@letrquan/book 0.2.0 → 0.3.0

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 (157) hide show
  1. package/CHANGELOG.md +2747 -40
  2. package/README.md +132 -1311
  3. package/dist/additional-roots-H7UNAEHS.js +31 -0
  4. package/dist/{app-MOXXWEOY.js → app-GE4OHJOC.js} +5740 -4045
  5. package/dist/app-GE4OHJOC.js.map +1 -0
  6. package/dist/{chunk-5GDZ22YP.js → chunk-224FJZLH.js} +99 -108
  7. package/dist/chunk-224FJZLH.js.map +1 -0
  8. package/dist/chunk-2AAOF5B5.js +28 -0
  9. package/dist/chunk-2AAOF5B5.js.map +1 -0
  10. package/dist/{chunk-CAGOB2N4.js → chunk-4AS7LKKQ.js} +6 -4
  11. package/dist/chunk-4AS7LKKQ.js.map +1 -0
  12. package/dist/chunk-525CAZPY.js +120 -0
  13. package/dist/chunk-525CAZPY.js.map +1 -0
  14. package/dist/{chunk-URLQAXTR.js → chunk-56SIEMDX.js} +2 -2
  15. package/dist/chunk-56SIEMDX.js.map +1 -0
  16. package/dist/{chunk-XIFFTATU.js → chunk-5YHTKWJN.js} +5420 -4031
  17. package/dist/chunk-5YHTKWJN.js.map +1 -0
  18. package/dist/chunk-62QL5QUP.js +283 -0
  19. package/dist/chunk-62QL5QUP.js.map +1 -0
  20. package/dist/{chunk-KWHQ2XXL.js → chunk-67UZEWWF.js} +3 -3
  21. package/dist/chunk-67UZEWWF.js.map +1 -0
  22. package/dist/chunk-6OQQOMZR.js +785 -0
  23. package/dist/chunk-6OQQOMZR.js.map +1 -0
  24. package/dist/{chunk-GTVB4HEM.js → chunk-AQEQX4FF.js} +344 -315
  25. package/dist/chunk-AQEQX4FF.js.map +1 -0
  26. package/dist/chunk-C7ADUNPR.js +273 -0
  27. package/dist/chunk-C7ADUNPR.js.map +1 -0
  28. package/dist/{chunk-E75IPST5.js → chunk-EFJ2NH6E.js} +12 -58
  29. package/dist/chunk-EFJ2NH6E.js.map +1 -0
  30. package/dist/{chunk-5KLVH3PY.js → chunk-HQI7NYGX.js} +5 -3
  31. package/dist/chunk-HQI7NYGX.js.map +1 -0
  32. package/dist/chunk-IME7R2UI.js +6541 -0
  33. package/dist/chunk-IME7R2UI.js.map +1 -0
  34. package/dist/chunk-J4F2QLM2.js +273 -0
  35. package/dist/chunk-J4F2QLM2.js.map +1 -0
  36. package/dist/{chunk-MAQXCUR4.js → chunk-J62VZYOB.js} +444 -112
  37. package/dist/chunk-J62VZYOB.js.map +1 -0
  38. package/dist/chunk-JTPOI3KU.js +826 -0
  39. package/dist/chunk-JTPOI3KU.js.map +1 -0
  40. package/dist/{chunk-WKBO5R4O.js → chunk-KBQS35OV.js} +2 -2
  41. package/dist/chunk-KBQS35OV.js.map +1 -0
  42. package/dist/chunk-KXAYUIKP.js +90 -0
  43. package/dist/chunk-KXAYUIKP.js.map +1 -0
  44. package/dist/{chunk-VZQXX3WW.js → chunk-QYBD22H5.js} +1 -1
  45. package/dist/chunk-QYBD22H5.js.map +1 -0
  46. package/dist/{chunk-V4BOYR52.js → chunk-RFJ6Q6LE.js} +763 -198
  47. package/dist/chunk-RFJ6Q6LE.js.map +1 -0
  48. package/dist/{chunk-7N4J557C.js → chunk-RT5SATF7.js} +1 -1
  49. package/dist/chunk-RT5SATF7.js.map +1 -0
  50. package/dist/{chunk-C7D4NJ2D.js → chunk-S4JMID4G.js} +68 -26
  51. package/dist/chunk-S4JMID4G.js.map +1 -0
  52. package/dist/{chunk-ANSJPYGI.js → chunk-SOW4WRQP.js} +3 -3
  53. package/dist/chunk-SOW4WRQP.js.map +1 -0
  54. package/dist/{chunk-J7MIZ5CA.js → chunk-TLX3H2CL.js} +518 -79
  55. package/dist/chunk-TLX3H2CL.js.map +1 -0
  56. package/dist/{chunk-DGMQEI6Y.js → chunk-VPNLCMVR.js} +22 -9
  57. package/dist/chunk-VPNLCMVR.js.map +1 -0
  58. package/dist/{chunk-BTZZUCLI.js → chunk-W3XFHEF4.js} +173 -29
  59. package/dist/chunk-W3XFHEF4.js.map +1 -0
  60. package/dist/{chunk-574C6MQR.js → chunk-X4NJT436.js} +130 -60
  61. package/dist/chunk-X4NJT436.js.map +1 -0
  62. package/dist/chunk-X66I2NQO.js +137 -0
  63. package/dist/chunk-X66I2NQO.js.map +1 -0
  64. package/dist/chunk-XHAO3K7I.js +18 -0
  65. package/dist/chunk-XHAO3K7I.js.map +1 -0
  66. package/dist/{chunk-3U5IBM24.js → chunk-Y2VHN7TC.js} +2 -2
  67. package/dist/chunk-Y2VHN7TC.js.map +1 -0
  68. package/dist/chunk-YC6W4EJH.js +947 -0
  69. package/dist/chunk-YC6W4EJH.js.map +1 -0
  70. package/dist/{chunk-24AXE6SP.js → chunk-ZPSZZUWF.js} +1 -1
  71. package/dist/chunk-ZPSZZUWF.js.map +1 -0
  72. package/dist/{command-approvals-6Y57VCWZ.js → command-approvals-VBIU6BLM.js} +6 -5
  73. package/dist/context-CBOBDY5Q.js +38 -0
  74. package/dist/{hook-approvals-CTUFKEJT.js → hook-approvals-ZVT4V7IM.js} +5 -5
  75. package/dist/index.js +996 -155
  76. package/dist/index.js.map +1 -1
  77. package/dist/{interactive-assets-CCAAWVKH.js → interactive-assets-T4I4SEHU.js} +13 -9
  78. package/dist/interactive-assets-T4I4SEHU.js.map +1 -0
  79. package/dist/job-runner.js +159 -111
  80. package/dist/job-runner.js.map +1 -1
  81. package/dist/job-supervisor.d.ts +2 -0
  82. package/dist/job-supervisor.js +112 -0
  83. package/dist/job-supervisor.js.map +1 -0
  84. package/dist/loader-S77OVTHH.js +18 -0
  85. package/dist/{mcp-approvals-WNPH65BK.js → mcp-approvals-HJ2VDYMC.js} +5 -5
  86. package/dist/{mcp-config-YIJIES2C.js → mcp-config-RNB5LEVE.js} +3 -3
  87. package/dist/memory-store-DZUT5HUH.js +75 -0
  88. package/dist/{model-window-store-YFVNXCIP.js → model-window-store-MTSQ5JDJ.js} +8 -4
  89. package/dist/openai-compatible-2OB54YST.js +14 -0
  90. package/dist/{permission-approvals-76PVREFD.js → permission-approvals-YLV6UC5W.js} +5 -5
  91. package/dist/permissions-DHOKF3W2.js +45 -0
  92. package/dist/sandbox-WRYTZH3D.js +41 -0
  93. package/dist/{scrollback-LUKLIBVI.js → scrollback-35ME2GFX.js} +32 -22
  94. package/dist/scrollback-35ME2GFX.js.map +1 -0
  95. package/dist/sdk.d.ts +2219 -2361
  96. package/dist/sdk.js +53 -38
  97. package/dist/sdk.js.map +1 -1
  98. package/dist/settings-loader-JUNPGAAM.js +35 -0
  99. package/dist/settings-loader-JUNPGAAM.js.map +1 -0
  100. package/dist/{shell-selection-V6QAYXPX.js → shell-selection-CDHGJ4GC.js} +2 -2
  101. package/dist/shell-selection-CDHGJ4GC.js.map +1 -0
  102. package/package.json +23 -22
  103. package/dist/app-MOXXWEOY.js.map +0 -1
  104. package/dist/chunk-24AXE6SP.js.map +0 -1
  105. package/dist/chunk-3U5IBM24.js.map +0 -1
  106. package/dist/chunk-574C6MQR.js.map +0 -1
  107. package/dist/chunk-5GDZ22YP.js.map +0 -1
  108. package/dist/chunk-5KLVH3PY.js.map +0 -1
  109. package/dist/chunk-7N4J557C.js.map +0 -1
  110. package/dist/chunk-A7JLBK2Y.js +0 -3506
  111. package/dist/chunk-A7JLBK2Y.js.map +0 -1
  112. package/dist/chunk-ANSJPYGI.js.map +0 -1
  113. package/dist/chunk-BTZZUCLI.js.map +0 -1
  114. package/dist/chunk-C7D4NJ2D.js.map +0 -1
  115. package/dist/chunk-CAGOB2N4.js.map +0 -1
  116. package/dist/chunk-DGMQEI6Y.js.map +0 -1
  117. package/dist/chunk-E75IPST5.js.map +0 -1
  118. package/dist/chunk-EYGD6X2T.js +0 -180
  119. package/dist/chunk-EYGD6X2T.js.map +0 -1
  120. package/dist/chunk-GTVB4HEM.js.map +0 -1
  121. package/dist/chunk-J7MIZ5CA.js.map +0 -1
  122. package/dist/chunk-KWHQ2XXL.js.map +0 -1
  123. package/dist/chunk-L6LK2QIQ.js +0 -657
  124. package/dist/chunk-L6LK2QIQ.js.map +0 -1
  125. package/dist/chunk-MAQXCUR4.js.map +0 -1
  126. package/dist/chunk-S4XL7HOM.js +0 -402
  127. package/dist/chunk-S4XL7HOM.js.map +0 -1
  128. package/dist/chunk-URLQAXTR.js.map +0 -1
  129. package/dist/chunk-V4BOYR52.js.map +0 -1
  130. package/dist/chunk-VZQXX3WW.js.map +0 -1
  131. package/dist/chunk-WKBO5R4O.js.map +0 -1
  132. package/dist/chunk-XIFFTATU.js.map +0 -1
  133. package/dist/context-ZC5IFEAX.js +0 -25
  134. package/dist/interactive-assets-CCAAWVKH.js.map +0 -1
  135. package/dist/loader-4IDS2RIE.js +0 -15
  136. package/dist/openai-compatible-RHZUBHV6.js +0 -10
  137. package/dist/permissions-CHCKTFQQ.js +0 -28
  138. package/dist/sandbox-CNVAQDA7.js +0 -23
  139. package/dist/scrollback-LUKLIBVI.js.map +0 -1
  140. package/dist/settings-loader-M2QP5TX2.js +0 -21
  141. package/patches/ink+6.8.0.patch +0 -13
  142. package/scripts/apply-ink-patch.mjs +0 -29
  143. package/scripts/ink-patch.mjs +0 -59
  144. package/scripts/verify-ink-patch.mjs +0 -11
  145. /package/dist/{command-approvals-6Y57VCWZ.js.map → additional-roots-H7UNAEHS.js.map} +0 -0
  146. /package/dist/{context-ZC5IFEAX.js.map → command-approvals-VBIU6BLM.js.map} +0 -0
  147. /package/dist/{hook-approvals-CTUFKEJT.js.map → context-CBOBDY5Q.js.map} +0 -0
  148. /package/dist/{loader-4IDS2RIE.js.map → hook-approvals-ZVT4V7IM.js.map} +0 -0
  149. /package/dist/{mcp-approvals-WNPH65BK.js.map → loader-S77OVTHH.js.map} +0 -0
  150. /package/dist/{mcp-config-YIJIES2C.js.map → mcp-approvals-HJ2VDYMC.js.map} +0 -0
  151. /package/dist/{model-window-store-YFVNXCIP.js.map → mcp-config-RNB5LEVE.js.map} +0 -0
  152. /package/dist/{openai-compatible-RHZUBHV6.js.map → memory-store-DZUT5HUH.js.map} +0 -0
  153. /package/dist/{permission-approvals-76PVREFD.js.map → model-window-store-MTSQ5JDJ.js.map} +0 -0
  154. /package/dist/{permissions-CHCKTFQQ.js.map → openai-compatible-2OB54YST.js.map} +0 -0
  155. /package/dist/{sandbox-CNVAQDA7.js.map → permission-approvals-YLV6UC5W.js.map} +0 -0
  156. /package/dist/{settings-loader-M2QP5TX2.js.map → permissions-DHOKF3W2.js.map} +0 -0
  157. /package/dist/{shell-selection-V6QAYXPX.js.map → sandbox-WRYTZH3D.js.map} +0 -0
package/dist/sdk.js CHANGED
@@ -3,13 +3,18 @@ import {
3
3
  connectMcpServers,
4
4
  disconnectMcpServers,
5
5
  runHeadless
6
- } from "./chunk-J7MIZ5CA.js";
6
+ } from "./chunk-TLX3H2CL.js";
7
+ import {
8
+ mcpServersToRecord,
9
+ partitionMcpServersByApproval
10
+ } from "./chunk-56SIEMDX.js";
7
11
  import {
8
12
  SessionStore,
9
13
  createAgentSessionSnapshot,
10
14
  reduceAgentSessionSnapshot,
11
- resolveSessionBootstrap
12
- } from "./chunk-V4BOYR52.js";
15
+ resolveSessionBootstrap,
16
+ resolveSessionDriver
17
+ } from "./chunk-RFJ6Q6LE.js";
13
18
  import {
14
19
  AgentManager,
15
20
  SessionRuntime,
@@ -17,14 +22,18 @@ import {
17
22
  SkillRegistryError,
18
23
  createDefaultRegistry,
19
24
  createRegistry,
20
- createToolSurface,
21
- freezeAgentConfig,
22
- loadConfig
23
- } from "./chunk-XIFFTATU.js";
25
+ createToolSurface
26
+ } from "./chunk-5YHTKWJN.js";
27
+ import {
28
+ resolveMcpServerList
29
+ } from "./chunk-Y2VHN7TC.js";
24
30
  import {
25
31
  buildSkillReport,
32
+ freezeAgentConfig,
33
+ loadConfig,
26
34
  resolvePermissionMode
27
- } from "./chunk-A7JLBK2Y.js";
35
+ } from "./chunk-IME7R2UI.js";
36
+ import "./chunk-AQEQX4FF.js";
28
37
  import {
29
38
  applySkillOverrides,
30
39
  buildSkillListing,
@@ -33,40 +42,40 @@ import {
33
42
  generateSkillListing,
34
43
  loadSkillBody,
35
44
  skillRoots
36
- } from "./chunk-KWHQ2XXL.js";
37
- import "./chunk-VZQXX3WW.js";
38
- import "./chunk-E75IPST5.js";
39
- import "./chunk-5KLVH3PY.js";
40
- import {
41
- resolveMcpServerList
42
- } from "./chunk-3U5IBM24.js";
43
- import {
44
- mcpServersToRecord,
45
- partitionMcpServersByApproval
46
- } from "./chunk-URLQAXTR.js";
47
- import "./chunk-MAQXCUR4.js";
48
- import {
49
- toolFailure,
50
- toolSuccess
51
- } from "./chunk-5GDZ22YP.js";
52
- import "./chunk-EYGD6X2T.js";
45
+ } from "./chunk-67UZEWWF.js";
46
+ import "./chunk-W3XFHEF4.js";
47
+ import "./chunk-6OQQOMZR.js";
48
+ import "./chunk-J62VZYOB.js";
49
+ import "./chunk-JTPOI3KU.js";
50
+ import "./chunk-4AS7LKKQ.js";
51
+ import "./chunk-QYBD22H5.js";
52
+ import "./chunk-EFJ2NH6E.js";
53
+ import "./chunk-KXAYUIKP.js";
54
+ import "./chunk-HQI7NYGX.js";
55
+ import "./chunk-2AAOF5B5.js";
56
+ import "./chunk-YC6W4EJH.js";
57
+ import "./chunk-C7ADUNPR.js";
58
+ import "./chunk-RT5SATF7.js";
59
+ import "./chunk-XHAO3K7I.js";
60
+ import "./chunk-525CAZPY.js";
61
+ import "./chunk-X66I2NQO.js";
53
62
  import {
54
63
  DEFAULT_LOCAL_DATA_RETENTION_DAYS,
55
64
  cleanupDebugLogs,
56
65
  getDebugLogPath
57
- } from "./chunk-GTVB4HEM.js";
58
- import "./chunk-S4XL7HOM.js";
59
- import "./chunk-7N4J557C.js";
60
- import "./chunk-BTZZUCLI.js";
61
- import "./chunk-WKBO5R4O.js";
62
- import "./chunk-ANSJPYGI.js";
63
- import "./chunk-DGMQEI6Y.js";
64
- import "./chunk-5RDRAO4B.js";
65
- import "./chunk-CAGOB2N4.js";
66
- import "./chunk-574C6MQR.js";
66
+ } from "./chunk-62QL5QUP.js";
67
+ import {
68
+ toolFailure,
69
+ toolSuccess
70
+ } from "./chunk-224FJZLH.js";
71
+ import "./chunk-KBQS35OV.js";
72
+ import "./chunk-SOW4WRQP.js";
73
+ import "./chunk-VPNLCMVR.js";
74
+ import "./chunk-X4NJT436.js";
67
75
  import {
68
76
  resolveBookHome
69
- } from "./chunk-24AXE6SP.js";
77
+ } from "./chunk-ZPSZZUWF.js";
78
+ import "./chunk-5RDRAO4B.js";
70
79
 
71
80
  // src/sdk.ts
72
81
  import { join } from "path";
@@ -514,6 +523,7 @@ async function* query(prompt, options = {}) {
514
523
  });
515
524
  const config = freezeAgentConfig({
516
525
  ...loadedConfig,
526
+ sessionDriver: resolveSessionDriver({ explicit: options.sessionDriver, env: process.env }).driver,
517
527
  ...options.model ? { model: options.model } : {},
518
528
  ...options.maxTurns ? { maxTurns: options.maxTurns } : {},
519
529
  ...options.agents ? {
@@ -531,7 +541,8 @@ async function* query(prompt, options = {}) {
531
541
  const sessionStore = persistSession ? options.sessionStore ?? new SessionStore(join(resolveBookHome(), "sessions")) : void 0;
532
542
  const bootstrap = resolveSessionBootstrap(sessionStore, {
533
543
  cwd: config.workspace,
534
- sessionId: options.sessionId
544
+ sessionId: options.sessionId,
545
+ driver: config.sessionDriver
535
546
  });
536
547
  if (persistSession && !options.sessionStore && sessionStore) {
537
548
  cleanupDebugLogs(DEFAULT_LOCAL_DATA_RETENTION_DAYS, getDebugLogPath());
@@ -569,6 +580,9 @@ async function* query(prompt, options = {}) {
569
580
  prompt,
570
581
  inputFormat: "text",
571
582
  outputFormat: "text",
583
+ // Progress lines are for a person watching `book -p` in a terminal. An
584
+ // SDK host owns its stderr and receives every tool call as an event.
585
+ quiet: true,
572
586
  history: bootstrap.history,
573
587
  transcript: bootstrap.transcript,
574
588
  compactBoundaries: bootstrap.compactBoundaries,
@@ -598,7 +612,8 @@ async function* query(prompt, options = {}) {
598
612
  runs: result.runs,
599
613
  // `stdout` above is a discard sink, so this is the only way an SDK
600
614
  // caller sees a slash command the host performed on its behalf.
601
- commandResults: result.commandResults
615
+ commandResults: result.commandResults,
616
+ answer: result.answer
602
617
  });
603
618
  } catch (error) {
604
619
  if (!controller.signal.aborted) {
package/dist/sdk.js.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/sdk.ts","../src/skill-evaluation.ts","../src/agents/evaluation.ts"],"sourcesContent":["/** Book Agent SDK — programmatic API for embedding Book as a library. */\r\n\r\nimport { join } from 'path';\r\nimport type { ElicitationHandler, UserQuestionHandler } from './types/tools.js';\r\nimport type { AgentConfig } from './types/runtime.js';\r\nimport type { SessionStoreInterface } from './types/sessions.js';\r\nimport { freezeAgentConfig, loadConfig, type LoadConfigOptions } from './config.js';\r\nimport { runHeadless } from './headless.js';\r\nimport { createDefaultRegistry } from './tools/registry.js';\r\nimport { connectMcpServers, disconnectMcpServers } from './mcp.js';\r\nimport { mcpServersToRecord, partitionMcpServersByApproval } from './mcp-approvals.js';\r\nimport { resolveMcpServerList } from './mcp-config.js';\r\nimport { collectWithheldProjectNotices } from './project-approval-notices.js';\r\nimport { AgentManager } from './agents/manager.js';\r\nimport { SessionStore } from './session/store.js';\r\nimport { resolveSessionBootstrap } from './session/resolve.js';\r\nimport type { AgentEvent } from './session/agent-events.js';\r\nimport {\r\n cleanupDebugLogs,\r\n DEFAULT_LOCAL_DATA_RETENTION_DAYS,\r\n getDebugLogPath,\r\n} from './debug-log.js';\r\nimport { resolvePermissionMode } from './permission-mode.js';\r\nimport { resolveBookHome } from './book-home.js';\r\n\r\nexport type QueryEvent = AgentEvent;\r\n\r\nexport interface QueryOptions {\r\n workspace?: string;\r\n model?: string;\r\n permissionMode?: string;\r\n maxTurns?: number;\r\n settingsPath?: string;\r\n noSettings?: boolean;\r\n persistSession?: boolean;\r\n sessionId?: string;\r\n sessionStore?: SessionStoreInterface;\r\n signal?: AbortSignal;\r\n onUserQuestionRequired?: UserQuestionHandler;\r\n /**\r\n * Answer MCP form elicitations. Supplying it declares the elicitation\r\n * capability to connected servers; without it they fail such requests\r\n * themselves rather than blocking on a prompt no one can see.\r\n */\r\n onElicit?: ElicitationHandler;\r\n agents?: 'adaptive' | 'manual' | 'off';\r\n /** Forward high-volume managed-agent text deltas. Defaults to false. */\r\n forwardSubagentText?: boolean;\r\n}\r\n\r\nclass AsyncEventQueue<T> {\r\n private values: T[] = [];\r\n private waiters: Array<(value: T | undefined) => void> = [];\r\n private closed = false;\r\n\r\n push(value: T): void {\r\n if (this.closed) return;\r\n const waiter = this.waiters.shift();\r\n if (waiter) waiter(value);\r\n else this.values.push(value);\r\n }\r\n\r\n close(): void {\r\n if (this.closed) return;\r\n this.closed = true;\r\n for (const waiter of this.waiters.splice(0)) waiter(undefined);\r\n }\r\n\r\n next(): Promise<T | undefined> {\r\n const value = this.values.shift();\r\n if (value !== undefined) return Promise.resolve(value);\r\n if (this.closed) return Promise.resolve(undefined);\r\n return new Promise((resolve) => this.waiters.push(resolve));\r\n }\r\n}\r\n\r\nexport async function* query(\r\n prompt: string,\r\n options: QueryOptions = {},\r\n): AsyncGenerator<QueryEvent, void, undefined> {\r\n const workspace = options.workspace || process.cwd();\r\n const loadedConfig = loadConfig(workspace, {\r\n settingsOverridePath: options.settingsPath,\r\n noSettings: options.noSettings,\r\n runMigrations: !options.noSettings,\r\n });\r\n const config = freezeAgentConfig({\r\n ...loadedConfig,\r\n ...(options.model ? { model: options.model } : {}),\r\n ...(options.maxTurns ? { maxTurns: options.maxTurns } : {}),\r\n ...(options.agents\r\n ? {\r\n settings: {\r\n ...loadedConfig.settings,\r\n agents: { ...loadedConfig.settings.agents, mode: options.agents },\r\n },\r\n }\r\n : {}),\r\n });\r\n\r\n const controller = new AbortController();\r\n const abortFromCaller = () => controller.abort(options.signal?.reason);\r\n if (options.signal?.aborted) abortFromCaller();\r\n else options.signal?.addEventListener('abort', abortFromCaller, { once: true });\r\n\r\n const persistSession = options.persistSession !== false;\r\n const sessionStore = persistSession\r\n ? (options.sessionStore ?? new SessionStore(join(resolveBookHome(), 'sessions')))\r\n : undefined;\r\n const bootstrap = resolveSessionBootstrap(sessionStore, {\r\n cwd: config.workspace,\r\n sessionId: options.sessionId,\r\n });\r\n if (persistSession && !options.sessionStore && sessionStore) {\r\n cleanupDebugLogs(DEFAULT_LOCAL_DATA_RETENTION_DAYS, getDebugLogPath());\r\n sessionStore.cleanup(DEFAULT_LOCAL_DATA_RETENTION_DAYS, new Set([bootstrap.sessionId]));\r\n }\r\n const queue = new AsyncEventQueue<QueryEvent>();\r\n let connections: Awaited<ReturnType<typeof connectMcpServers>>['connections'] = [];\r\n\r\n queue.push({ type: 'system', model: config.model, cwd: config.workspace });\r\n queue.push({ type: 'session', sessionId: bootstrap.sessionId });\r\n\r\n const running = (async () => {\r\n try {\r\n // Project-declared MCP servers require a prior interactive approval;\r\n // SDK runs cannot prompt, so unapproved servers are skipped with a warning.\r\n // Repository-declared allow rules and hooks are withheld until the user\r\n // decides on them; an SDK run cannot ask, so it reports and continues.\r\n for (const notice of collectWithheldProjectNotices({\r\n workspace: config.workspace,\r\n settings: config.settings,\r\n settingsEnabled: !options.noSettings,\r\n })) {\r\n console.warn(notice);\r\n }\r\n const declaredMcpServers = resolveMcpServerList(config.workspace);\r\n const mcpPartition = partitionMcpServersByApproval(declaredMcpServers, config.settings);\r\n for (const server of mcpPartition.pending) {\r\n console.warn(\r\n `⚠ Skipping MCP server \"${server.name}\" (${server.path}): project-declared servers require one-time approval. Approve it in an interactive session first.`,\r\n );\r\n }\r\n const mcp = await connectMcpServers(config.workspace, {\r\n signal: controller.signal,\r\n servers: mcpServersToRecord(mcpPartition.allowed),\r\n onElicit: options.onElicit,\r\n });\r\n connections = mcp.connections;\r\n const registry = createDefaultRegistry({ agents: config.settings.agents.mode !== 'off' });\r\n if (mcp.tools.length > 0) registry.registerAll(mcp.tools);\r\n\r\n const result = await runHeadless(config, registry, {\r\n prompt,\r\n inputFormat: 'text',\r\n outputFormat: 'text',\r\n history: bootstrap.history,\r\n transcript: bootstrap.transcript,\r\n compactBoundaries: bootstrap.compactBoundaries,\r\n mode: resolvePermissionMode(config.settings, options.permissionMode),\r\n maxTurns: options.maxTurns,\r\n persistSession,\r\n sessionStore,\r\n sessionId: bootstrap.sessionId,\r\n sessionCreated: bootstrap.created,\r\n runSource: 'sdk',\r\n signal: controller.signal,\r\n onUserQuestionRequired: options.onUserQuestionRequired,\r\n forwardSubagentText: options.forwardSubagentText,\r\n stdout: { write: () => true },\r\n onAgentEvent: (event) => {\r\n if (\r\n event.type !== 'system' &&\r\n event.type !== 'session' &&\r\n event.type !== 'result' &&\r\n event.type !== 'done' &&\r\n (event.type !== 'agent_text_delta' || options.forwardSubagentText === true)\r\n ) {\r\n queue.push(event);\r\n }\r\n },\r\n });\r\n queue.push({\r\n type: 'result',\r\n messages: result.messages,\r\n usage: result.usage,\r\n sessionId: bootstrap.sessionId,\r\n outcome: result.outcome,\r\n runs: result.runs,\r\n // `stdout` above is a discard sink, so this is the only way an SDK\r\n // caller sees a slash command the host performed on its behalf.\r\n commandResults: result.commandResults,\r\n });\r\n } catch (error) {\r\n if (!controller.signal.aborted) {\r\n queue.push({\r\n type: 'error',\r\n error: error instanceof Error ? error.message : String(error),\r\n });\r\n }\r\n } finally {\r\n await disconnectMcpServers(connections);\r\n queue.push({ type: 'done' });\r\n queue.close();\r\n }\r\n })();\r\n\r\n try {\r\n while (true) {\r\n const event = await queue.next();\r\n if (!event) break;\r\n yield event;\r\n }\r\n } finally {\r\n controller.abort(new Error('SDK event consumer stopped'));\r\n options.signal?.removeEventListener('abort', abortFromCaller);\r\n await running;\r\n }\r\n}\r\n\r\nexport { loadConfig, type LoadConfigOptions };\r\nexport type { ToolDiscoverySettings } from './settings.js';\r\nexport { AgentManager };\r\nexport { createDefaultRegistry, createRegistry } from './tools/registry.js';\r\nexport { createToolSurface } from './tools/catalog.js';\r\nexport { toolFailure, toolSuccess } from './tools/result.js';\r\nexport {\r\n applySkillOverrides,\r\n buildSkillListing,\r\n discoverSkills,\r\n explicitSkillMentions,\r\n generateSkillListing,\r\n loadSkillBody,\r\n skillRoots,\r\n} from './skills.js';\r\nexport type {\r\n DiscoverSkillsOptions,\r\n LoadedSkillBody,\r\n ShadowedSkill,\r\n Skill,\r\n SkillIssue,\r\n SkillListingResult,\r\n SkillResource,\r\n SkillRoot,\r\n SkillRootKind,\r\n SkillSource,\r\n} from './skills.js';\r\nexport {\r\n SkillRegistry,\r\n SkillRegistryError,\r\n type SkillActivationFrame,\r\n type SkillActivationFrameSummary,\r\n type SkillActivationReason,\r\n type SkillLifecycleEvent,\r\n type SkillLifecycleEventType,\r\n type SkillRegistrySnapshot,\r\n} from './skill-registry.js';\r\nexport {\r\n DEFAULT_SKILL_EVALUATION_THRESHOLDS,\r\n DEFAULT_SKILL_EVALUATION_FIXTURES,\r\n SKILL_EVALUATION_CATEGORIES,\r\n evaluateSkillActivation,\r\n observeSkillEvaluation,\r\n renderSkillEvaluationReport,\r\n runSkillActivationEvaluation,\r\n writeSkillEvaluationReport,\r\n} from './skill-evaluation.js';\r\nexport type {\r\n SkillEvaluationCategory,\r\n SkillEvaluationFixture,\r\n SkillEvaluationObservation,\r\n SkillEvaluationOutcome,\r\n SkillEvaluationPromptFixture,\r\n SkillEvaluationReport,\r\n SkillEvaluationThresholds,\r\n} from './skill-evaluation.js';\r\nexport { SessionRuntime } from './session/runtime.js';\r\nexport { buildSkillReport } from './skill-report.js';\r\nexport function createAgentManager(config: AgentConfig): AgentManager {\r\n return new AgentManager(config, createDefaultRegistry({ agents: true }).getDefinitions());\r\n}\r\nexport { runPairedEvaluation, evaluateSuccess } from './agents/evaluation.js';\r\nexport { createAgentSessionSnapshot, reduceAgentSessionSnapshot } from './session/agent-events.js';\r\nexport type {\r\n AgentEvent,\r\n AgentSessionSnapshot,\r\n AgentSessionStatus,\r\n} from './session/agent-events.js';\r\nexport type {\r\n EvaluationFixture,\r\n EvaluationMetric,\r\n EvaluationRunResult,\r\n} from './agents/evaluation.js';\r\nexport type {\r\n AgentPlanRecord,\r\n AgentApplyResult,\r\n AgentRecord,\r\n AgentSummary,\r\n AgentCompletion,\r\n AgentCompletionNotification,\r\n AgentProfile,\r\n AgentActivity,\r\n AgentRuntimeEvent,\r\n AgentRunMetrics,\r\n AgentSnapshot,\r\n AgentSpawnRequest,\r\n EvidenceItem,\r\n} from './agents/types.js';\r\nexport type {\r\n AgentTerminalOutcome,\r\n AgentTerminalReason,\r\n AgentTerminalStatus,\r\n} from './types/terminal.js';\r\nexport type {\r\n AgentModelIdentity,\r\n AgentModelIdentityStatus,\r\n AgentRunAmbientSnapshot,\r\n AgentRunAccounting,\r\n AgentRunContext,\r\n AgentRunResult,\r\n AgentRunSource,\r\n} from './types/runs.js';\r\nexport type {\r\n JsonSchemaObject,\r\n ToolCatalogMetadata,\r\n ToolCategory,\r\n ToolDefinition,\r\n ToolDiscoveryContext,\r\n ToolEffect,\r\n ToolPolicy,\r\n ToolResult,\r\n ToolResultArtifacts,\r\n ToolResultError,\r\n ToolResultPresentation,\r\n ToolResultStatus,\r\n ToolSearchMatch,\r\n UserQuestion,\r\n UserQuestionOption,\r\n UserQuestionRequest,\r\n UserQuestionResponse,\r\n UserQuestionSource,\r\n UserQuestionHandler,\r\n ElicitationField,\r\n ElicitationRequest,\r\n ElicitationResponse,\r\n ElicitationValue,\r\n ElicitationHandler,\r\n} from './types/tools.js';\r\n","import { createHash } from 'node:crypto';\r\nimport { mkdirSync, writeFileSync } from 'node:fs';\r\nimport { dirname } from 'node:path';\r\nimport type { SkillLifecycleEvent } from './skill-registry.js';\r\n\r\nexport const SKILL_EVALUATION_CATEGORIES = [\r\n 'direct',\r\n 'indirect',\r\n 'negative',\r\n 'ambiguous',\r\n 'conflicting',\r\n 'disabled',\r\n 'invalid',\r\n 'missing-body',\r\n 'missing-resource',\r\n] as const;\r\n\r\nexport type SkillEvaluationCategory = (typeof SKILL_EVALUATION_CATEGORIES)[number];\r\n\r\nexport interface SkillEvaluationFixture {\r\n id: string;\r\n category: SkillEvaluationCategory;\r\n promptHash: string;\r\n expectedSkills: string[];\r\n expectedBlockedCodes?: string[];\r\n}\r\n\r\nexport interface SkillEvaluationPromptFixture {\r\n id: string;\r\n category: SkillEvaluationCategory;\r\n prompt: string;\r\n expectedSkills: string[];\r\n expectedBlockedCodes?: string[];\r\n}\r\n\r\nexport interface SkillEvaluationOutcome {\r\n promptChars?: number;\r\n promptTokens?: number;\r\n bodyTokens?: number;\r\n taskCompleted?: boolean;\r\n userCorrections?: number;\r\n skillToolFailures?: number;\r\n exposure?: 'eager' | 'deferred';\r\n activationTurns?: number;\r\n}\r\n\r\nexport interface SkillEvaluationObservation extends SkillEvaluationFixture {\r\n activatedSkills: string[];\r\n blockedCodes: string[];\r\n promptChars: number;\r\n promptTokens: number;\r\n bodyBytes: number;\r\n bodyTokens: number;\r\n falseActivationBytes: number;\r\n activationLatencyMs: number[];\r\n consentRequests: number;\r\n unnecessaryPermissionPrompts: number;\r\n taskCompleted?: boolean;\r\n userCorrections: number;\r\n skillToolFailures: number;\r\n blockingMismatch: boolean;\r\n exposure?: 'eager' | 'deferred';\r\n activationTurns?: number;\r\n}\r\n\r\nexport interface SkillEvaluationThresholds {\r\n minimumPrecision: number;\r\n minimumRecall: number;\r\n maximumFalseActivations: number;\r\n maximumUnnecessaryPermissionPrompts: number;\r\n maximumSkillToolFailures: number;\r\n}\r\n\r\nexport interface SkillEvaluationExecution {\r\n evidenceKind: 'offline-observation';\r\n providerRunEligibility: 'not-applicable';\r\n controls: {\r\n evaluationDate: string;\r\n randomSeed: string;\r\n runtimeRevision: string;\r\n fixtureRevision: string;\r\n fixtureRevisionStatus: 'captured' | 'incomplete';\r\n };\r\n}\r\n\r\nexport interface SkillEvaluationReport {\r\n generatedAt: string;\r\n fixtureCount: number;\r\n categoryCounts: Record<SkillEvaluationCategory, number>;\r\n truePositives: number;\r\n falsePositives: number;\r\n falseNegatives: number;\r\n precision: number;\r\n recall: number;\r\n falseActivationBytes: number;\r\n promptChars: { total: number; median: number };\r\n promptTokens: { total: number; median: number };\r\n bodyBytes: { total: number; median: number };\r\n bodyTokens: { total: number; median: number };\r\n activationLatencyMs: { median: number; p95: number };\r\n consentRequests: number;\r\n unnecessaryPermissionPrompts: number;\r\n taskCompletionRate?: number;\r\n userCorrections: number;\r\n skillToolFailures: number;\r\n blockingMismatches: number;\r\n exposure: Record<\r\n 'eager' | 'deferred',\r\n { samples: number; activationRate: number; medianActivationTurns: number }\r\n >;\r\n thresholds: SkillEvaluationThresholds;\r\n rolloutReady: boolean;\r\n reasons: string[];\r\n observations: SkillEvaluationObservation[];\r\n evaluation?: SkillEvaluationExecution;\r\n}\r\n\r\nexport const DEFAULT_SKILL_EVALUATION_THRESHOLDS: SkillEvaluationThresholds = {\r\n minimumPrecision: 0.95,\r\n minimumRecall: 0.8,\r\n maximumFalseActivations: 0,\r\n maximumUnnecessaryPermissionPrompts: 0,\r\n maximumSkillToolFailures: 0,\r\n};\r\n\r\nexport const DEFAULT_SKILL_EVALUATION_FIXTURES: readonly SkillEvaluationPromptFixture[] = [\r\n {\r\n id: 'direct-review',\r\n category: 'direct',\r\n prompt: '$review inspect the current change',\r\n expectedSkills: ['review'],\r\n },\r\n {\r\n id: 'indirect-review',\r\n category: 'indirect',\r\n prompt: 'Inspect this diff for correctness regressions and missing tests',\r\n expectedSkills: ['review'],\r\n },\r\n {\r\n id: 'negative-general-question',\r\n category: 'negative',\r\n prompt: 'Explain what this function returns',\r\n expectedSkills: [],\r\n },\r\n {\r\n id: 'ambiguous-review-word',\r\n category: 'ambiguous',\r\n prompt: 'Review the available deployment choices with me',\r\n expectedSkills: [],\r\n },\r\n {\r\n id: 'conflicting-tool-ceilings',\r\n category: 'conflicting',\r\n prompt: '$read-review $write-review inspect the change',\r\n expectedSkills: ['read-review'],\r\n expectedBlockedCodes: ['skill_tool_intersection_empty'],\r\n },\r\n {\r\n id: 'disabled-skill',\r\n category: 'disabled',\r\n prompt: '$disabled-skill run the workflow',\r\n expectedSkills: [],\r\n expectedBlockedCodes: ['skill_disabled'],\r\n },\r\n {\r\n id: 'invalid-skill',\r\n category: 'invalid',\r\n prompt: '$invalid-skill run the workflow',\r\n expectedSkills: [],\r\n expectedBlockedCodes: ['skill_invalid'],\r\n },\r\n {\r\n id: 'missing-body',\r\n category: 'missing-body',\r\n prompt: '$missing-body run the workflow',\r\n expectedSkills: [],\r\n expectedBlockedCodes: ['skill_load_failed'],\r\n },\r\n {\r\n id: 'missing-resource',\r\n category: 'missing-resource',\r\n prompt: '$missing-resource load references/checklist.md',\r\n expectedSkills: ['missing-resource'],\r\n expectedBlockedCodes: ['skill_resource_unreadable'],\r\n },\r\n];\r\n\r\nfunction numberDetail(event: SkillLifecycleEvent, key: string): number {\r\n const value = event.details?.[key];\r\n return typeof value === 'number' && Number.isFinite(value) ? value : 0;\r\n}\r\n\r\nfunction median(values: readonly number[]): number {\r\n if (values.length === 0) return 0;\r\n const sorted = [...values].sort((left, right) => left - right);\r\n return sorted[Math.floor(sorted.length / 2)] ?? 0;\r\n}\r\n\r\nfunction percentile(values: readonly number[], fraction: number): number {\r\n if (values.length === 0) return 0;\r\n const sorted = [...values].sort((left, right) => left - right);\r\n const index = Math.min(sorted.length - 1, Math.ceil(sorted.length * fraction) - 1);\r\n return sorted[index] ?? 0;\r\n}\r\n\r\n/** Convert bounded lifecycle evidence into one prompt-free evaluation observation. */\r\nexport function observeSkillEvaluation(\r\n fixture: SkillEvaluationFixture,\r\n events: readonly SkillLifecycleEvent[],\r\n outcome: SkillEvaluationOutcome = {},\r\n): SkillEvaluationObservation {\r\n const applied = events.filter(\r\n (event) => event.type === 'skill_activation_applied' && event.skill,\r\n );\r\n const activatedSkills = [...new Set(applied.map((event) => event.skill!))];\r\n const expected = new Set(fixture.expectedSkills);\r\n const falseActivations = applied.filter((event) => event.skill && !expected.has(event.skill));\r\n const consentEvents = events.filter((event) => event.type === 'skill_consent_requested');\r\n const blockedCodes = events\r\n .filter(\r\n (event) =>\r\n event.type === 'skill_activation_blocked' || event.type === 'skill_resource_blocked',\r\n )\r\n .map((event) => event.details?.code)\r\n .filter((code): code is string => typeof code === 'string');\r\n const expectedBlockedCodes = fixture.expectedBlockedCodes ?? [];\r\n const promptChars = outcome.promptChars ?? 0;\r\n const bodyBytes = applied.reduce(\r\n (total, event) => total + numberDetail(event, 'bodyByteSize'),\r\n 0,\r\n );\r\n return {\r\n ...fixture,\r\n expectedSkills: [...fixture.expectedSkills],\r\n activatedSkills,\r\n expectedBlockedCodes: [...expectedBlockedCodes],\r\n blockedCodes,\r\n promptChars,\r\n promptTokens: outcome.promptTokens ?? Math.ceil(promptChars / 4),\r\n bodyBytes,\r\n bodyTokens: outcome.bodyTokens ?? Math.ceil(bodyBytes / 4),\r\n falseActivationBytes: falseActivations.reduce(\r\n (total, event) => total + numberDetail(event, 'bodyByteSize'),\r\n 0,\r\n ),\r\n activationLatencyMs: applied.map((event) => numberDetail(event, 'activationLatencyMs')),\r\n consentRequests: consentEvents.length,\r\n unnecessaryPermissionPrompts: consentEvents.filter(\r\n (event) => !event.skill || !expected.has(event.skill),\r\n ).length,\r\n taskCompleted: outcome.taskCompleted,\r\n userCorrections: outcome.userCorrections ?? 0,\r\n skillToolFailures: outcome.skillToolFailures ?? 0,\r\n blockingMismatch:\r\n expectedBlockedCodes.some((code) => !blockedCodes.includes(code)) ||\r\n blockedCodes.some((code) => !expectedBlockedCodes.includes(code)),\r\n exposure: outcome.exposure,\r\n activationTurns: outcome.activationTurns,\r\n };\r\n}\r\n\r\nexport function evaluateSkillActivation(\r\n observations: readonly SkillEvaluationObservation[],\r\n thresholds: SkillEvaluationThresholds = DEFAULT_SKILL_EVALUATION_THRESHOLDS,\r\n): SkillEvaluationReport {\r\n const categoryCounts = Object.fromEntries(\r\n SKILL_EVALUATION_CATEGORIES.map((category) => [\r\n category,\r\n observations.filter((observation) => observation.category === category).length,\r\n ]),\r\n ) as Record<SkillEvaluationCategory, number>;\r\n let truePositives = 0;\r\n let falsePositives = 0;\r\n let falseNegatives = 0;\r\n for (const observation of observations) {\r\n const expected = new Set(observation.expectedSkills);\r\n const activated = new Set(observation.activatedSkills);\r\n truePositives += [...activated].filter((skill) => expected.has(skill)).length;\r\n falsePositives += [...activated].filter((skill) => !expected.has(skill)).length;\r\n falseNegatives += [...expected].filter((skill) => !activated.has(skill)).length;\r\n }\r\n const precision =\r\n truePositives + falsePositives === 0 ? 1 : truePositives / (truePositives + falsePositives);\r\n const recall =\r\n truePositives + falseNegatives === 0 ? 1 : truePositives / (truePositives + falseNegatives);\r\n const completed = observations.filter(\r\n (observation): observation is SkillEvaluationObservation & { taskCompleted: boolean } =>\r\n observation.taskCompleted !== undefined,\r\n );\r\n const exposure = Object.fromEntries(\r\n (['eager', 'deferred'] as const).map((mode) => {\r\n const samples = observations.filter((observation) => observation.exposure === mode);\r\n return [\r\n mode,\r\n {\r\n samples: samples.length,\r\n activationRate:\r\n samples.length === 0\r\n ? 0\r\n : samples.filter((observation) => observation.activatedSkills.length > 0).length /\r\n samples.length,\r\n medianActivationTurns: median(\r\n samples.map((observation) => observation.activationTurns ?? 0),\r\n ),\r\n },\r\n ];\r\n }),\r\n ) as SkillEvaluationReport['exposure'];\r\n const reasons: string[] = [];\r\n const missingCategories = SKILL_EVALUATION_CATEGORIES.filter(\r\n (category) => categoryCounts[category] === 0,\r\n );\r\n if (missingCategories.length)\r\n reasons.push(`Missing fixture categories: ${missingCategories.join(', ')}.`);\r\n if (precision < thresholds.minimumPrecision) {\r\n reasons.push(\r\n `Activation precision ${precision.toFixed(3)} is below ${thresholds.minimumPrecision}.`,\r\n );\r\n }\r\n if (recall < thresholds.minimumRecall) {\r\n reasons.push(`Activation recall ${recall.toFixed(3)} is below ${thresholds.minimumRecall}.`);\r\n }\r\n if (falsePositives > thresholds.maximumFalseActivations) {\r\n reasons.push(\r\n `False activations ${falsePositives} exceed ${thresholds.maximumFalseActivations}.`,\r\n );\r\n }\r\n const unnecessaryPermissionPrompts = observations.reduce(\r\n (total, observation) => total + observation.unnecessaryPermissionPrompts,\r\n 0,\r\n );\r\n if (unnecessaryPermissionPrompts > thresholds.maximumUnnecessaryPermissionPrompts) {\r\n reasons.push(\r\n `Unnecessary permission prompts ${unnecessaryPermissionPrompts} exceed ${thresholds.maximumUnnecessaryPermissionPrompts}.`,\r\n );\r\n }\r\n const skillToolFailures = observations.reduce(\r\n (total, observation) => total + observation.skillToolFailures,\r\n 0,\r\n );\r\n if (skillToolFailures > thresholds.maximumSkillToolFailures) {\r\n reasons.push(\r\n `Skill-caused tool failures ${skillToolFailures} exceed ${thresholds.maximumSkillToolFailures}.`,\r\n );\r\n }\r\n const blockingMismatches = observations.filter(\r\n (observation) => observation.blockingMismatch,\r\n ).length;\r\n if (blockingMismatches > 0) {\r\n reasons.push(`${blockingMismatches} fixture(s) had unexpected or missing block outcomes.`);\r\n }\r\n const latencies = observations.flatMap((observation) => observation.activationLatencyMs);\r\n return {\r\n generatedAt: new Date().toISOString(),\r\n fixtureCount: observations.length,\r\n categoryCounts,\r\n truePositives,\r\n falsePositives,\r\n falseNegatives,\r\n precision,\r\n recall,\r\n falseActivationBytes: observations.reduce(\r\n (total, observation) => total + observation.falseActivationBytes,\r\n 0,\r\n ),\r\n promptChars: {\r\n total: observations.reduce((total, observation) => total + observation.promptChars, 0),\r\n median: median(observations.map((observation) => observation.promptChars)),\r\n },\r\n promptTokens: {\r\n total: observations.reduce((total, observation) => total + observation.promptTokens, 0),\r\n median: median(observations.map((observation) => observation.promptTokens)),\r\n },\r\n bodyBytes: {\r\n total: observations.reduce((total, observation) => total + observation.bodyBytes, 0),\r\n median: median(observations.map((observation) => observation.bodyBytes)),\r\n },\r\n bodyTokens: {\r\n total: observations.reduce((total, observation) => total + observation.bodyTokens, 0),\r\n median: median(observations.map((observation) => observation.bodyTokens)),\r\n },\r\n activationLatencyMs: { median: median(latencies), p95: percentile(latencies, 0.95) },\r\n consentRequests: observations.reduce(\r\n (total, observation) => total + observation.consentRequests,\r\n 0,\r\n ),\r\n unnecessaryPermissionPrompts,\r\n taskCompletionRate:\r\n completed.length === 0\r\n ? undefined\r\n : completed.filter((observation) => observation.taskCompleted).length / completed.length,\r\n userCorrections: observations.reduce(\r\n (total, observation) => total + observation.userCorrections,\r\n 0,\r\n ),\r\n skillToolFailures,\r\n blockingMismatches,\r\n exposure,\r\n thresholds: { ...thresholds },\r\n rolloutReady: reasons.length === 0,\r\n reasons,\r\n observations: observations.map((observation) => ({\r\n ...observation,\r\n expectedSkills: [...observation.expectedSkills],\r\n expectedBlockedCodes: observation.expectedBlockedCodes\r\n ? [...observation.expectedBlockedCodes]\r\n : undefined,\r\n activatedSkills: [...observation.activatedSkills],\r\n blockedCodes: [...observation.blockedCodes],\r\n activationLatencyMs: [...observation.activationLatencyMs],\r\n })),\r\n };\r\n}\r\n\r\nexport async function runSkillActivationEvaluation(\r\n fixtures: readonly SkillEvaluationPromptFixture[],\r\n execute: (\r\n fixture: SkillEvaluationPromptFixture,\r\n ) => Promise<{ events: readonly SkillLifecycleEvent[]; outcome?: SkillEvaluationOutcome }>,\r\n thresholds: SkillEvaluationThresholds = DEFAULT_SKILL_EVALUATION_THRESHOLDS,\r\n): Promise<SkillEvaluationReport> {\r\n const observations: SkillEvaluationObservation[] = [];\r\n for (const fixture of fixtures) {\r\n const result = await execute(fixture);\r\n observations.push(\r\n observeSkillEvaluation(\r\n {\r\n id: fixture.id,\r\n category: fixture.category,\r\n promptHash: createHash('sha256').update(fixture.prompt).digest('hex'),\r\n expectedSkills: [...fixture.expectedSkills],\r\n expectedBlockedCodes: fixture.expectedBlockedCodes\r\n ? [...fixture.expectedBlockedCodes]\r\n : undefined,\r\n },\r\n result.events,\r\n result.outcome,\r\n ),\r\n );\r\n }\r\n return evaluateSkillActivation(observations, thresholds);\r\n}\r\n\r\nexport function renderSkillEvaluationReport(report: SkillEvaluationReport): string {\r\n const categories = SKILL_EVALUATION_CATEGORIES.map(\r\n (category) => `- ${category}: ${report.categoryCounts[category]}`,\r\n ).join('\\n');\r\n return [\r\n '# Skill Activation Evaluation',\r\n '',\r\n `Generated: ${report.generatedAt}`,\r\n `Rollout ready: ${report.rolloutReady ? 'yes' : 'no'}`,\r\n `Precision: ${report.precision.toFixed(3)}`,\r\n `Recall: ${report.recall.toFixed(3)}`,\r\n `False activations: ${report.falsePositives}`,\r\n `False-activation body bytes: ${report.falseActivationBytes}`,\r\n `Median prompt tokens: ${report.promptTokens.median}`,\r\n `Median activated body tokens: ${report.bodyTokens.median}`,\r\n `Median activation latency: ${report.activationLatencyMs.median} ms`,\r\n `Unnecessary permission prompts: ${report.unnecessaryPermissionPrompts}`,\r\n `Skill-caused tool failures: ${report.skillToolFailures}`,\r\n `Blocking mismatches: ${report.blockingMismatches}`,\r\n ...(report.evaluation\r\n ? [\r\n `Evidence kind: ${report.evaluation.evidenceKind}`,\r\n `Provider run eligibility: ${report.evaluation.providerRunEligibility}`,\r\n `Evaluation date: ${report.evaluation.controls.evaluationDate}`,\r\n `Evaluation seed: ${report.evaluation.controls.randomSeed}`,\r\n `Runtime revision: ${report.evaluation.controls.runtimeRevision}`,\r\n `Fixture revision: ${report.evaluation.controls.fixtureRevision} (${report.evaluation.controls.fixtureRevisionStatus})`,\r\n ]\r\n : []),\r\n '',\r\n '## Fixture Coverage',\r\n '',\r\n categories,\r\n '',\r\n '## Gate',\r\n '',\r\n ...(report.reasons.length\r\n ? report.reasons.map((reason) => `- ${reason}`)\r\n : ['- All thresholds passed.']),\r\n '',\r\n ].join('\\n');\r\n}\r\n\r\nexport function writeSkillEvaluationReport(\r\n report: SkillEvaluationReport,\r\n jsonPath: string,\r\n markdownPath?: string,\r\n): void {\r\n mkdirSync(dirname(jsonPath), { recursive: true });\r\n writeFileSync(jsonPath, `${JSON.stringify(report, null, 2)}\\n`, 'utf8');\r\n if (markdownPath) {\r\n mkdirSync(dirname(markdownPath), { recursive: true });\r\n writeFileSync(markdownPath, renderSkillEvaluationReport(report), 'utf8');\r\n }\r\n}\r\n","import { createHash } from 'crypto';\r\nimport { mkdirSync, writeFileSync } from 'fs';\r\nimport { dirname } from 'path';\r\nimport type { AgentMode, AgentTopology, IssueQuality } from './types.js';\r\n\r\nexport interface EvaluationFixture {\r\n id: string;\r\n cohort: 'decomposable' | 'ambiguous' | 'sequential';\r\n prompt: string;\r\n}\r\n\r\nexport interface EvaluationRunResult {\r\n passed: boolean;\r\n validationCorrect?: boolean;\r\n wallTimeMs: number;\r\n totalTokens?: number;\r\n costUsd?: number;\r\n spawnCount?: number;\r\n route?: AgentTopology;\r\n issueQuality?: IssueQuality;\r\n conflicts?: number;\r\n interruptions?: number;\r\n unvalidatedApplications?: number;\r\n}\r\n\r\nexport interface EvaluationMetric extends Omit<EvaluationRunResult, 'route' | 'issueQuality'> {\r\n fixtureId: string;\r\n promptHash: string;\r\n cohort: EvaluationFixture['cohort'];\r\n mode: Extract<AgentMode, 'adaptive' | 'off'>;\r\n route?: AgentTopology;\r\n issueQuality?: IssueQuality;\r\n}\r\n\r\n/** Run each fixture with the single-agent control and adaptive treatment. */\r\nexport async function runPairedEvaluation(\r\n fixtures: EvaluationFixture[],\r\n execute: (\r\n fixture: EvaluationFixture,\r\n mode: Extract<AgentMode, 'adaptive' | 'off'>,\r\n ) => Promise<EvaluationRunResult>,\r\n outputPath: string,\r\n): Promise<EvaluationMetric[]> {\r\n const metrics: EvaluationMetric[] = [];\r\n for (const fixture of fixtures) {\r\n for (const mode of ['off', 'adaptive'] as const) {\r\n const result = await execute(fixture, mode);\r\n metrics.push({\r\n fixtureId: fixture.id,\r\n promptHash: createHash('sha256').update(fixture.prompt).digest('hex'),\r\n cohort: fixture.cohort,\r\n mode,\r\n ...result,\r\n });\r\n }\r\n }\r\n mkdirSync(dirname(outputPath), { recursive: true });\r\n writeFileSync(\r\n outputPath,\r\n `${metrics.map((metric) => JSON.stringify(metric)).join('\\n')}\\n`,\r\n 'utf8',\r\n );\r\n return metrics;\r\n}\r\n\r\nexport function evaluateSuccess(metrics: EvaluationMetric[]): {\r\n successful: boolean;\r\n reasons: string[];\r\n} {\r\n const reasons: string[] = [];\r\n const byMode = (cohorts: EvaluationFixture['cohort'][], mode: EvaluationMetric['mode']) =>\r\n metrics.filter((metric) => cohorts.includes(metric.cohort) && metric.mode === mode);\r\n const rate = (items: EvaluationMetric[]) =>\r\n items.length === 0 ? 0 : items.filter((item) => item.passed).length / items.length;\r\n const median = (values: number[]) => {\r\n if (values.length === 0) return 0;\r\n const sorted = [...values].sort((left, right) => left - right);\r\n return sorted[Math.floor(sorted.length / 2)];\r\n };\r\n\r\n const treatment = byMode(['decomposable', 'ambiguous'], 'adaptive');\r\n const control = byMode(['decomposable', 'ambiguous'], 'off');\r\n const passGain = rate(treatment) - rate(control);\r\n const treatmentWall = median(treatment.map((item) => item.wallTimeMs));\r\n const controlWall = median(control.map((item) => item.wallTimeMs));\r\n if (passGain < 0.05 && !(passGain >= 0 && treatmentWall <= controlWall * 0.8)) {\r\n reasons.push('Decomposable/ambiguous cohort missed the pass-rate or wall-time target.');\r\n }\r\n\r\n const sequentialTreatment = byMode(['sequential'], 'adaptive');\r\n const sequentialControl = byMode(['sequential'], 'off');\r\n if (rate(sequentialTreatment) < rate(sequentialControl) - 0.02) {\r\n reasons.push('Sequential cohort regressed by more than two pass-rate points.');\r\n }\r\n\r\n const adaptiveTokens = median(\r\n metrics.filter((item) => item.mode === 'adaptive').map((item) => item.totalTokens ?? 0),\r\n );\r\n const controlTokens = median(\r\n metrics.filter((item) => item.mode === 'off').map((item) => item.totalTokens ?? 0),\r\n );\r\n if (controlTokens > 0 && adaptiveTokens >= controlTokens * 3) {\r\n reasons.push('Median adaptive token overhead is not below 3x.');\r\n }\r\n if (metrics.some((item) => (item.unvalidatedApplications ?? 0) > 0)) {\r\n reasons.push('At least one unvalidated patch was applied.');\r\n }\r\n return { successful: reasons.length === 0, reasons };\r\n}\r\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAEA,SAAS,YAAY;;;ACFrB,SAAS,kBAAkB;AAC3B,SAAS,WAAW,qBAAqB;AACzC,SAAS,eAAe;AAGjB,IAAM,8BAA8B;AAAA,EACzC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAsGO,IAAM,sCAAiE;AAAA,EAC5E,kBAAkB;AAAA,EAClB,eAAe;AAAA,EACf,yBAAyB;AAAA,EACzB,qCAAqC;AAAA,EACrC,0BAA0B;AAC5B;AAEO,IAAM,oCAA6E;AAAA,EACxF;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,QAAQ;AAAA,EAC3B;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,QAAQ;AAAA,EAC3B;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,EACnB;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,EACnB;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,aAAa;AAAA,IAC9B,sBAAsB,CAAC,+BAA+B;AAAA,EACxD;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,IACjB,sBAAsB,CAAC,gBAAgB;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,IACjB,sBAAsB,CAAC,eAAe;AAAA,EACxC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,IACjB,sBAAsB,CAAC,mBAAmB;AAAA,EAC5C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,kBAAkB;AAAA,IACnC,sBAAsB,CAAC,2BAA2B;AAAA,EACpD;AACF;AAEA,SAAS,aAAa,OAA4B,KAAqB;AACrE,QAAM,QAAQ,MAAM,UAAU,GAAG;AACjC,SAAO,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,IAAI,QAAQ;AACvE;AAEA,SAAS,OAAO,QAAmC;AACjD,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,MAAM,UAAU,OAAO,KAAK;AAC7D,SAAO,OAAO,KAAK,MAAM,OAAO,SAAS,CAAC,CAAC,KAAK;AAClD;AAEA,SAAS,WAAW,QAA2B,UAA0B;AACvE,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,MAAM,UAAU,OAAO,KAAK;AAC7D,QAAM,QAAQ,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,KAAK,OAAO,SAAS,QAAQ,IAAI,CAAC;AACjF,SAAO,OAAO,KAAK,KAAK;AAC1B;AAGO,SAAS,uBACd,SACA,QACA,UAAkC,CAAC,GACP;AAC5B,QAAM,UAAU,OAAO;AAAA,IACrB,CAAC,UAAU,MAAM,SAAS,8BAA8B,MAAM;AAAA,EAChE;AACA,QAAM,kBAAkB,CAAC,GAAG,IAAI,IAAI,QAAQ,IAAI,CAAC,UAAU,MAAM,KAAM,CAAC,CAAC;AACzE,QAAM,WAAW,IAAI,IAAI,QAAQ,cAAc;AAC/C,QAAM,mBAAmB,QAAQ,OAAO,CAAC,UAAU,MAAM,SAAS,CAAC,SAAS,IAAI,MAAM,KAAK,CAAC;AAC5F,QAAM,gBAAgB,OAAO,OAAO,CAAC,UAAU,MAAM,SAAS,yBAAyB;AACvF,QAAM,eAAe,OAClB;AAAA,IACC,CAAC,UACC,MAAM,SAAS,8BAA8B,MAAM,SAAS;AAAA,EAChE,EACC,IAAI,CAAC,UAAU,MAAM,SAAS,IAAI,EAClC,OAAO,CAAC,SAAyB,OAAO,SAAS,QAAQ;AAC5D,QAAM,uBAAuB,QAAQ,wBAAwB,CAAC;AAC9D,QAAM,cAAc,QAAQ,eAAe;AAC3C,QAAM,YAAY,QAAQ;AAAA,IACxB,CAAC,OAAO,UAAU,QAAQ,aAAa,OAAO,cAAc;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AAAA,IACL,GAAG;AAAA,IACH,gBAAgB,CAAC,GAAG,QAAQ,cAAc;AAAA,IAC1C;AAAA,IACA,sBAAsB,CAAC,GAAG,oBAAoB;AAAA,IAC9C;AAAA,IACA;AAAA,IACA,cAAc,QAAQ,gBAAgB,KAAK,KAAK,cAAc,CAAC;AAAA,IAC/D;AAAA,IACA,YAAY,QAAQ,cAAc,KAAK,KAAK,YAAY,CAAC;AAAA,IACzD,sBAAsB,iBAAiB;AAAA,MACrC,CAAC,OAAO,UAAU,QAAQ,aAAa,OAAO,cAAc;AAAA,MAC5D;AAAA,IACF;AAAA,IACA,qBAAqB,QAAQ,IAAI,CAAC,UAAU,aAAa,OAAO,qBAAqB,CAAC;AAAA,IACtF,iBAAiB,cAAc;AAAA,IAC/B,8BAA8B,cAAc;AAAA,MAC1C,CAAC,UAAU,CAAC,MAAM,SAAS,CAAC,SAAS,IAAI,MAAM,KAAK;AAAA,IACtD,EAAE;AAAA,IACF,eAAe,QAAQ;AAAA,IACvB,iBAAiB,QAAQ,mBAAmB;AAAA,IAC5C,mBAAmB,QAAQ,qBAAqB;AAAA,IAChD,kBACE,qBAAqB,KAAK,CAAC,SAAS,CAAC,aAAa,SAAS,IAAI,CAAC,KAChE,aAAa,KAAK,CAAC,SAAS,CAAC,qBAAqB,SAAS,IAAI,CAAC;AAAA,IAClE,UAAU,QAAQ;AAAA,IAClB,iBAAiB,QAAQ;AAAA,EAC3B;AACF;AAEO,SAAS,wBACd,cACA,aAAwC,qCACjB;AACvB,QAAM,iBAAiB,OAAO;AAAA,IAC5B,4BAA4B,IAAI,CAAC,aAAa;AAAA,MAC5C;AAAA,MACA,aAAa,OAAO,CAAC,gBAAgB,YAAY,aAAa,QAAQ,EAAE;AAAA,IAC1E,CAAC;AAAA,EACH;AACA,MAAI,gBAAgB;AACpB,MAAI,iBAAiB;AACrB,MAAI,iBAAiB;AACrB,aAAW,eAAe,cAAc;AACtC,UAAM,WAAW,IAAI,IAAI,YAAY,cAAc;AACnD,UAAM,YAAY,IAAI,IAAI,YAAY,eAAe;AACrD,qBAAiB,CAAC,GAAG,SAAS,EAAE,OAAO,CAAC,UAAU,SAAS,IAAI,KAAK,CAAC,EAAE;AACvE,sBAAkB,CAAC,GAAG,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,SAAS,IAAI,KAAK,CAAC,EAAE;AACzE,sBAAkB,CAAC,GAAG,QAAQ,EAAE,OAAO,CAAC,UAAU,CAAC,UAAU,IAAI,KAAK,CAAC,EAAE;AAAA,EAC3E;AACA,QAAM,YACJ,gBAAgB,mBAAmB,IAAI,IAAI,iBAAiB,gBAAgB;AAC9E,QAAM,SACJ,gBAAgB,mBAAmB,IAAI,IAAI,iBAAiB,gBAAgB;AAC9E,QAAM,YAAY,aAAa;AAAA,IAC7B,CAAC,gBACC,YAAY,kBAAkB;AAAA,EAClC;AACA,QAAM,WAAW,OAAO;AAAA,IACrB,CAAC,SAAS,UAAU,EAAY,IAAI,CAAC,SAAS;AAC7C,YAAM,UAAU,aAAa,OAAO,CAAC,gBAAgB,YAAY,aAAa,IAAI;AAClF,aAAO;AAAA,QACL;AAAA,QACA;AAAA,UACE,SAAS,QAAQ;AAAA,UACjB,gBACE,QAAQ,WAAW,IACf,IACA,QAAQ,OAAO,CAAC,gBAAgB,YAAY,gBAAgB,SAAS,CAAC,EAAE,SACxE,QAAQ;AAAA,UACd,uBAAuB;AAAA,YACrB,QAAQ,IAAI,CAAC,gBAAgB,YAAY,mBAAmB,CAAC;AAAA,UAC/D;AAAA,QACF;AAAA,MACF;AAAA,IACF,CAAC;AAAA,EACH;AACA,QAAM,UAAoB,CAAC;AAC3B,QAAM,oBAAoB,4BAA4B;AAAA,IACpD,CAAC,aAAa,eAAe,QAAQ,MAAM;AAAA,EAC7C;AACA,MAAI,kBAAkB;AACpB,YAAQ,KAAK,+BAA+B,kBAAkB,KAAK,IAAI,CAAC,GAAG;AAC7E,MAAI,YAAY,WAAW,kBAAkB;AAC3C,YAAQ;AAAA,MACN,wBAAwB,UAAU,QAAQ,CAAC,CAAC,aAAa,WAAW,gBAAgB;AAAA,IACtF;AAAA,EACF;AACA,MAAI,SAAS,WAAW,eAAe;AACrC,YAAQ,KAAK,qBAAqB,OAAO,QAAQ,CAAC,CAAC,aAAa,WAAW,aAAa,GAAG;AAAA,EAC7F;AACA,MAAI,iBAAiB,WAAW,yBAAyB;AACvD,YAAQ;AAAA,MACN,qBAAqB,cAAc,WAAW,WAAW,uBAAuB;AAAA,IAClF;AAAA,EACF;AACA,QAAM,+BAA+B,aAAa;AAAA,IAChD,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,IAC5C;AAAA,EACF;AACA,MAAI,+BAA+B,WAAW,qCAAqC;AACjF,YAAQ;AAAA,MACN,kCAAkC,4BAA4B,WAAW,WAAW,mCAAmC;AAAA,IACzH;AAAA,EACF;AACA,QAAM,oBAAoB,aAAa;AAAA,IACrC,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,IAC5C;AAAA,EACF;AACA,MAAI,oBAAoB,WAAW,0BAA0B;AAC3D,YAAQ;AAAA,MACN,8BAA8B,iBAAiB,WAAW,WAAW,wBAAwB;AAAA,IAC/F;AAAA,EACF;AACA,QAAM,qBAAqB,aAAa;AAAA,IACtC,CAAC,gBAAgB,YAAY;AAAA,EAC/B,EAAE;AACF,MAAI,qBAAqB,GAAG;AAC1B,YAAQ,KAAK,GAAG,kBAAkB,uDAAuD;AAAA,EAC3F;AACA,QAAM,YAAY,aAAa,QAAQ,CAAC,gBAAgB,YAAY,mBAAmB;AACvF,SAAO;AAAA,IACL,cAAa,oBAAI,KAAK,GAAE,YAAY;AAAA,IACpC,cAAc,aAAa;AAAA,IAC3B;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,sBAAsB,aAAa;AAAA,MACjC,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,MAC5C;AAAA,IACF;AAAA,IACA,aAAa;AAAA,MACX,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,aAAa,CAAC;AAAA,MACrF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,WAAW,CAAC;AAAA,IAC3E;AAAA,IACA,cAAc;AAAA,MACZ,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,cAAc,CAAC;AAAA,MACtF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,YAAY,CAAC;AAAA,IAC5E;AAAA,IACA,WAAW;AAAA,MACT,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,WAAW,CAAC;AAAA,MACnF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,SAAS,CAAC;AAAA,IACzE;AAAA,IACA,YAAY;AAAA,MACV,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,YAAY,CAAC;AAAA,MACpF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,UAAU,CAAC;AAAA,IAC1E;AAAA,IACA,qBAAqB,EAAE,QAAQ,OAAO,SAAS,GAAG,KAAK,WAAW,WAAW,IAAI,EAAE;AAAA,IACnF,iBAAiB,aAAa;AAAA,MAC5B,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,MAC5C;AAAA,IACF;AAAA,IACA;AAAA,IACA,oBACE,UAAU,WAAW,IACjB,SACA,UAAU,OAAO,CAAC,gBAAgB,YAAY,aAAa,EAAE,SAAS,UAAU;AAAA,IACtF,iBAAiB,aAAa;AAAA,MAC5B,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,MAC5C;AAAA,IACF;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,YAAY,EAAE,GAAG,WAAW;AAAA,IAC5B,cAAc,QAAQ,WAAW;AAAA,IACjC;AAAA,IACA,cAAc,aAAa,IAAI,CAAC,iBAAiB;AAAA,MAC/C,GAAG;AAAA,MACH,gBAAgB,CAAC,GAAG,YAAY,cAAc;AAAA,MAC9C,sBAAsB,YAAY,uBAC9B,CAAC,GAAG,YAAY,oBAAoB,IACpC;AAAA,MACJ,iBAAiB,CAAC,GAAG,YAAY,eAAe;AAAA,MAChD,cAAc,CAAC,GAAG,YAAY,YAAY;AAAA,MAC1C,qBAAqB,CAAC,GAAG,YAAY,mBAAmB;AAAA,IAC1D,EAAE;AAAA,EACJ;AACF;AAEA,eAAsB,6BACpB,UACA,SAGA,aAAwC,qCACR;AAChC,QAAM,eAA6C,CAAC;AACpD,aAAW,WAAW,UAAU;AAC9B,UAAM,SAAS,MAAM,QAAQ,OAAO;AACpC,iBAAa;AAAA,MACX;AAAA,QACE;AAAA,UACE,IAAI,QAAQ;AAAA,UACZ,UAAU,QAAQ;AAAA,UAClB,YAAY,WAAW,QAAQ,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,KAAK;AAAA,UACpE,gBAAgB,CAAC,GAAG,QAAQ,cAAc;AAAA,UAC1C,sBAAsB,QAAQ,uBAC1B,CAAC,GAAG,QAAQ,oBAAoB,IAChC;AAAA,QACN;AAAA,QACA,OAAO;AAAA,QACP,OAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AACA,SAAO,wBAAwB,cAAc,UAAU;AACzD;AAEO,SAAS,4BAA4B,QAAuC;AACjF,QAAM,aAAa,4BAA4B;AAAA,IAC7C,CAAC,aAAa,KAAK,QAAQ,KAAK,OAAO,eAAe,QAAQ,CAAC;AAAA,EACjE,EAAE,KAAK,IAAI;AACX,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,cAAc,OAAO,WAAW;AAAA,IAChC,kBAAkB,OAAO,eAAe,QAAQ,IAAI;AAAA,IACpD,cAAc,OAAO,UAAU,QAAQ,CAAC,CAAC;AAAA,IACzC,WAAW,OAAO,OAAO,QAAQ,CAAC,CAAC;AAAA,IACnC,sBAAsB,OAAO,cAAc;AAAA,IAC3C,gCAAgC,OAAO,oBAAoB;AAAA,IAC3D,yBAAyB,OAAO,aAAa,MAAM;AAAA,IACnD,iCAAiC,OAAO,WAAW,MAAM;AAAA,IACzD,8BAA8B,OAAO,oBAAoB,MAAM;AAAA,IAC/D,mCAAmC,OAAO,4BAA4B;AAAA,IACtE,+BAA+B,OAAO,iBAAiB;AAAA,IACvD,wBAAwB,OAAO,kBAAkB;AAAA,IACjD,GAAI,OAAO,aACP;AAAA,MACE,kBAAkB,OAAO,WAAW,YAAY;AAAA,MAChD,6BAA6B,OAAO,WAAW,sBAAsB;AAAA,MACrE,oBAAoB,OAAO,WAAW,SAAS,cAAc;AAAA,MAC7D,oBAAoB,OAAO,WAAW,SAAS,UAAU;AAAA,MACzD,qBAAqB,OAAO,WAAW,SAAS,eAAe;AAAA,MAC/D,qBAAqB,OAAO,WAAW,SAAS,eAAe,KAAK,OAAO,WAAW,SAAS,qBAAqB;AAAA,IACtH,IACA,CAAC;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAI,OAAO,QAAQ,SACf,OAAO,QAAQ,IAAI,CAAC,WAAW,KAAK,MAAM,EAAE,IAC5C,CAAC,0BAA0B;AAAA,IAC/B;AAAA,EACF,EAAE,KAAK,IAAI;AACb;AAEO,SAAS,2BACd,QACA,UACA,cACM;AACN,YAAU,QAAQ,QAAQ,GAAG,EAAE,WAAW,KAAK,CAAC;AAChD,gBAAc,UAAU,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,GAAM,MAAM;AACtE,MAAI,cAAc;AAChB,cAAU,QAAQ,YAAY,GAAG,EAAE,WAAW,KAAK,CAAC;AACpD,kBAAc,cAAc,4BAA4B,MAAM,GAAG,MAAM;AAAA,EACzE;AACF;;;ACjfA,SAAS,cAAAA,mBAAkB;AAC3B,SAAS,aAAAC,YAAW,iBAAAC,sBAAqB;AACzC,SAAS,WAAAC,gBAAe;AAiCxB,eAAsB,oBACpB,UACA,SAIA,YAC6B;AAC7B,QAAM,UAA8B,CAAC;AACrC,aAAW,WAAW,UAAU;AAC9B,eAAW,QAAQ,CAAC,OAAO,UAAU,GAAY;AAC/C,YAAM,SAAS,MAAM,QAAQ,SAAS,IAAI;AAC1C,cAAQ,KAAK;AAAA,QACX,WAAW,QAAQ;AAAA,QACnB,YAAYH,YAAW,QAAQ,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,KAAK;AAAA,QACpE,QAAQ,QAAQ;AAAA,QAChB;AAAA,QACA,GAAG;AAAA,MACL,CAAC;AAAA,IACH;AAAA,EACF;AACA,EAAAC,WAAUE,SAAQ,UAAU,GAAG,EAAE,WAAW,KAAK,CAAC;AAClD,EAAAD;AAAA,IACE;AAAA,IACA,GAAG,QAAQ,IAAI,CAAC,WAAW,KAAK,UAAU,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC;AAAA;AAAA,IAC7D;AAAA,EACF;AACA,SAAO;AACT;AAEO,SAAS,gBAAgB,SAG9B;AACA,QAAM,UAAoB,CAAC;AAC3B,QAAM,SAAS,CAAC,SAAwC,SACtD,QAAQ,OAAO,CAAC,WAAW,QAAQ,SAAS,OAAO,MAAM,KAAK,OAAO,SAAS,IAAI;AACpF,QAAM,OAAO,CAAC,UACZ,MAAM,WAAW,IAAI,IAAI,MAAM,OAAO,CAAC,SAAS,KAAK,MAAM,EAAE,SAAS,MAAM;AAC9E,QAAME,UAAS,CAAC,WAAqB;AACnC,QAAI,OAAO,WAAW,EAAG,QAAO;AAChC,UAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,MAAM,UAAU,OAAO,KAAK;AAC7D,WAAO,OAAO,KAAK,MAAM,OAAO,SAAS,CAAC,CAAC;AAAA,EAC7C;AAEA,QAAM,YAAY,OAAO,CAAC,gBAAgB,WAAW,GAAG,UAAU;AAClE,QAAM,UAAU,OAAO,CAAC,gBAAgB,WAAW,GAAG,KAAK;AAC3D,QAAM,WAAW,KAAK,SAAS,IAAI,KAAK,OAAO;AAC/C,QAAM,gBAAgBA,QAAO,UAAU,IAAI,CAAC,SAAS,KAAK,UAAU,CAAC;AACrE,QAAM,cAAcA,QAAO,QAAQ,IAAI,CAAC,SAAS,KAAK,UAAU,CAAC;AACjE,MAAI,WAAW,QAAQ,EAAE,YAAY,KAAK,iBAAiB,cAAc,MAAM;AAC7E,YAAQ,KAAK,yEAAyE;AAAA,EACxF;AAEA,QAAM,sBAAsB,OAAO,CAAC,YAAY,GAAG,UAAU;AAC7D,QAAM,oBAAoB,OAAO,CAAC,YAAY,GAAG,KAAK;AACtD,MAAI,KAAK,mBAAmB,IAAI,KAAK,iBAAiB,IAAI,MAAM;AAC9D,YAAQ,KAAK,gEAAgE;AAAA,EAC/E;AAEA,QAAM,iBAAiBA;AAAA,IACrB,QAAQ,OAAO,CAAC,SAAS,KAAK,SAAS,UAAU,EAAE,IAAI,CAAC,SAAS,KAAK,eAAe,CAAC;AAAA,EACxF;AACA,QAAM,gBAAgBA;AAAA,IACpB,QAAQ,OAAO,CAAC,SAAS,KAAK,SAAS,KAAK,EAAE,IAAI,CAAC,SAAS,KAAK,eAAe,CAAC;AAAA,EACnF;AACA,MAAI,gBAAgB,KAAK,kBAAkB,gBAAgB,GAAG;AAC5D,YAAQ,KAAK,iDAAiD;AAAA,EAChE;AACA,MAAI,QAAQ,KAAK,CAAC,UAAU,KAAK,2BAA2B,KAAK,CAAC,GAAG;AACnE,YAAQ,KAAK,6CAA6C;AAAA,EAC5D;AACA,SAAO,EAAE,YAAY,QAAQ,WAAW,GAAG,QAAQ;AACrD;;;AF1DA,IAAM,kBAAN,MAAyB;AAAA,EACf,SAAc,CAAC;AAAA,EACf,UAAiD,CAAC;AAAA,EAClD,SAAS;AAAA,EAEjB,KAAK,OAAgB;AACnB,QAAI,KAAK,OAAQ;AACjB,UAAM,SAAS,KAAK,QAAQ,MAAM;AAClC,QAAI,OAAQ,QAAO,KAAK;AAAA,QACnB,MAAK,OAAO,KAAK,KAAK;AAAA,EAC7B;AAAA,EAEA,QAAc;AACZ,QAAI,KAAK,OAAQ;AACjB,SAAK,SAAS;AACd,eAAW,UAAU,KAAK,QAAQ,OAAO,CAAC,EAAG,QAAO,MAAS;AAAA,EAC/D;AAAA,EAEA,OAA+B;AAC7B,UAAM,QAAQ,KAAK,OAAO,MAAM;AAChC,QAAI,UAAU,OAAW,QAAO,QAAQ,QAAQ,KAAK;AACrD,QAAI,KAAK,OAAQ,QAAO,QAAQ,QAAQ,MAAS;AACjD,WAAO,IAAI,QAAQ,CAAC,YAAY,KAAK,QAAQ,KAAK,OAAO,CAAC;AAAA,EAC5D;AACF;AAEA,gBAAuB,MACrB,QACA,UAAwB,CAAC,GACoB;AAC7C,QAAM,YAAY,QAAQ,aAAa,QAAQ,IAAI;AACnD,QAAM,eAAe,WAAW,WAAW;AAAA,IACzC,sBAAsB,QAAQ;AAAA,IAC9B,YAAY,QAAQ;AAAA,IACpB,eAAe,CAAC,QAAQ;AAAA,EAC1B,CAAC;AACD,QAAM,SAAS,kBAAkB;AAAA,IAC/B,GAAG;AAAA,IACH,GAAI,QAAQ,QAAQ,EAAE,OAAO,QAAQ,MAAM,IAAI,CAAC;AAAA,IAChD,GAAI,QAAQ,WAAW,EAAE,UAAU,QAAQ,SAAS,IAAI,CAAC;AAAA,IACzD,GAAI,QAAQ,SACR;AAAA,MACE,UAAU;AAAA,QACR,GAAG,aAAa;AAAA,QAChB,QAAQ,EAAE,GAAG,aAAa,SAAS,QAAQ,MAAM,QAAQ,OAAO;AAAA,MAClE;AAAA,IACF,IACA,CAAC;AAAA,EACP,CAAC;AAED,QAAM,aAAa,IAAI,gBAAgB;AACvC,QAAM,kBAAkB,MAAM,WAAW,MAAM,QAAQ,QAAQ,MAAM;AACrE,MAAI,QAAQ,QAAQ,QAAS,iBAAgB;AAAA,MACxC,SAAQ,QAAQ,iBAAiB,SAAS,iBAAiB,EAAE,MAAM,KAAK,CAAC;AAE9E,QAAM,iBAAiB,QAAQ,mBAAmB;AAClD,QAAM,eAAe,iBAChB,QAAQ,gBAAgB,IAAI,aAAa,KAAK,gBAAgB,GAAG,UAAU,CAAC,IAC7E;AACJ,QAAM,YAAY,wBAAwB,cAAc;AAAA,IACtD,KAAK,OAAO;AAAA,IACZ,WAAW,QAAQ;AAAA,EACrB,CAAC;AACD,MAAI,kBAAkB,CAAC,QAAQ,gBAAgB,cAAc;AAC3D,qBAAiB,mCAAmC,gBAAgB,CAAC;AACrE,iBAAa,QAAQ,mCAAmC,oBAAI,IAAI,CAAC,UAAU,SAAS,CAAC,CAAC;AAAA,EACxF;AACA,QAAM,QAAQ,IAAI,gBAA4B;AAC9C,MAAI,cAA4E,CAAC;AAEjF,QAAM,KAAK,EAAE,MAAM,UAAU,OAAO,OAAO,OAAO,KAAK,OAAO,UAAU,CAAC;AACzE,QAAM,KAAK,EAAE,MAAM,WAAW,WAAW,UAAU,UAAU,CAAC;AAE9D,QAAM,WAAW,YAAY;AAC3B,QAAI;AAKF,iBAAW,UAAU,8BAA8B;AAAA,QACjD,WAAW,OAAO;AAAA,QAClB,UAAU,OAAO;AAAA,QACjB,iBAAiB,CAAC,QAAQ;AAAA,MAC5B,CAAC,GAAG;AACF,gBAAQ,KAAK,MAAM;AAAA,MACrB;AACA,YAAM,qBAAqB,qBAAqB,OAAO,SAAS;AAChE,YAAM,eAAe,8BAA8B,oBAAoB,OAAO,QAAQ;AACtF,iBAAW,UAAU,aAAa,SAAS;AACzC,gBAAQ;AAAA,UACN,gCAA2B,OAAO,IAAI,MAAM,OAAO,IAAI;AAAA,QACzD;AAAA,MACF;AACA,YAAM,MAAM,MAAM,kBAAkB,OAAO,WAAW;AAAA,QACpD,QAAQ,WAAW;AAAA,QACnB,SAAS,mBAAmB,aAAa,OAAO;AAAA,QAChD,UAAU,QAAQ;AAAA,MACpB,CAAC;AACD,oBAAc,IAAI;AAClB,YAAM,WAAW,sBAAsB,EAAE,QAAQ,OAAO,SAAS,OAAO,SAAS,MAAM,CAAC;AACxF,UAAI,IAAI,MAAM,SAAS,EAAG,UAAS,YAAY,IAAI,KAAK;AAExD,YAAM,SAAS,MAAM,YAAY,QAAQ,UAAU;AAAA,QACjD;AAAA,QACA,aAAa;AAAA,QACb,cAAc;AAAA,QACd,SAAS,UAAU;AAAA,QACnB,YAAY,UAAU;AAAA,QACtB,mBAAmB,UAAU;AAAA,QAC7B,MAAM,sBAAsB,OAAO,UAAU,QAAQ,cAAc;AAAA,QACnE,UAAU,QAAQ;AAAA,QAClB;AAAA,QACA;AAAA,QACA,WAAW,UAAU;AAAA,QACrB,gBAAgB,UAAU;AAAA,QAC1B,WAAW;AAAA,QACX,QAAQ,WAAW;AAAA,QACnB,wBAAwB,QAAQ;AAAA,QAChC,qBAAqB,QAAQ;AAAA,QAC7B,QAAQ,EAAE,OAAO,MAAM,KAAK;AAAA,QAC5B,cAAc,CAAC,UAAU;AACvB,cACE,MAAM,SAAS,YACf,MAAM,SAAS,aACf,MAAM,SAAS,YACf,MAAM,SAAS,WACd,MAAM,SAAS,sBAAsB,QAAQ,wBAAwB,OACtE;AACA,kBAAM,KAAK,KAAK;AAAA,UAClB;AAAA,QACF;AAAA,MACF,CAAC;AACD,YAAM,KAAK;AAAA,QACT,MAAM;AAAA,QACN,UAAU,OAAO;AAAA,QACjB,OAAO,OAAO;AAAA,QACd,WAAW,UAAU;AAAA,QACrB,SAAS,OAAO;AAAA,QAChB,MAAM,OAAO;AAAA;AAAA;AAAA,QAGb,gBAAgB,OAAO;AAAA,MACzB,CAAC;AAAA,IACH,SAAS,OAAO;AACd,UAAI,CAAC,WAAW,OAAO,SAAS;AAC9B,cAAM,KAAK;AAAA,UACT,MAAM;AAAA,UACN,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;AAAA,QAC9D,CAAC;AAAA,MACH;AAAA,IACF,UAAE;AACA,YAAM,qBAAqB,WAAW;AACtC,YAAM,KAAK,EAAE,MAAM,OAAO,CAAC;AAC3B,YAAM,MAAM;AAAA,IACd;AAAA,EACF,GAAG;AAEH,MAAI;AACF,WAAO,MAAM;AACX,YAAM,QAAQ,MAAM,MAAM,KAAK;AAC/B,UAAI,CAAC,MAAO;AACZ,YAAM;AAAA,IACR;AAAA,EACF,UAAE;AACA,eAAW,MAAM,IAAI,MAAM,4BAA4B,CAAC;AACxD,YAAQ,QAAQ,oBAAoB,SAAS,eAAe;AAC5D,UAAM;AAAA,EACR;AACF;AA4DO,SAAS,mBAAmB,QAAmC;AACpE,SAAO,IAAI,aAAa,QAAQ,sBAAsB,EAAE,QAAQ,KAAK,CAAC,EAAE,eAAe,CAAC;AAC1F;","names":["createHash","mkdirSync","writeFileSync","dirname","median"]}
1
+ {"version":3,"sources":["../src/sdk.ts","../src/skill-evaluation.ts","../src/agents/evaluation.ts"],"sourcesContent":["/** Book Agent SDK — programmatic API for embedding Book as a library. */\n\nimport { join } from 'path';\nimport type { ElicitationHandler, UserQuestionHandler } from './types/tools.js';\nimport type { AgentConfig, SessionDriver } from './types/runtime.js';\nimport type { SessionStoreInterface } from './types/sessions.js';\nimport { freezeAgentConfig, loadConfig, type LoadConfigOptions } from './config.js';\nimport { runHeadless } from './headless.js';\nimport { createDefaultRegistry } from './tools/registry.js';\nimport { connectMcpServers, disconnectMcpServers } from './mcp.js';\nimport { mcpServersToRecord, partitionMcpServersByApproval } from './mcp-approvals.js';\nimport { resolveMcpServerList } from './mcp-config.js';\nimport { collectWithheldProjectNotices } from './project-approval-notices.js';\nimport { AgentManager } from './agents/manager.js';\nimport { SessionStore } from './session/store.js';\nimport { resolveSessionBootstrap } from './session/resolve.js';\nimport type { AgentEvent } from './session/agent-events.js';\nimport {\n cleanupDebugLogs,\n DEFAULT_LOCAL_DATA_RETENTION_DAYS,\n getDebugLogPath,\n} from './debug-log.js';\nimport { resolvePermissionMode } from './permission-mode.js';\nimport { resolveBookHome } from './book-home.js';\nimport { resolveSessionDriver } from './session-driver.js';\n\nexport type QueryEvent = AgentEvent;\n\nexport interface QueryOptions {\n workspace?: string;\n model?: string;\n permissionMode?: string;\n maxTurns?: number;\n settingsPath?: string;\n noSettings?: boolean;\n persistSession?: boolean;\n sessionId?: string;\n sessionStore?: SessionStoreInterface;\n signal?: AbortSignal;\n onUserQuestionRequired?: UserQuestionHandler;\n /**\n * Answer MCP form elicitations. Supplying it declares the elicitation\n * capability to connected servers; without it they fail such requests\n * themselves rather than blocking on a prompt no one can see.\n */\n onElicit?: ElicitationHandler;\n agents?: 'adaptive' | 'manual' | 'off';\n /** Forward high-volume managed-agent text deltas. Defaults to false. */\n forwardSubagentText?: boolean;\n /**\n * Who drives the session. Defaults to `BOOK_SESSION_DRIVER`, then to `agent` under another agent\n * harness and `human` otherwise; an agent-driven session saves only what it learned, as\n * project or reference memory, never its delegator's instructions.\n */\n sessionDriver?: SessionDriver;\n}\n\nclass AsyncEventQueue<T> {\n private values: T[] = [];\n private waiters: Array<(value: T | undefined) => void> = [];\n private closed = false;\n\n push(value: T): void {\n if (this.closed) return;\n const waiter = this.waiters.shift();\n if (waiter) waiter(value);\n else this.values.push(value);\n }\n\n close(): void {\n if (this.closed) return;\n this.closed = true;\n for (const waiter of this.waiters.splice(0)) waiter(undefined);\n }\n\n next(): Promise<T | undefined> {\n const value = this.values.shift();\n if (value !== undefined) return Promise.resolve(value);\n if (this.closed) return Promise.resolve(undefined);\n return new Promise((resolve) => this.waiters.push(resolve));\n }\n}\n\nexport async function* query(\n prompt: string,\n options: QueryOptions = {},\n): AsyncGenerator<QueryEvent, void, undefined> {\n const workspace = options.workspace || process.cwd();\n const loadedConfig = loadConfig(workspace, {\n settingsOverridePath: options.settingsPath,\n noSettings: options.noSettings,\n runMigrations: !options.noSettings,\n });\n const config = freezeAgentConfig({\n ...loadedConfig,\n sessionDriver: resolveSessionDriver({ explicit: options.sessionDriver, env: process.env })\n .driver,\n ...(options.model ? { model: options.model } : {}),\n ...(options.maxTurns ? { maxTurns: options.maxTurns } : {}),\n ...(options.agents\n ? {\n settings: {\n ...loadedConfig.settings,\n agents: { ...loadedConfig.settings.agents, mode: options.agents },\n },\n }\n : {}),\n });\n\n const controller = new AbortController();\n const abortFromCaller = () => controller.abort(options.signal?.reason);\n if (options.signal?.aborted) abortFromCaller();\n else options.signal?.addEventListener('abort', abortFromCaller, { once: true });\n\n const persistSession = options.persistSession !== false;\n const sessionStore = persistSession\n ? (options.sessionStore ?? new SessionStore(join(resolveBookHome(), 'sessions')))\n : undefined;\n const bootstrap = resolveSessionBootstrap(sessionStore, {\n cwd: config.workspace,\n sessionId: options.sessionId,\n driver: config.sessionDriver,\n });\n if (persistSession && !options.sessionStore && sessionStore) {\n cleanupDebugLogs(DEFAULT_LOCAL_DATA_RETENTION_DAYS, getDebugLogPath());\n sessionStore.cleanup(DEFAULT_LOCAL_DATA_RETENTION_DAYS, new Set([bootstrap.sessionId]));\n }\n const queue = new AsyncEventQueue<QueryEvent>();\n let connections: Awaited<ReturnType<typeof connectMcpServers>>['connections'] = [];\n\n queue.push({ type: 'system', model: config.model, cwd: config.workspace });\n queue.push({ type: 'session', sessionId: bootstrap.sessionId });\n\n const running = (async () => {\n try {\n // Project-declared MCP servers require a prior interactive approval;\n // SDK runs cannot prompt, so unapproved servers are skipped with a warning.\n // Repository-declared allow rules and hooks are withheld until the user\n // decides on them; an SDK run cannot ask, so it reports and continues.\n for (const notice of collectWithheldProjectNotices({\n workspace: config.workspace,\n settings: config.settings,\n settingsEnabled: !options.noSettings,\n })) {\n console.warn(notice);\n }\n const declaredMcpServers = resolveMcpServerList(config.workspace);\n const mcpPartition = partitionMcpServersByApproval(declaredMcpServers, config.settings);\n for (const server of mcpPartition.pending) {\n console.warn(\n `⚠ Skipping MCP server \"${server.name}\" (${server.path}): project-declared servers require one-time approval. Approve it in an interactive session first.`,\n );\n }\n const mcp = await connectMcpServers(config.workspace, {\n signal: controller.signal,\n servers: mcpServersToRecord(mcpPartition.allowed),\n onElicit: options.onElicit,\n });\n connections = mcp.connections;\n const registry = createDefaultRegistry({ agents: config.settings.agents.mode !== 'off' });\n if (mcp.tools.length > 0) registry.registerAll(mcp.tools);\n\n const result = await runHeadless(config, registry, {\n prompt,\n inputFormat: 'text',\n outputFormat: 'text',\n // Progress lines are for a person watching `book -p` in a terminal. An\n // SDK host owns its stderr and receives every tool call as an event.\n quiet: true,\n history: bootstrap.history,\n transcript: bootstrap.transcript,\n compactBoundaries: bootstrap.compactBoundaries,\n mode: resolvePermissionMode(config.settings, options.permissionMode),\n maxTurns: options.maxTurns,\n persistSession,\n sessionStore,\n sessionId: bootstrap.sessionId,\n sessionCreated: bootstrap.created,\n runSource: 'sdk',\n signal: controller.signal,\n onUserQuestionRequired: options.onUserQuestionRequired,\n forwardSubagentText: options.forwardSubagentText,\n stdout: { write: () => true },\n onAgentEvent: (event) => {\n if (\n event.type !== 'system' &&\n event.type !== 'session' &&\n event.type !== 'result' &&\n event.type !== 'done' &&\n (event.type !== 'agent_text_delta' || options.forwardSubagentText === true)\n ) {\n queue.push(event);\n }\n },\n });\n queue.push({\n type: 'result',\n messages: result.messages,\n usage: result.usage,\n sessionId: bootstrap.sessionId,\n outcome: result.outcome,\n runs: result.runs,\n // `stdout` above is a discard sink, so this is the only way an SDK\n // caller sees a slash command the host performed on its behalf.\n commandResults: result.commandResults,\n answer: result.answer,\n });\n } catch (error) {\n if (!controller.signal.aborted) {\n queue.push({\n type: 'error',\n error: error instanceof Error ? error.message : String(error),\n });\n }\n } finally {\n await disconnectMcpServers(connections);\n queue.push({ type: 'done' });\n queue.close();\n }\n })();\n\n try {\n while (true) {\n const event = await queue.next();\n if (!event) break;\n yield event;\n }\n } finally {\n controller.abort(new Error('SDK event consumer stopped'));\n options.signal?.removeEventListener('abort', abortFromCaller);\n await running;\n }\n}\n\nexport { loadConfig, type LoadConfigOptions };\nexport type { ToolDiscoverySettings } from './settings.js';\nexport { AgentManager };\nexport { createDefaultRegistry, createRegistry } from './tools/registry.js';\nexport { createToolSurface } from './tools/catalog.js';\nexport { toolFailure, toolSuccess } from './tools/result.js';\nexport {\n applySkillOverrides,\n buildSkillListing,\n discoverSkills,\n explicitSkillMentions,\n generateSkillListing,\n loadSkillBody,\n skillRoots,\n} from './skills.js';\nexport type {\n DiscoverSkillsOptions,\n LoadedSkillBody,\n ShadowedSkill,\n Skill,\n SkillIssue,\n SkillListingResult,\n SkillResource,\n SkillRoot,\n SkillRootKind,\n SkillSource,\n} from './skills.js';\nexport {\n SkillRegistry,\n SkillRegistryError,\n type SkillActivationFrame,\n type SkillActivationFrameSummary,\n type SkillActivationReason,\n type SkillLifecycleEvent,\n type SkillLifecycleEventType,\n type SkillRegistrySnapshot,\n} from './skill-registry.js';\nexport {\n DEFAULT_SKILL_EVALUATION_THRESHOLDS,\n DEFAULT_SKILL_EVALUATION_FIXTURES,\n SKILL_EVALUATION_CATEGORIES,\n evaluateSkillActivation,\n observeSkillEvaluation,\n renderSkillEvaluationReport,\n runSkillActivationEvaluation,\n writeSkillEvaluationReport,\n} from './skill-evaluation.js';\nexport type {\n SkillEvaluationCategory,\n SkillEvaluationFixture,\n SkillEvaluationObservation,\n SkillEvaluationOutcome,\n SkillEvaluationPromptFixture,\n SkillEvaluationReport,\n SkillEvaluationThresholds,\n} from './skill-evaluation.js';\nexport { SessionRuntime } from './session/runtime.js';\nexport { buildSkillReport } from './skill-report.js';\nexport function createAgentManager(config: AgentConfig): AgentManager {\n return new AgentManager(config, createDefaultRegistry({ agents: true }).getDefinitions());\n}\nexport { runPairedEvaluation, evaluateSuccess } from './agents/evaluation.js';\nexport { createAgentSessionSnapshot, reduceAgentSessionSnapshot } from './session/agent-events.js';\nexport type {\n AgentEvent,\n AgentSessionSnapshot,\n AgentSessionStatus,\n} from './session/agent-events.js';\nexport type {\n EvaluationFixture,\n EvaluationMetric,\n EvaluationRunResult,\n} from './agents/evaluation.js';\nexport type {\n AgentPlanRecord,\n AgentApplyResult,\n AgentRecord,\n AgentSummary,\n AgentCompletion,\n AgentCompletionNotification,\n AgentProfile,\n AgentActivity,\n AgentRuntimeEvent,\n AgentRunMetrics,\n AgentSnapshot,\n AgentSpawnRequest,\n EvidenceItem,\n} from './agents/types.js';\nexport type {\n AgentTerminalOutcome,\n AgentTerminalReason,\n AgentTerminalStatus,\n} from './types/terminal.js';\nexport type {\n AgentModelIdentity,\n AgentModelIdentityStatus,\n AgentRunAmbientSnapshot,\n AgentRunAccounting,\n AgentRunContext,\n AgentRunResult,\n AgentRunSource,\n} from './types/runs.js';\nexport type {\n JsonSchemaObject,\n ToolCatalogMetadata,\n ToolCategory,\n ToolDefinition,\n ToolDiscoveryContext,\n ToolEffect,\n ToolPolicy,\n ToolResult,\n ToolResultArtifacts,\n ToolResultError,\n ToolResultPresentation,\n ToolResultStatus,\n ToolSearchMatch,\n UserQuestion,\n UserQuestionOption,\n UserQuestionRequest,\n UserQuestionResponse,\n UserQuestionSource,\n UserQuestionHandler,\n ElicitationField,\n ElicitationRequest,\n ElicitationResponse,\n ElicitationValue,\n ElicitationHandler,\n} from './types/tools.js';\n","import { createHash } from 'node:crypto';\nimport { mkdirSync, writeFileSync } from 'node:fs';\nimport { dirname } from 'node:path';\nimport type { SkillLifecycleEvent } from './skill-registry.js';\n\nexport const SKILL_EVALUATION_CATEGORIES = [\n 'direct',\n 'indirect',\n 'negative',\n 'ambiguous',\n 'conflicting',\n 'disabled',\n 'invalid',\n 'missing-body',\n 'missing-resource',\n] as const;\n\nexport type SkillEvaluationCategory = (typeof SKILL_EVALUATION_CATEGORIES)[number];\n\nexport interface SkillEvaluationFixture {\n id: string;\n category: SkillEvaluationCategory;\n promptHash: string;\n expectedSkills: string[];\n expectedBlockedCodes?: string[];\n}\n\nexport interface SkillEvaluationPromptFixture {\n id: string;\n category: SkillEvaluationCategory;\n prompt: string;\n expectedSkills: string[];\n expectedBlockedCodes?: string[];\n}\n\nexport interface SkillEvaluationOutcome {\n promptChars?: number;\n promptTokens?: number;\n bodyTokens?: number;\n taskCompleted?: boolean;\n userCorrections?: number;\n skillToolFailures?: number;\n exposure?: 'eager' | 'deferred';\n activationTurns?: number;\n}\n\nexport interface SkillEvaluationObservation extends SkillEvaluationFixture {\n activatedSkills: string[];\n blockedCodes: string[];\n promptChars: number;\n promptTokens: number;\n bodyBytes: number;\n bodyTokens: number;\n falseActivationBytes: number;\n activationLatencyMs: number[];\n consentRequests: number;\n unnecessaryPermissionPrompts: number;\n taskCompleted?: boolean;\n userCorrections: number;\n skillToolFailures: number;\n blockingMismatch: boolean;\n exposure?: 'eager' | 'deferred';\n activationTurns?: number;\n}\n\nexport interface SkillEvaluationThresholds {\n minimumPrecision: number;\n minimumRecall: number;\n maximumFalseActivations: number;\n maximumUnnecessaryPermissionPrompts: number;\n maximumSkillToolFailures: number;\n}\n\nexport interface SkillEvaluationExecution {\n evidenceKind: 'offline-observation';\n providerRunEligibility: 'not-applicable';\n controls: {\n evaluationDate: string;\n randomSeed: string;\n runtimeRevision: string;\n fixtureRevision: string;\n fixtureRevisionStatus: 'captured' | 'incomplete';\n };\n}\n\nexport interface SkillEvaluationReport {\n generatedAt: string;\n fixtureCount: number;\n categoryCounts: Record<SkillEvaluationCategory, number>;\n truePositives: number;\n falsePositives: number;\n falseNegatives: number;\n precision: number;\n recall: number;\n falseActivationBytes: number;\n promptChars: { total: number; median: number };\n promptTokens: { total: number; median: number };\n bodyBytes: { total: number; median: number };\n bodyTokens: { total: number; median: number };\n activationLatencyMs: { median: number; p95: number };\n consentRequests: number;\n unnecessaryPermissionPrompts: number;\n taskCompletionRate?: number;\n userCorrections: number;\n skillToolFailures: number;\n blockingMismatches: number;\n exposure: Record<\n 'eager' | 'deferred',\n { samples: number; activationRate: number; medianActivationTurns: number }\n >;\n thresholds: SkillEvaluationThresholds;\n rolloutReady: boolean;\n reasons: string[];\n observations: SkillEvaluationObservation[];\n evaluation?: SkillEvaluationExecution;\n}\n\nexport const DEFAULT_SKILL_EVALUATION_THRESHOLDS: SkillEvaluationThresholds = {\n minimumPrecision: 0.95,\n minimumRecall: 0.8,\n maximumFalseActivations: 0,\n maximumUnnecessaryPermissionPrompts: 0,\n maximumSkillToolFailures: 0,\n};\n\nexport const DEFAULT_SKILL_EVALUATION_FIXTURES: readonly SkillEvaluationPromptFixture[] = [\n {\n id: 'direct-review',\n category: 'direct',\n prompt: '$review inspect the current change',\n expectedSkills: ['review'],\n },\n {\n id: 'indirect-review',\n category: 'indirect',\n prompt: 'Inspect this diff for correctness regressions and missing tests',\n expectedSkills: ['review'],\n },\n {\n id: 'negative-general-question',\n category: 'negative',\n prompt: 'Explain what this function returns',\n expectedSkills: [],\n },\n {\n id: 'ambiguous-review-word',\n category: 'ambiguous',\n prompt: 'Review the available deployment choices with me',\n expectedSkills: [],\n },\n {\n id: 'conflicting-tool-ceilings',\n category: 'conflicting',\n prompt: '$read-review $write-review inspect the change',\n expectedSkills: ['read-review'],\n expectedBlockedCodes: ['skill_tool_intersection_empty'],\n },\n {\n id: 'disabled-skill',\n category: 'disabled',\n prompt: '$disabled-skill run the workflow',\n expectedSkills: [],\n expectedBlockedCodes: ['skill_disabled'],\n },\n {\n id: 'invalid-skill',\n category: 'invalid',\n prompt: '$invalid-skill run the workflow',\n expectedSkills: [],\n expectedBlockedCodes: ['skill_invalid'],\n },\n {\n id: 'missing-body',\n category: 'missing-body',\n prompt: '$missing-body run the workflow',\n expectedSkills: [],\n expectedBlockedCodes: ['skill_load_failed'],\n },\n {\n id: 'missing-resource',\n category: 'missing-resource',\n prompt: '$missing-resource load references/checklist.md',\n expectedSkills: ['missing-resource'],\n expectedBlockedCodes: ['skill_resource_unreadable'],\n },\n];\n\nfunction numberDetail(event: SkillLifecycleEvent, key: string): number {\n const value = event.details?.[key];\n return typeof value === 'number' && Number.isFinite(value) ? value : 0;\n}\n\nfunction median(values: readonly number[]): number {\n if (values.length === 0) return 0;\n const sorted = [...values].sort((left, right) => left - right);\n return sorted[Math.floor(sorted.length / 2)] ?? 0;\n}\n\nfunction percentile(values: readonly number[], fraction: number): number {\n if (values.length === 0) return 0;\n const sorted = [...values].sort((left, right) => left - right);\n const index = Math.min(sorted.length - 1, Math.ceil(sorted.length * fraction) - 1);\n return sorted[index] ?? 0;\n}\n\n/** Convert bounded lifecycle evidence into one prompt-free evaluation observation. */\nexport function observeSkillEvaluation(\n fixture: SkillEvaluationFixture,\n events: readonly SkillLifecycleEvent[],\n outcome: SkillEvaluationOutcome = {},\n): SkillEvaluationObservation {\n const applied = events.filter(\n (event) => event.type === 'skill_activation_applied' && event.skill,\n );\n const activatedSkills = [...new Set(applied.map((event) => event.skill!))];\n const expected = new Set(fixture.expectedSkills);\n const falseActivations = applied.filter((event) => event.skill && !expected.has(event.skill));\n const consentEvents = events.filter((event) => event.type === 'skill_consent_requested');\n const blockedCodes = events\n .filter(\n (event) =>\n event.type === 'skill_activation_blocked' || event.type === 'skill_resource_blocked',\n )\n .map((event) => event.details?.code)\n .filter((code): code is string => typeof code === 'string');\n const expectedBlockedCodes = fixture.expectedBlockedCodes ?? [];\n const promptChars = outcome.promptChars ?? 0;\n const bodyBytes = applied.reduce(\n (total, event) => total + numberDetail(event, 'bodyByteSize'),\n 0,\n );\n return {\n ...fixture,\n expectedSkills: [...fixture.expectedSkills],\n activatedSkills,\n expectedBlockedCodes: [...expectedBlockedCodes],\n blockedCodes,\n promptChars,\n promptTokens: outcome.promptTokens ?? Math.ceil(promptChars / 4),\n bodyBytes,\n bodyTokens: outcome.bodyTokens ?? Math.ceil(bodyBytes / 4),\n falseActivationBytes: falseActivations.reduce(\n (total, event) => total + numberDetail(event, 'bodyByteSize'),\n 0,\n ),\n activationLatencyMs: applied.map((event) => numberDetail(event, 'activationLatencyMs')),\n consentRequests: consentEvents.length,\n unnecessaryPermissionPrompts: consentEvents.filter(\n (event) => !event.skill || !expected.has(event.skill),\n ).length,\n taskCompleted: outcome.taskCompleted,\n userCorrections: outcome.userCorrections ?? 0,\n skillToolFailures: outcome.skillToolFailures ?? 0,\n blockingMismatch:\n expectedBlockedCodes.some((code) => !blockedCodes.includes(code)) ||\n blockedCodes.some((code) => !expectedBlockedCodes.includes(code)),\n exposure: outcome.exposure,\n activationTurns: outcome.activationTurns,\n };\n}\n\nexport function evaluateSkillActivation(\n observations: readonly SkillEvaluationObservation[],\n thresholds: SkillEvaluationThresholds = DEFAULT_SKILL_EVALUATION_THRESHOLDS,\n): SkillEvaluationReport {\n const categoryCounts = Object.fromEntries(\n SKILL_EVALUATION_CATEGORIES.map((category) => [\n category,\n observations.filter((observation) => observation.category === category).length,\n ]),\n ) as Record<SkillEvaluationCategory, number>;\n let truePositives = 0;\n let falsePositives = 0;\n let falseNegatives = 0;\n for (const observation of observations) {\n const expected = new Set(observation.expectedSkills);\n const activated = new Set(observation.activatedSkills);\n truePositives += [...activated].filter((skill) => expected.has(skill)).length;\n falsePositives += [...activated].filter((skill) => !expected.has(skill)).length;\n falseNegatives += [...expected].filter((skill) => !activated.has(skill)).length;\n }\n const precision =\n truePositives + falsePositives === 0 ? 1 : truePositives / (truePositives + falsePositives);\n const recall =\n truePositives + falseNegatives === 0 ? 1 : truePositives / (truePositives + falseNegatives);\n const completed = observations.filter(\n (observation): observation is SkillEvaluationObservation & { taskCompleted: boolean } =>\n observation.taskCompleted !== undefined,\n );\n const exposure = Object.fromEntries(\n (['eager', 'deferred'] as const).map((mode) => {\n const samples = observations.filter((observation) => observation.exposure === mode);\n return [\n mode,\n {\n samples: samples.length,\n activationRate:\n samples.length === 0\n ? 0\n : samples.filter((observation) => observation.activatedSkills.length > 0).length /\n samples.length,\n medianActivationTurns: median(\n samples.map((observation) => observation.activationTurns ?? 0),\n ),\n },\n ];\n }),\n ) as SkillEvaluationReport['exposure'];\n const reasons: string[] = [];\n const missingCategories = SKILL_EVALUATION_CATEGORIES.filter(\n (category) => categoryCounts[category] === 0,\n );\n if (missingCategories.length)\n reasons.push(`Missing fixture categories: ${missingCategories.join(', ')}.`);\n if (precision < thresholds.minimumPrecision) {\n reasons.push(\n `Activation precision ${precision.toFixed(3)} is below ${thresholds.minimumPrecision}.`,\n );\n }\n if (recall < thresholds.minimumRecall) {\n reasons.push(`Activation recall ${recall.toFixed(3)} is below ${thresholds.minimumRecall}.`);\n }\n if (falsePositives > thresholds.maximumFalseActivations) {\n reasons.push(\n `False activations ${falsePositives} exceed ${thresholds.maximumFalseActivations}.`,\n );\n }\n const unnecessaryPermissionPrompts = observations.reduce(\n (total, observation) => total + observation.unnecessaryPermissionPrompts,\n 0,\n );\n if (unnecessaryPermissionPrompts > thresholds.maximumUnnecessaryPermissionPrompts) {\n reasons.push(\n `Unnecessary permission prompts ${unnecessaryPermissionPrompts} exceed ${thresholds.maximumUnnecessaryPermissionPrompts}.`,\n );\n }\n const skillToolFailures = observations.reduce(\n (total, observation) => total + observation.skillToolFailures,\n 0,\n );\n if (skillToolFailures > thresholds.maximumSkillToolFailures) {\n reasons.push(\n `Skill-caused tool failures ${skillToolFailures} exceed ${thresholds.maximumSkillToolFailures}.`,\n );\n }\n const blockingMismatches = observations.filter(\n (observation) => observation.blockingMismatch,\n ).length;\n if (blockingMismatches > 0) {\n reasons.push(`${blockingMismatches} fixture(s) had unexpected or missing block outcomes.`);\n }\n const latencies = observations.flatMap((observation) => observation.activationLatencyMs);\n return {\n generatedAt: new Date().toISOString(),\n fixtureCount: observations.length,\n categoryCounts,\n truePositives,\n falsePositives,\n falseNegatives,\n precision,\n recall,\n falseActivationBytes: observations.reduce(\n (total, observation) => total + observation.falseActivationBytes,\n 0,\n ),\n promptChars: {\n total: observations.reduce((total, observation) => total + observation.promptChars, 0),\n median: median(observations.map((observation) => observation.promptChars)),\n },\n promptTokens: {\n total: observations.reduce((total, observation) => total + observation.promptTokens, 0),\n median: median(observations.map((observation) => observation.promptTokens)),\n },\n bodyBytes: {\n total: observations.reduce((total, observation) => total + observation.bodyBytes, 0),\n median: median(observations.map((observation) => observation.bodyBytes)),\n },\n bodyTokens: {\n total: observations.reduce((total, observation) => total + observation.bodyTokens, 0),\n median: median(observations.map((observation) => observation.bodyTokens)),\n },\n activationLatencyMs: { median: median(latencies), p95: percentile(latencies, 0.95) },\n consentRequests: observations.reduce(\n (total, observation) => total + observation.consentRequests,\n 0,\n ),\n unnecessaryPermissionPrompts,\n taskCompletionRate:\n completed.length === 0\n ? undefined\n : completed.filter((observation) => observation.taskCompleted).length / completed.length,\n userCorrections: observations.reduce(\n (total, observation) => total + observation.userCorrections,\n 0,\n ),\n skillToolFailures,\n blockingMismatches,\n exposure,\n thresholds: { ...thresholds },\n rolloutReady: reasons.length === 0,\n reasons,\n observations: observations.map((observation) => ({\n ...observation,\n expectedSkills: [...observation.expectedSkills],\n expectedBlockedCodes: observation.expectedBlockedCodes\n ? [...observation.expectedBlockedCodes]\n : undefined,\n activatedSkills: [...observation.activatedSkills],\n blockedCodes: [...observation.blockedCodes],\n activationLatencyMs: [...observation.activationLatencyMs],\n })),\n };\n}\n\nexport async function runSkillActivationEvaluation(\n fixtures: readonly SkillEvaluationPromptFixture[],\n execute: (\n fixture: SkillEvaluationPromptFixture,\n ) => Promise<{ events: readonly SkillLifecycleEvent[]; outcome?: SkillEvaluationOutcome }>,\n thresholds: SkillEvaluationThresholds = DEFAULT_SKILL_EVALUATION_THRESHOLDS,\n): Promise<SkillEvaluationReport> {\n const observations: SkillEvaluationObservation[] = [];\n for (const fixture of fixtures) {\n const result = await execute(fixture);\n observations.push(\n observeSkillEvaluation(\n {\n id: fixture.id,\n category: fixture.category,\n promptHash: createHash('sha256').update(fixture.prompt).digest('hex'),\n expectedSkills: [...fixture.expectedSkills],\n expectedBlockedCodes: fixture.expectedBlockedCodes\n ? [...fixture.expectedBlockedCodes]\n : undefined,\n },\n result.events,\n result.outcome,\n ),\n );\n }\n return evaluateSkillActivation(observations, thresholds);\n}\n\nexport function renderSkillEvaluationReport(report: SkillEvaluationReport): string {\n const categories = SKILL_EVALUATION_CATEGORIES.map(\n (category) => `- ${category}: ${report.categoryCounts[category]}`,\n ).join('\\n');\n return [\n '# Skill Activation Evaluation',\n '',\n `Generated: ${report.generatedAt}`,\n `Rollout ready: ${report.rolloutReady ? 'yes' : 'no'}`,\n `Precision: ${report.precision.toFixed(3)}`,\n `Recall: ${report.recall.toFixed(3)}`,\n `False activations: ${report.falsePositives}`,\n `False-activation body bytes: ${report.falseActivationBytes}`,\n `Median prompt tokens: ${report.promptTokens.median}`,\n `Median activated body tokens: ${report.bodyTokens.median}`,\n `Median activation latency: ${report.activationLatencyMs.median} ms`,\n `Unnecessary permission prompts: ${report.unnecessaryPermissionPrompts}`,\n `Skill-caused tool failures: ${report.skillToolFailures}`,\n `Blocking mismatches: ${report.blockingMismatches}`,\n ...(report.evaluation\n ? [\n `Evidence kind: ${report.evaluation.evidenceKind}`,\n `Provider run eligibility: ${report.evaluation.providerRunEligibility}`,\n `Evaluation date: ${report.evaluation.controls.evaluationDate}`,\n `Evaluation seed: ${report.evaluation.controls.randomSeed}`,\n `Runtime revision: ${report.evaluation.controls.runtimeRevision}`,\n `Fixture revision: ${report.evaluation.controls.fixtureRevision} (${report.evaluation.controls.fixtureRevisionStatus})`,\n ]\n : []),\n '',\n '## Fixture Coverage',\n '',\n categories,\n '',\n '## Gate',\n '',\n ...(report.reasons.length\n ? report.reasons.map((reason) => `- ${reason}`)\n : ['- All thresholds passed.']),\n '',\n ].join('\\n');\n}\n\nexport function writeSkillEvaluationReport(\n report: SkillEvaluationReport,\n jsonPath: string,\n markdownPath?: string,\n): void {\n mkdirSync(dirname(jsonPath), { recursive: true });\n writeFileSync(jsonPath, `${JSON.stringify(report, null, 2)}\\n`, 'utf8');\n if (markdownPath) {\n mkdirSync(dirname(markdownPath), { recursive: true });\n writeFileSync(markdownPath, renderSkillEvaluationReport(report), 'utf8');\n }\n}\n","import { createHash } from 'crypto';\nimport { mkdirSync, writeFileSync } from 'fs';\nimport { dirname } from 'path';\nimport type { AgentMode, AgentTopology, IssueQuality } from './types.js';\n\nexport interface EvaluationFixture {\n id: string;\n cohort: 'decomposable' | 'ambiguous' | 'sequential';\n prompt: string;\n}\n\nexport interface EvaluationRunResult {\n passed: boolean;\n validationCorrect?: boolean;\n wallTimeMs: number;\n totalTokens?: number;\n costUsd?: number;\n spawnCount?: number;\n route?: AgentTopology;\n issueQuality?: IssueQuality;\n conflicts?: number;\n interruptions?: number;\n unvalidatedApplications?: number;\n}\n\nexport interface EvaluationMetric extends Omit<EvaluationRunResult, 'route' | 'issueQuality'> {\n fixtureId: string;\n promptHash: string;\n cohort: EvaluationFixture['cohort'];\n mode: Extract<AgentMode, 'adaptive' | 'off'>;\n route?: AgentTopology;\n issueQuality?: IssueQuality;\n}\n\n/** Run each fixture with the single-agent control and adaptive treatment. */\nexport async function runPairedEvaluation(\n fixtures: EvaluationFixture[],\n execute: (\n fixture: EvaluationFixture,\n mode: Extract<AgentMode, 'adaptive' | 'off'>,\n ) => Promise<EvaluationRunResult>,\n outputPath: string,\n): Promise<EvaluationMetric[]> {\n const metrics: EvaluationMetric[] = [];\n for (const fixture of fixtures) {\n for (const mode of ['off', 'adaptive'] as const) {\n const result = await execute(fixture, mode);\n metrics.push({\n fixtureId: fixture.id,\n promptHash: createHash('sha256').update(fixture.prompt).digest('hex'),\n cohort: fixture.cohort,\n mode,\n ...result,\n });\n }\n }\n mkdirSync(dirname(outputPath), { recursive: true });\n writeFileSync(\n outputPath,\n `${metrics.map((metric) => JSON.stringify(metric)).join('\\n')}\\n`,\n 'utf8',\n );\n return metrics;\n}\n\nexport function evaluateSuccess(metrics: EvaluationMetric[]): {\n successful: boolean;\n reasons: string[];\n} {\n const reasons: string[] = [];\n const byMode = (cohorts: EvaluationFixture['cohort'][], mode: EvaluationMetric['mode']) =>\n metrics.filter((metric) => cohorts.includes(metric.cohort) && metric.mode === mode);\n const rate = (items: EvaluationMetric[]) =>\n items.length === 0 ? 0 : items.filter((item) => item.passed).length / items.length;\n const median = (values: number[]) => {\n if (values.length === 0) return 0;\n const sorted = [...values].sort((left, right) => left - right);\n return sorted[Math.floor(sorted.length / 2)];\n };\n\n const treatment = byMode(['decomposable', 'ambiguous'], 'adaptive');\n const control = byMode(['decomposable', 'ambiguous'], 'off');\n const passGain = rate(treatment) - rate(control);\n const treatmentWall = median(treatment.map((item) => item.wallTimeMs));\n const controlWall = median(control.map((item) => item.wallTimeMs));\n if (passGain < 0.05 && !(passGain >= 0 && treatmentWall <= controlWall * 0.8)) {\n reasons.push('Decomposable/ambiguous cohort missed the pass-rate or wall-time target.');\n }\n\n const sequentialTreatment = byMode(['sequential'], 'adaptive');\n const sequentialControl = byMode(['sequential'], 'off');\n if (rate(sequentialTreatment) < rate(sequentialControl) - 0.02) {\n reasons.push('Sequential cohort regressed by more than two pass-rate points.');\n }\n\n const adaptiveTokens = median(\n metrics.filter((item) => item.mode === 'adaptive').map((item) => item.totalTokens ?? 0),\n );\n const controlTokens = median(\n metrics.filter((item) => item.mode === 'off').map((item) => item.totalTokens ?? 0),\n );\n if (controlTokens > 0 && adaptiveTokens >= controlTokens * 3) {\n reasons.push('Median adaptive token overhead is not below 3x.');\n }\n if (metrics.some((item) => (item.unvalidatedApplications ?? 0) > 0)) {\n reasons.push('At least one unvalidated patch was applied.');\n }\n return { successful: reasons.length === 0, reasons };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAEA,SAAS,YAAY;;;ACFrB,SAAS,kBAAkB;AAC3B,SAAS,WAAW,qBAAqB;AACzC,SAAS,eAAe;AAGjB,IAAM,8BAA8B;AAAA,EACzC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAsGO,IAAM,sCAAiE;AAAA,EAC5E,kBAAkB;AAAA,EAClB,eAAe;AAAA,EACf,yBAAyB;AAAA,EACzB,qCAAqC;AAAA,EACrC,0BAA0B;AAC5B;AAEO,IAAM,oCAA6E;AAAA,EACxF;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,QAAQ;AAAA,EAC3B;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,QAAQ;AAAA,EAC3B;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,EACnB;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,EACnB;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,aAAa;AAAA,IAC9B,sBAAsB,CAAC,+BAA+B;AAAA,EACxD;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,IACjB,sBAAsB,CAAC,gBAAgB;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,IACjB,sBAAsB,CAAC,eAAe;AAAA,EACxC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC;AAAA,IACjB,sBAAsB,CAAC,mBAAmB;AAAA,EAC5C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,gBAAgB,CAAC,kBAAkB;AAAA,IACnC,sBAAsB,CAAC,2BAA2B;AAAA,EACpD;AACF;AAEA,SAAS,aAAa,OAA4B,KAAqB;AACrE,QAAM,QAAQ,MAAM,UAAU,GAAG;AACjC,SAAO,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,IAAI,QAAQ;AACvE;AAEA,SAAS,OAAO,QAAmC;AACjD,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,MAAM,UAAU,OAAO,KAAK;AAC7D,SAAO,OAAO,KAAK,MAAM,OAAO,SAAS,CAAC,CAAC,KAAK;AAClD;AAEA,SAAS,WAAW,QAA2B,UAA0B;AACvE,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,MAAM,UAAU,OAAO,KAAK;AAC7D,QAAM,QAAQ,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,KAAK,OAAO,SAAS,QAAQ,IAAI,CAAC;AACjF,SAAO,OAAO,KAAK,KAAK;AAC1B;AAGO,SAAS,uBACd,SACA,QACA,UAAkC,CAAC,GACP;AAC5B,QAAM,UAAU,OAAO;AAAA,IACrB,CAAC,UAAU,MAAM,SAAS,8BAA8B,MAAM;AAAA,EAChE;AACA,QAAM,kBAAkB,CAAC,GAAG,IAAI,IAAI,QAAQ,IAAI,CAAC,UAAU,MAAM,KAAM,CAAC,CAAC;AACzE,QAAM,WAAW,IAAI,IAAI,QAAQ,cAAc;AAC/C,QAAM,mBAAmB,QAAQ,OAAO,CAAC,UAAU,MAAM,SAAS,CAAC,SAAS,IAAI,MAAM,KAAK,CAAC;AAC5F,QAAM,gBAAgB,OAAO,OAAO,CAAC,UAAU,MAAM,SAAS,yBAAyB;AACvF,QAAM,eAAe,OAClB;AAAA,IACC,CAAC,UACC,MAAM,SAAS,8BAA8B,MAAM,SAAS;AAAA,EAChE,EACC,IAAI,CAAC,UAAU,MAAM,SAAS,IAAI,EAClC,OAAO,CAAC,SAAyB,OAAO,SAAS,QAAQ;AAC5D,QAAM,uBAAuB,QAAQ,wBAAwB,CAAC;AAC9D,QAAM,cAAc,QAAQ,eAAe;AAC3C,QAAM,YAAY,QAAQ;AAAA,IACxB,CAAC,OAAO,UAAU,QAAQ,aAAa,OAAO,cAAc;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AAAA,IACL,GAAG;AAAA,IACH,gBAAgB,CAAC,GAAG,QAAQ,cAAc;AAAA,IAC1C;AAAA,IACA,sBAAsB,CAAC,GAAG,oBAAoB;AAAA,IAC9C;AAAA,IACA;AAAA,IACA,cAAc,QAAQ,gBAAgB,KAAK,KAAK,cAAc,CAAC;AAAA,IAC/D;AAAA,IACA,YAAY,QAAQ,cAAc,KAAK,KAAK,YAAY,CAAC;AAAA,IACzD,sBAAsB,iBAAiB;AAAA,MACrC,CAAC,OAAO,UAAU,QAAQ,aAAa,OAAO,cAAc;AAAA,MAC5D;AAAA,IACF;AAAA,IACA,qBAAqB,QAAQ,IAAI,CAAC,UAAU,aAAa,OAAO,qBAAqB,CAAC;AAAA,IACtF,iBAAiB,cAAc;AAAA,IAC/B,8BAA8B,cAAc;AAAA,MAC1C,CAAC,UAAU,CAAC,MAAM,SAAS,CAAC,SAAS,IAAI,MAAM,KAAK;AAAA,IACtD,EAAE;AAAA,IACF,eAAe,QAAQ;AAAA,IACvB,iBAAiB,QAAQ,mBAAmB;AAAA,IAC5C,mBAAmB,QAAQ,qBAAqB;AAAA,IAChD,kBACE,qBAAqB,KAAK,CAAC,SAAS,CAAC,aAAa,SAAS,IAAI,CAAC,KAChE,aAAa,KAAK,CAAC,SAAS,CAAC,qBAAqB,SAAS,IAAI,CAAC;AAAA,IAClE,UAAU,QAAQ;AAAA,IAClB,iBAAiB,QAAQ;AAAA,EAC3B;AACF;AAEO,SAAS,wBACd,cACA,aAAwC,qCACjB;AACvB,QAAM,iBAAiB,OAAO;AAAA,IAC5B,4BAA4B,IAAI,CAAC,aAAa;AAAA,MAC5C;AAAA,MACA,aAAa,OAAO,CAAC,gBAAgB,YAAY,aAAa,QAAQ,EAAE;AAAA,IAC1E,CAAC;AAAA,EACH;AACA,MAAI,gBAAgB;AACpB,MAAI,iBAAiB;AACrB,MAAI,iBAAiB;AACrB,aAAW,eAAe,cAAc;AACtC,UAAM,WAAW,IAAI,IAAI,YAAY,cAAc;AACnD,UAAM,YAAY,IAAI,IAAI,YAAY,eAAe;AACrD,qBAAiB,CAAC,GAAG,SAAS,EAAE,OAAO,CAAC,UAAU,SAAS,IAAI,KAAK,CAAC,EAAE;AACvE,sBAAkB,CAAC,GAAG,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,SAAS,IAAI,KAAK,CAAC,EAAE;AACzE,sBAAkB,CAAC,GAAG,QAAQ,EAAE,OAAO,CAAC,UAAU,CAAC,UAAU,IAAI,KAAK,CAAC,EAAE;AAAA,EAC3E;AACA,QAAM,YACJ,gBAAgB,mBAAmB,IAAI,IAAI,iBAAiB,gBAAgB;AAC9E,QAAM,SACJ,gBAAgB,mBAAmB,IAAI,IAAI,iBAAiB,gBAAgB;AAC9E,QAAM,YAAY,aAAa;AAAA,IAC7B,CAAC,gBACC,YAAY,kBAAkB;AAAA,EAClC;AACA,QAAM,WAAW,OAAO;AAAA,IACrB,CAAC,SAAS,UAAU,EAAY,IAAI,CAAC,SAAS;AAC7C,YAAM,UAAU,aAAa,OAAO,CAAC,gBAAgB,YAAY,aAAa,IAAI;AAClF,aAAO;AAAA,QACL;AAAA,QACA;AAAA,UACE,SAAS,QAAQ;AAAA,UACjB,gBACE,QAAQ,WAAW,IACf,IACA,QAAQ,OAAO,CAAC,gBAAgB,YAAY,gBAAgB,SAAS,CAAC,EAAE,SACxE,QAAQ;AAAA,UACd,uBAAuB;AAAA,YACrB,QAAQ,IAAI,CAAC,gBAAgB,YAAY,mBAAmB,CAAC;AAAA,UAC/D;AAAA,QACF;AAAA,MACF;AAAA,IACF,CAAC;AAAA,EACH;AACA,QAAM,UAAoB,CAAC;AAC3B,QAAM,oBAAoB,4BAA4B;AAAA,IACpD,CAAC,aAAa,eAAe,QAAQ,MAAM;AAAA,EAC7C;AACA,MAAI,kBAAkB;AACpB,YAAQ,KAAK,+BAA+B,kBAAkB,KAAK,IAAI,CAAC,GAAG;AAC7E,MAAI,YAAY,WAAW,kBAAkB;AAC3C,YAAQ;AAAA,MACN,wBAAwB,UAAU,QAAQ,CAAC,CAAC,aAAa,WAAW,gBAAgB;AAAA,IACtF;AAAA,EACF;AACA,MAAI,SAAS,WAAW,eAAe;AACrC,YAAQ,KAAK,qBAAqB,OAAO,QAAQ,CAAC,CAAC,aAAa,WAAW,aAAa,GAAG;AAAA,EAC7F;AACA,MAAI,iBAAiB,WAAW,yBAAyB;AACvD,YAAQ;AAAA,MACN,qBAAqB,cAAc,WAAW,WAAW,uBAAuB;AAAA,IAClF;AAAA,EACF;AACA,QAAM,+BAA+B,aAAa;AAAA,IAChD,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,IAC5C;AAAA,EACF;AACA,MAAI,+BAA+B,WAAW,qCAAqC;AACjF,YAAQ;AAAA,MACN,kCAAkC,4BAA4B,WAAW,WAAW,mCAAmC;AAAA,IACzH;AAAA,EACF;AACA,QAAM,oBAAoB,aAAa;AAAA,IACrC,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,IAC5C;AAAA,EACF;AACA,MAAI,oBAAoB,WAAW,0BAA0B;AAC3D,YAAQ;AAAA,MACN,8BAA8B,iBAAiB,WAAW,WAAW,wBAAwB;AAAA,IAC/F;AAAA,EACF;AACA,QAAM,qBAAqB,aAAa;AAAA,IACtC,CAAC,gBAAgB,YAAY;AAAA,EAC/B,EAAE;AACF,MAAI,qBAAqB,GAAG;AAC1B,YAAQ,KAAK,GAAG,kBAAkB,uDAAuD;AAAA,EAC3F;AACA,QAAM,YAAY,aAAa,QAAQ,CAAC,gBAAgB,YAAY,mBAAmB;AACvF,SAAO;AAAA,IACL,cAAa,oBAAI,KAAK,GAAE,YAAY;AAAA,IACpC,cAAc,aAAa;AAAA,IAC3B;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,sBAAsB,aAAa;AAAA,MACjC,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,MAC5C;AAAA,IACF;AAAA,IACA,aAAa;AAAA,MACX,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,aAAa,CAAC;AAAA,MACrF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,WAAW,CAAC;AAAA,IAC3E;AAAA,IACA,cAAc;AAAA,MACZ,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,cAAc,CAAC;AAAA,MACtF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,YAAY,CAAC;AAAA,IAC5E;AAAA,IACA,WAAW;AAAA,MACT,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,WAAW,CAAC;AAAA,MACnF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,SAAS,CAAC;AAAA,IACzE;AAAA,IACA,YAAY;AAAA,MACV,OAAO,aAAa,OAAO,CAAC,OAAO,gBAAgB,QAAQ,YAAY,YAAY,CAAC;AAAA,MACpF,QAAQ,OAAO,aAAa,IAAI,CAAC,gBAAgB,YAAY,UAAU,CAAC;AAAA,IAC1E;AAAA,IACA,qBAAqB,EAAE,QAAQ,OAAO,SAAS,GAAG,KAAK,WAAW,WAAW,IAAI,EAAE;AAAA,IACnF,iBAAiB,aAAa;AAAA,MAC5B,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,MAC5C;AAAA,IACF;AAAA,IACA;AAAA,IACA,oBACE,UAAU,WAAW,IACjB,SACA,UAAU,OAAO,CAAC,gBAAgB,YAAY,aAAa,EAAE,SAAS,UAAU;AAAA,IACtF,iBAAiB,aAAa;AAAA,MAC5B,CAAC,OAAO,gBAAgB,QAAQ,YAAY;AAAA,MAC5C;AAAA,IACF;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,YAAY,EAAE,GAAG,WAAW;AAAA,IAC5B,cAAc,QAAQ,WAAW;AAAA,IACjC;AAAA,IACA,cAAc,aAAa,IAAI,CAAC,iBAAiB;AAAA,MAC/C,GAAG;AAAA,MACH,gBAAgB,CAAC,GAAG,YAAY,cAAc;AAAA,MAC9C,sBAAsB,YAAY,uBAC9B,CAAC,GAAG,YAAY,oBAAoB,IACpC;AAAA,MACJ,iBAAiB,CAAC,GAAG,YAAY,eAAe;AAAA,MAChD,cAAc,CAAC,GAAG,YAAY,YAAY;AAAA,MAC1C,qBAAqB,CAAC,GAAG,YAAY,mBAAmB;AAAA,IAC1D,EAAE;AAAA,EACJ;AACF;AAEA,eAAsB,6BACpB,UACA,SAGA,aAAwC,qCACR;AAChC,QAAM,eAA6C,CAAC;AACpD,aAAW,WAAW,UAAU;AAC9B,UAAM,SAAS,MAAM,QAAQ,OAAO;AACpC,iBAAa;AAAA,MACX;AAAA,QACE;AAAA,UACE,IAAI,QAAQ;AAAA,UACZ,UAAU,QAAQ;AAAA,UAClB,YAAY,WAAW,QAAQ,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,KAAK;AAAA,UACpE,gBAAgB,CAAC,GAAG,QAAQ,cAAc;AAAA,UAC1C,sBAAsB,QAAQ,uBAC1B,CAAC,GAAG,QAAQ,oBAAoB,IAChC;AAAA,QACN;AAAA,QACA,OAAO;AAAA,QACP,OAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AACA,SAAO,wBAAwB,cAAc,UAAU;AACzD;AAEO,SAAS,4BAA4B,QAAuC;AACjF,QAAM,aAAa,4BAA4B;AAAA,IAC7C,CAAC,aAAa,KAAK,QAAQ,KAAK,OAAO,eAAe,QAAQ,CAAC;AAAA,EACjE,EAAE,KAAK,IAAI;AACX,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,cAAc,OAAO,WAAW;AAAA,IAChC,kBAAkB,OAAO,eAAe,QAAQ,IAAI;AAAA,IACpD,cAAc,OAAO,UAAU,QAAQ,CAAC,CAAC;AAAA,IACzC,WAAW,OAAO,OAAO,QAAQ,CAAC,CAAC;AAAA,IACnC,sBAAsB,OAAO,cAAc;AAAA,IAC3C,gCAAgC,OAAO,oBAAoB;AAAA,IAC3D,yBAAyB,OAAO,aAAa,MAAM;AAAA,IACnD,iCAAiC,OAAO,WAAW,MAAM;AAAA,IACzD,8BAA8B,OAAO,oBAAoB,MAAM;AAAA,IAC/D,mCAAmC,OAAO,4BAA4B;AAAA,IACtE,+BAA+B,OAAO,iBAAiB;AAAA,IACvD,wBAAwB,OAAO,kBAAkB;AAAA,IACjD,GAAI,OAAO,aACP;AAAA,MACE,kBAAkB,OAAO,WAAW,YAAY;AAAA,MAChD,6BAA6B,OAAO,WAAW,sBAAsB;AAAA,MACrE,oBAAoB,OAAO,WAAW,SAAS,cAAc;AAAA,MAC7D,oBAAoB,OAAO,WAAW,SAAS,UAAU;AAAA,MACzD,qBAAqB,OAAO,WAAW,SAAS,eAAe;AAAA,MAC/D,qBAAqB,OAAO,WAAW,SAAS,eAAe,KAAK,OAAO,WAAW,SAAS,qBAAqB;AAAA,IACtH,IACA,CAAC;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAI,OAAO,QAAQ,SACf,OAAO,QAAQ,IAAI,CAAC,WAAW,KAAK,MAAM,EAAE,IAC5C,CAAC,0BAA0B;AAAA,IAC/B;AAAA,EACF,EAAE,KAAK,IAAI;AACb;AAEO,SAAS,2BACd,QACA,UACA,cACM;AACN,YAAU,QAAQ,QAAQ,GAAG,EAAE,WAAW,KAAK,CAAC;AAChD,gBAAc,UAAU,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,GAAM,MAAM;AACtE,MAAI,cAAc;AAChB,cAAU,QAAQ,YAAY,GAAG,EAAE,WAAW,KAAK,CAAC;AACpD,kBAAc,cAAc,4BAA4B,MAAM,GAAG,MAAM;AAAA,EACzE;AACF;;;ACjfA,SAAS,cAAAA,mBAAkB;AAC3B,SAAS,aAAAC,YAAW,iBAAAC,sBAAqB;AACzC,SAAS,WAAAC,gBAAe;AAiCxB,eAAsB,oBACpB,UACA,SAIA,YAC6B;AAC7B,QAAM,UAA8B,CAAC;AACrC,aAAW,WAAW,UAAU;AAC9B,eAAW,QAAQ,CAAC,OAAO,UAAU,GAAY;AAC/C,YAAM,SAAS,MAAM,QAAQ,SAAS,IAAI;AAC1C,cAAQ,KAAK;AAAA,QACX,WAAW,QAAQ;AAAA,QACnB,YAAYH,YAAW,QAAQ,EAAE,OAAO,QAAQ,MAAM,EAAE,OAAO,KAAK;AAAA,QACpE,QAAQ,QAAQ;AAAA,QAChB;AAAA,QACA,GAAG;AAAA,MACL,CAAC;AAAA,IACH;AAAA,EACF;AACA,EAAAC,WAAUE,SAAQ,UAAU,GAAG,EAAE,WAAW,KAAK,CAAC;AAClD,EAAAD;AAAA,IACE;AAAA,IACA,GAAG,QAAQ,IAAI,CAAC,WAAW,KAAK,UAAU,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC;AAAA;AAAA,IAC7D;AAAA,EACF;AACA,SAAO;AACT;AAEO,SAAS,gBAAgB,SAG9B;AACA,QAAM,UAAoB,CAAC;AAC3B,QAAM,SAAS,CAAC,SAAwC,SACtD,QAAQ,OAAO,CAAC,WAAW,QAAQ,SAAS,OAAO,MAAM,KAAK,OAAO,SAAS,IAAI;AACpF,QAAM,OAAO,CAAC,UACZ,MAAM,WAAW,IAAI,IAAI,MAAM,OAAO,CAAC,SAAS,KAAK,MAAM,EAAE,SAAS,MAAM;AAC9E,QAAME,UAAS,CAAC,WAAqB;AACnC,QAAI,OAAO,WAAW,EAAG,QAAO;AAChC,UAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,MAAM,UAAU,OAAO,KAAK;AAC7D,WAAO,OAAO,KAAK,MAAM,OAAO,SAAS,CAAC,CAAC;AAAA,EAC7C;AAEA,QAAM,YAAY,OAAO,CAAC,gBAAgB,WAAW,GAAG,UAAU;AAClE,QAAM,UAAU,OAAO,CAAC,gBAAgB,WAAW,GAAG,KAAK;AAC3D,QAAM,WAAW,KAAK,SAAS,IAAI,KAAK,OAAO;AAC/C,QAAM,gBAAgBA,QAAO,UAAU,IAAI,CAAC,SAAS,KAAK,UAAU,CAAC;AACrE,QAAM,cAAcA,QAAO,QAAQ,IAAI,CAAC,SAAS,KAAK,UAAU,CAAC;AACjE,MAAI,WAAW,QAAQ,EAAE,YAAY,KAAK,iBAAiB,cAAc,MAAM;AAC7E,YAAQ,KAAK,yEAAyE;AAAA,EACxF;AAEA,QAAM,sBAAsB,OAAO,CAAC,YAAY,GAAG,UAAU;AAC7D,QAAM,oBAAoB,OAAO,CAAC,YAAY,GAAG,KAAK;AACtD,MAAI,KAAK,mBAAmB,IAAI,KAAK,iBAAiB,IAAI,MAAM;AAC9D,YAAQ,KAAK,gEAAgE;AAAA,EAC/E;AAEA,QAAM,iBAAiBA;AAAA,IACrB,QAAQ,OAAO,CAAC,SAAS,KAAK,SAAS,UAAU,EAAE,IAAI,CAAC,SAAS,KAAK,eAAe,CAAC;AAAA,EACxF;AACA,QAAM,gBAAgBA;AAAA,IACpB,QAAQ,OAAO,CAAC,SAAS,KAAK,SAAS,KAAK,EAAE,IAAI,CAAC,SAAS,KAAK,eAAe,CAAC;AAAA,EACnF;AACA,MAAI,gBAAgB,KAAK,kBAAkB,gBAAgB,GAAG;AAC5D,YAAQ,KAAK,iDAAiD;AAAA,EAChE;AACA,MAAI,QAAQ,KAAK,CAAC,UAAU,KAAK,2BAA2B,KAAK,CAAC,GAAG;AACnE,YAAQ,KAAK,6CAA6C;AAAA,EAC5D;AACA,SAAO,EAAE,YAAY,QAAQ,WAAW,GAAG,QAAQ;AACrD;;;AFnDA,IAAM,kBAAN,MAAyB;AAAA,EACf,SAAc,CAAC;AAAA,EACf,UAAiD,CAAC;AAAA,EAClD,SAAS;AAAA,EAEjB,KAAK,OAAgB;AACnB,QAAI,KAAK,OAAQ;AACjB,UAAM,SAAS,KAAK,QAAQ,MAAM;AAClC,QAAI,OAAQ,QAAO,KAAK;AAAA,QACnB,MAAK,OAAO,KAAK,KAAK;AAAA,EAC7B;AAAA,EAEA,QAAc;AACZ,QAAI,KAAK,OAAQ;AACjB,SAAK,SAAS;AACd,eAAW,UAAU,KAAK,QAAQ,OAAO,CAAC,EAAG,QAAO,MAAS;AAAA,EAC/D;AAAA,EAEA,OAA+B;AAC7B,UAAM,QAAQ,KAAK,OAAO,MAAM;AAChC,QAAI,UAAU,OAAW,QAAO,QAAQ,QAAQ,KAAK;AACrD,QAAI,KAAK,OAAQ,QAAO,QAAQ,QAAQ,MAAS;AACjD,WAAO,IAAI,QAAQ,CAAC,YAAY,KAAK,QAAQ,KAAK,OAAO,CAAC;AAAA,EAC5D;AACF;AAEA,gBAAuB,MACrB,QACA,UAAwB,CAAC,GACoB;AAC7C,QAAM,YAAY,QAAQ,aAAa,QAAQ,IAAI;AACnD,QAAM,eAAe,WAAW,WAAW;AAAA,IACzC,sBAAsB,QAAQ;AAAA,IAC9B,YAAY,QAAQ;AAAA,IACpB,eAAe,CAAC,QAAQ;AAAA,EAC1B,CAAC;AACD,QAAM,SAAS,kBAAkB;AAAA,IAC/B,GAAG;AAAA,IACH,eAAe,qBAAqB,EAAE,UAAU,QAAQ,eAAe,KAAK,QAAQ,IAAI,CAAC,EACtF;AAAA,IACH,GAAI,QAAQ,QAAQ,EAAE,OAAO,QAAQ,MAAM,IAAI,CAAC;AAAA,IAChD,GAAI,QAAQ,WAAW,EAAE,UAAU,QAAQ,SAAS,IAAI,CAAC;AAAA,IACzD,GAAI,QAAQ,SACR;AAAA,MACE,UAAU;AAAA,QACR,GAAG,aAAa;AAAA,QAChB,QAAQ,EAAE,GAAG,aAAa,SAAS,QAAQ,MAAM,QAAQ,OAAO;AAAA,MAClE;AAAA,IACF,IACA,CAAC;AAAA,EACP,CAAC;AAED,QAAM,aAAa,IAAI,gBAAgB;AACvC,QAAM,kBAAkB,MAAM,WAAW,MAAM,QAAQ,QAAQ,MAAM;AACrE,MAAI,QAAQ,QAAQ,QAAS,iBAAgB;AAAA,MACxC,SAAQ,QAAQ,iBAAiB,SAAS,iBAAiB,EAAE,MAAM,KAAK,CAAC;AAE9E,QAAM,iBAAiB,QAAQ,mBAAmB;AAClD,QAAM,eAAe,iBAChB,QAAQ,gBAAgB,IAAI,aAAa,KAAK,gBAAgB,GAAG,UAAU,CAAC,IAC7E;AACJ,QAAM,YAAY,wBAAwB,cAAc;AAAA,IACtD,KAAK,OAAO;AAAA,IACZ,WAAW,QAAQ;AAAA,IACnB,QAAQ,OAAO;AAAA,EACjB,CAAC;AACD,MAAI,kBAAkB,CAAC,QAAQ,gBAAgB,cAAc;AAC3D,qBAAiB,mCAAmC,gBAAgB,CAAC;AACrE,iBAAa,QAAQ,mCAAmC,oBAAI,IAAI,CAAC,UAAU,SAAS,CAAC,CAAC;AAAA,EACxF;AACA,QAAM,QAAQ,IAAI,gBAA4B;AAC9C,MAAI,cAA4E,CAAC;AAEjF,QAAM,KAAK,EAAE,MAAM,UAAU,OAAO,OAAO,OAAO,KAAK,OAAO,UAAU,CAAC;AACzE,QAAM,KAAK,EAAE,MAAM,WAAW,WAAW,UAAU,UAAU,CAAC;AAE9D,QAAM,WAAW,YAAY;AAC3B,QAAI;AAKF,iBAAW,UAAU,8BAA8B;AAAA,QACjD,WAAW,OAAO;AAAA,QAClB,UAAU,OAAO;AAAA,QACjB,iBAAiB,CAAC,QAAQ;AAAA,MAC5B,CAAC,GAAG;AACF,gBAAQ,KAAK,MAAM;AAAA,MACrB;AACA,YAAM,qBAAqB,qBAAqB,OAAO,SAAS;AAChE,YAAM,eAAe,8BAA8B,oBAAoB,OAAO,QAAQ;AACtF,iBAAW,UAAU,aAAa,SAAS;AACzC,gBAAQ;AAAA,UACN,gCAA2B,OAAO,IAAI,MAAM,OAAO,IAAI;AAAA,QACzD;AAAA,MACF;AACA,YAAM,MAAM,MAAM,kBAAkB,OAAO,WAAW;AAAA,QACpD,QAAQ,WAAW;AAAA,QACnB,SAAS,mBAAmB,aAAa,OAAO;AAAA,QAChD,UAAU,QAAQ;AAAA,MACpB,CAAC;AACD,oBAAc,IAAI;AAClB,YAAM,WAAW,sBAAsB,EAAE,QAAQ,OAAO,SAAS,OAAO,SAAS,MAAM,CAAC;AACxF,UAAI,IAAI,MAAM,SAAS,EAAG,UAAS,YAAY,IAAI,KAAK;AAExD,YAAM,SAAS,MAAM,YAAY,QAAQ,UAAU;AAAA,QACjD;AAAA,QACA,aAAa;AAAA,QACb,cAAc;AAAA;AAAA;AAAA,QAGd,OAAO;AAAA,QACP,SAAS,UAAU;AAAA,QACnB,YAAY,UAAU;AAAA,QACtB,mBAAmB,UAAU;AAAA,QAC7B,MAAM,sBAAsB,OAAO,UAAU,QAAQ,cAAc;AAAA,QACnE,UAAU,QAAQ;AAAA,QAClB;AAAA,QACA;AAAA,QACA,WAAW,UAAU;AAAA,QACrB,gBAAgB,UAAU;AAAA,QAC1B,WAAW;AAAA,QACX,QAAQ,WAAW;AAAA,QACnB,wBAAwB,QAAQ;AAAA,QAChC,qBAAqB,QAAQ;AAAA,QAC7B,QAAQ,EAAE,OAAO,MAAM,KAAK;AAAA,QAC5B,cAAc,CAAC,UAAU;AACvB,cACE,MAAM,SAAS,YACf,MAAM,SAAS,aACf,MAAM,SAAS,YACf,MAAM,SAAS,WACd,MAAM,SAAS,sBAAsB,QAAQ,wBAAwB,OACtE;AACA,kBAAM,KAAK,KAAK;AAAA,UAClB;AAAA,QACF;AAAA,MACF,CAAC;AACD,YAAM,KAAK;AAAA,QACT,MAAM;AAAA,QACN,UAAU,OAAO;AAAA,QACjB,OAAO,OAAO;AAAA,QACd,WAAW,UAAU;AAAA,QACrB,SAAS,OAAO;AAAA,QAChB,MAAM,OAAO;AAAA;AAAA;AAAA,QAGb,gBAAgB,OAAO;AAAA,QACvB,QAAQ,OAAO;AAAA,MACjB,CAAC;AAAA,IACH,SAAS,OAAO;AACd,UAAI,CAAC,WAAW,OAAO,SAAS;AAC9B,cAAM,KAAK;AAAA,UACT,MAAM;AAAA,UACN,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;AAAA,QAC9D,CAAC;AAAA,MACH;AAAA,IACF,UAAE;AACA,YAAM,qBAAqB,WAAW;AACtC,YAAM,KAAK,EAAE,MAAM,OAAO,CAAC;AAC3B,YAAM,MAAM;AAAA,IACd;AAAA,EACF,GAAG;AAEH,MAAI;AACF,WAAO,MAAM;AACX,YAAM,QAAQ,MAAM,MAAM,KAAK;AAC/B,UAAI,CAAC,MAAO;AACZ,YAAM;AAAA,IACR;AAAA,EACF,UAAE;AACA,eAAW,MAAM,IAAI,MAAM,4BAA4B,CAAC;AACxD,YAAQ,QAAQ,oBAAoB,SAAS,eAAe;AAC5D,UAAM;AAAA,EACR;AACF;AA4DO,SAAS,mBAAmB,QAAmC;AACpE,SAAO,IAAI,aAAa,QAAQ,sBAAsB,EAAE,QAAQ,KAAK,CAAC,EAAE,eAAe,CAAC;AAC1F;","names":["createHash","mkdirSync","writeFileSync","dirname","median"]}
@@ -0,0 +1,35 @@
1
+ import {
2
+ applySettingsEnvOverrides,
3
+ formatIgnoredWorkspaceSandboxKey,
4
+ ignoredWorkspaceSandboxKeys,
5
+ loadSettingsFile,
6
+ mergeSettings,
7
+ migrateLegacyPermissions,
8
+ resolveSettings,
9
+ settingsLayerPaths,
10
+ startupAnimationEnvNote,
11
+ workspaceLayerThatEnabledSandbox
12
+ } from "./chunk-W3XFHEF4.js";
13
+ import "./chunk-525CAZPY.js";
14
+ import "./chunk-X66I2NQO.js";
15
+ import "./chunk-62QL5QUP.js";
16
+ import "./chunk-224FJZLH.js";
17
+ import "./chunk-KBQS35OV.js";
18
+ import "./chunk-SOW4WRQP.js";
19
+ import "./chunk-VPNLCMVR.js";
20
+ import "./chunk-X4NJT436.js";
21
+ import "./chunk-ZPSZZUWF.js";
22
+ import "./chunk-5RDRAO4B.js";
23
+ export {
24
+ applySettingsEnvOverrides,
25
+ formatIgnoredWorkspaceSandboxKey,
26
+ ignoredWorkspaceSandboxKeys,
27
+ loadSettingsFile,
28
+ mergeSettings,
29
+ migrateLegacyPermissions,
30
+ resolveSettings,
31
+ settingsLayerPaths,
32
+ startupAnimationEnvNote,
33
+ workspaceLayerThatEnabledSandbox
34
+ };
35
+ //# sourceMappingURL=settings-loader-JUNPGAAM.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}
@@ -5,7 +5,7 @@ import {
5
5
  resolveShell,
6
6
  shellExecution,
7
7
  shellPromptLine
8
- } from "./chunk-VZQXX3WW.js";
8
+ } from "./chunk-QYBD22H5.js";
9
9
  export {
10
10
  decodePowerShellCommand,
11
11
  describeShell,
@@ -14,4 +14,4 @@ export {
14
14
  shellExecution,
15
15
  shellPromptLine
16
16
  };
17
- //# sourceMappingURL=shell-selection-V6QAYXPX.js.map
17
+ //# sourceMappingURL=shell-selection-CDHGJ4GC.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@letrquan/book",
3
- "version": "0.2.0",
3
+ "version": "0.3.0",
4
4
  "description": "AI coding agent CLI with rich TUI",
5
5
  "type": "module",
6
6
  "license": "PolyForm-Small-Business-1.0.0",
@@ -24,18 +24,14 @@
24
24
  "agent"
25
25
  ],
26
26
  "engines": {
27
- "node": ">=22.13.0"
27
+ "node": ">=22.19.0"
28
28
  },
29
29
  "packageManager": "npm@11.16.0",
30
30
  "files": [
31
31
  "dist",
32
32
  "README.md",
33
33
  "CHANGELOG.md",
34
- "LICENSE",
35
- "patches/ink+6.8.0.patch",
36
- "scripts/apply-ink-patch.mjs",
37
- "scripts/ink-patch.mjs",
38
- "scripts/verify-ink-patch.mjs"
34
+ "LICENSE"
39
35
  ],
40
36
  "publishConfig": {
41
37
  "access": "public"
@@ -58,8 +54,7 @@
58
54
  "scripts": {
59
55
  "build": "tsup",
60
56
  "dev": "tsx src/index.ts",
61
- "postinstall": "node scripts/apply-ink-patch.mjs",
62
- "typecheck": "tsc --noEmit",
57
+ "typecheck": "tsc --noEmit && tsc -p tsconfig.scripts.json",
63
58
  "pretest": "npm run build",
64
59
  "test": "npm run test:unit && npm run test:contract && npm run test:integration",
65
60
  "test:unit": "vitest run --config vitest.unit.config.ts",
@@ -68,16 +63,18 @@
68
63
  "test:watch": "vitest",
69
64
  "test:coverage": "vitest run --coverage --config vitest.coverage.config.ts",
70
65
  "bench:ui": "node scripts/run-ui-benchmark.mjs",
71
- "verify:ink-patch": "node scripts/verify-ink-patch.mjs",
66
+ "verify:ink": "tsx scripts/verify-ink.ts",
72
67
  "bench:runtime": "tsx src/tui/__benchmarks__/runtime.bench.ts",
73
68
  "eval:edit": "tsx scripts/edit-eval.ts",
74
69
  "eval:compact": "tsx scripts/compact-eval.ts",
70
+ "eval:compact-replay": "tsx scripts/compact-replay.ts",
71
+ "eval:memory": "tsx scripts/memory-eval.ts",
75
72
  "eval:skills": "tsx scripts/skill-eval.ts",
76
73
  "eval:review": "tsx scripts/review-eval.ts",
77
- "apply:ink-patch": "node scripts/apply-ink-patch.mjs",
74
+ "eval:prompt": "tsx scripts/prompt-eval.ts",
78
75
  "lint": "eslint src/ --ext .ts,.tsx --max-warnings 0",
79
- "format": "prettier --write \"src/**/*.{ts,tsx}\"",
80
- "format:check": "prettier --check \"{src,scripts}/**/*.{ts,tsx}\" \"*.config.ts\"",
76
+ "format": "prettier --write \"{src,scripts}/**/*.{ts,tsx}\" \"*.config.ts\" \"*.md\" \"docs/**/*.md\"",
77
+ "format:check": "prettier --check \"{src,scripts}/**/*.{ts,tsx}\" \"*.config.ts\" \"*.md\" \"docs/**/*.md\"",
81
78
  "architecture:check": "tsx scripts/check-architecture.ts",
82
79
  "deadcode:check": "knip",
83
80
  "deadcode:report": "knip --reporter markdown --no-exit-code",
@@ -94,26 +91,28 @@
94
91
  "@modelcontextprotocol/sdk": "^1.30.0",
95
92
  "commander": "15.0.0",
96
93
  "cross-spawn": "^7.0.6",
97
- "fast-glob": "^3.3.0",
98
94
  "fuse.js": "^7.5.0",
99
95
  "htmlparser2": "^12.0.0",
100
96
  "ignore": "^7.0.6",
101
- "ink": "6.8.0",
97
+ "ink": "7.1.1",
102
98
  "ink-text-input": "^6.0.0",
103
99
  "marked": "^18.0.9",
100
+ "picomatch": "^4.0.7",
104
101
  "react": "^19.2.8",
102
+ "tinyglobby": "^0.2.17",
105
103
  "turndown": "^7.2.4",
106
- "undici": "^6.28.0",
107
- "zod": "3.25.76"
104
+ "undici": "^8.11.2",
105
+ "zod": "4.6.5"
108
106
  },
109
107
  "devDependencies": {
110
108
  "@eslint/js": "^10.0.1",
111
109
  "@types/cross-spawn": "^6.0.6",
112
110
  "@types/node": "^26.2.0",
111
+ "@types/picomatch": "^4.0.3",
113
112
  "@types/react": "^19.2.18",
114
113
  "@types/turndown": "^5.0.6",
115
- "@typescript-eslint/eslint-plugin": "8.67.0",
116
- "@typescript-eslint/parser": "8.67.0",
114
+ "@typescript-eslint/eslint-plugin": "8.70.1",
115
+ "@typescript-eslint/parser": "8.70.1",
117
116
  "@vitest/coverage-v8": "^4.1.9",
118
117
  "@xterm/headless": "^6.0.0",
119
118
  "eslint": "^10.8.1",
@@ -122,18 +121,20 @@
122
121
  "ink-testing-library": "^4.0.0",
123
122
  "knip": "^6.32.2",
124
123
  "node-pty": "^1.1.0",
125
- "patch-package": "^8.0.1",
126
124
  "prettier": "^3.9.6",
125
+ "sharp": "^0.35.4",
127
126
  "tinybench": "^6.1.3",
128
127
  "tsup": "^8.5.0",
129
128
  "tsx": "^4.23.12",
130
129
  "typescript": "5.9.3",
131
- "vitest": "^4.1.9"
130
+ "vitest": "^4.1.11"
132
131
  },
133
132
  "overrides": {
134
133
  "postcss": "8.5.26",
134
+ "sharp": "$sharp",
135
135
  "adm-zip": "^0.6.0",
136
- "fast-uri": "^3.1.7"
136
+ "fast-uri": "^3.1.7",
137
+ "esbuild": "0.28.1"
137
138
  },
138
139
  "allowScripts": {
139
140
  "esbuild@0.27.7": true,