
+ import { createCleanPermit, formatWorktreeSnapshot, inspectWorktrees } from '../worktree-safety.ts';
17
+ import type { WorktreeRemovalPermit } from '../worktree-safety.ts';
16
18
 
17
19
  const USAGE = `Usage: ai sandbox prune [--dry-run]`;
18
20
 
21
+ type PruneDependencies = {
22
+ config?: SandboxConfig;
23
+ tools?: SandboxTool[];
24
+ engine?: string;
25
+ labelsOutput?: string;
26
+ confirm?: typeof p.confirm;
27
+ isCancel?: typeof p.isCancel;
28
+ };
29
+
19
30
  type OrphanKind = 'shell' | 'worktree' | 'share' | 'tool';
20
31
 
21
32
  export type OrphanGroup = {
@@ -107,12 +118,19 @@ export function collectOrphanGroups(
107
118
  return groups;
108
119
  }
109
120
 
110
- export function removeOrphanGroups(config: SandboxConfig, groups: OrphanGroup[]): boolean {
121
+ export function removeOrphanGroups(
122
+ config: SandboxConfig,
123
+ groups: OrphanGroup[],
124
+ permits: ReadonlyMap<string, WorktreeRemovalPermit>,
125
+ allowRegisteredPathFallback = false
126
+ ): boolean {
111
127
  let removedWorktrees = false;
112
128
  for (const group of groups) {
113
129
  for (const dir of group.dirs) {
114
130
  if (group.kind === 'worktree') {
115
- removeWorktreeDir(config.repoRoot, group.base, dir);
131
+ const permit = permits.get(path.resolve(dir));
132
+ if (!permit) throw new Error(`Missing worktree removal permit: ${dir}`);
133
+ removeWorktreeDir(config.repoRoot, group.base, dir, permit, { allowRegisteredPathFallback });
116
134
  removedWorktrees = true;
117
135
  } else {
118
136
  removeManagedDir(group.base, dir);
@@ -142,7 +160,7 @@ function writeGroups(groups: OrphanGroup[]): void {
142
160
  }
143
161
  }
144
162
 
145
- export async function prune(args: string[]): Promise<void> {
163
+ export async function prune(args: string[], dependencies: PruneDependencies = {}): Promise<void> {
146
164
  const { values } = parseArgs({
147
165
  args,
148
166
  strict: true,
@@ -157,9 +175,9 @@ export async function prune(args: string[]): Promise<void> {
157
175
  return;
158
176
  }
159
177
 
160
- const config = loadConfig();
161
- const tools = [...createSandboxCapabilityPlan(config).cleanupInventory];
162
- const engine = detectEngine(config);
178
+ const config = dependencies.config ?? loadConfig();
179
+ const tools = dependencies.tools ?? [...createSandboxCapabilityPlan(config).cleanupInventory];
180
+ const engine = dependencies.engine ?? detectEngine(config);
163
181
  const psArgs = [
164
182
  'ps',
165
183
  '-a',
@@ -168,14 +186,25 @@ export async function prune(args: string[]): Promise<void> {
168
186
  '--format',
169
187
  '{{.Labels}}'
170
188
  ];
171
- let labelsOutput: string;
172
- try {
173
- labelsOutput = runEngine(engine, 'docker', psArgs);
174
- } catch {
175
- throw new Error('Unable to determine active sandbox branches: docker ps failed');
189
+ let labelsOutput = dependencies.labelsOutput;
190
+ if (labelsOutput === undefined) {
191
+ try {
192
+ labelsOutput = runEngine(engine, 'docker', psArgs);
193
+ } catch {
194
+ throw new Error('Unable to determine active sandbox branches: docker ps failed');
195
+ }
176
196
  }
197
+ const confirm = dependencies.confirm ?? p.confirm;
198
+ const isCancel = dependencies.isCancel ?? p.isCancel;
177
199
  const groups = collectOrphanGroups(config, tools, activeBranchesFromLabels(config, labelsOutput));
178
200
  const count = orphanCount(groups);
201
+ const worktrees = groups.filter((group) => group.kind === 'worktree').flatMap((group) => group.dirs);
202
+ const inspections = inspectWorktrees(worktrees);
203
+ const blockers = inspections.filter((inspection) => inspection.status !== 'clean');
204
+ const permits = new Map<string, WorktreeRemovalPermit>();
205
+ for (const inspection of inspections) {
206
+ if (inspection.status === 'clean') permits.set(inspection.snapshot.worktree, createCleanPermit(inspection.snapshot));
207
+ }
179
208
 
180
209
  p.intro(pc.cyan(`Pruning orphaned sandbox state for ${config.project}`));
181
210
 
@@ -186,23 +215,32 @@ export async function prune(args: string[]): Promise<void> {
186
215
  }
187
216
 
188
217
  writeGroups(groups);
218
+ for (const blocker of blockers) {
219
+ p.log.error(blocker.status === 'failed'
220
+ ? `${JSON.stringify(blocker.worktree)}: ${blocker.message}`
221
+ : formatWorktreeSnapshot(blocker.snapshot));
222
+ }
189
223
 
190
224
  if (values['dry-run']) {
191
225
  p.outro(pc.green('Dry run complete'));
192
226
  return;
193
227
  }
194
228
 
195
- const shouldRemove = await p.confirm({
229
+ if (blockers.length > 0) {
230
+ throw new Error('Refusing to prune because worktree preflight found blocker(s); no orphan state was deleted');
231
+ }
232
+
233
+ const shouldRemove = await confirm({
196
234
  message: `Remove ${count} orphaned sandbox state dirs?`,
197
235
  initialValue: true
198
236
  });
199
237
 
200
- if (p.isCancel(shouldRemove) || !shouldRemove) {
238
+ if (isCancel(shouldRemove) || !shouldRemove) {
201
239
  p.outro('Cancelled');
202
240
  return;
203
241
  }
204
242
 
205
- const removedWorktrees = removeOrphanGroups(config, groups);
243
+ const removedWorktrees = removeOrphanGroups(config, groups, permits, engine === ENGINES.WSL2);
206
244
  if (removedWorktrees) {
207
245
  runSafe('git', ['-C', config.repoRoot, 'worktree', 'prune']);
208
246
  }