@itaila/archetype 0.3.30

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 (319) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +475 -0
  3. package/dist/audit/audit-persona.d.ts +163 -0
  4. package/dist/audit/audit-persona.d.ts.map +1 -0
  5. package/dist/audit/audit-persona.js +415 -0
  6. package/dist/audit/audit-persona.js.map +1 -0
  7. package/dist/audit/brain-reflection.d.ts +33 -0
  8. package/dist/audit/brain-reflection.d.ts.map +1 -0
  9. package/dist/audit/brain-reflection.js +148 -0
  10. package/dist/audit/brain-reflection.js.map +1 -0
  11. package/dist/audit/conversation-audit.d.ts +12 -0
  12. package/dist/audit/conversation-audit.d.ts.map +1 -0
  13. package/dist/audit/conversation-audit.js +76 -0
  14. package/dist/audit/conversation-audit.js.map +1 -0
  15. package/dist/audit/prompt-audit.d.ts +10 -0
  16. package/dist/audit/prompt-audit.d.ts.map +1 -0
  17. package/dist/audit/prompt-audit.js +153 -0
  18. package/dist/audit/prompt-audit.js.map +1 -0
  19. package/dist/audit/prompt-dump.d.ts +137 -0
  20. package/dist/audit/prompt-dump.d.ts.map +1 -0
  21. package/dist/audit/prompt-dump.js +269 -0
  22. package/dist/audit/prompt-dump.js.map +1 -0
  23. package/dist/audit/trace-integrity.d.ts +33 -0
  24. package/dist/audit/trace-integrity.d.ts.map +1 -0
  25. package/dist/audit/trace-integrity.js +109 -0
  26. package/dist/audit/trace-integrity.js.map +1 -0
  27. package/dist/audit/types.d.ts +92 -0
  28. package/dist/audit/types.d.ts.map +1 -0
  29. package/dist/audit/types.js +2 -0
  30. package/dist/audit/types.js.map +1 -0
  31. package/dist/audit/version.d.ts +14 -0
  32. package/dist/audit/version.d.ts.map +1 -0
  33. package/dist/audit/version.js +65 -0
  34. package/dist/audit/version.js.map +1 -0
  35. package/dist/brain.d.ts +7 -0
  36. package/dist/brain.d.ts.map +1 -0
  37. package/dist/brain.js +83 -0
  38. package/dist/brain.js.map +1 -0
  39. package/dist/builder/actions.d.ts +60 -0
  40. package/dist/builder/actions.d.ts.map +1 -0
  41. package/dist/builder/actions.js +257 -0
  42. package/dist/builder/actions.js.map +1 -0
  43. package/dist/builder/browser.d.ts +140 -0
  44. package/dist/builder/browser.d.ts.map +1 -0
  45. package/dist/builder/browser.js +232 -0
  46. package/dist/builder/browser.js.map +1 -0
  47. package/dist/builder/executor.d.ts +228 -0
  48. package/dist/builder/executor.d.ts.map +1 -0
  49. package/dist/builder/executor.js +1548 -0
  50. package/dist/builder/executor.js.map +1 -0
  51. package/dist/builder/index.d.ts +24 -0
  52. package/dist/builder/index.d.ts.map +1 -0
  53. package/dist/builder/index.js +24 -0
  54. package/dist/builder/index.js.map +1 -0
  55. package/dist/builder/node-test-discovery.d.ts +13 -0
  56. package/dist/builder/node-test-discovery.d.ts.map +1 -0
  57. package/dist/builder/node-test-discovery.js +45 -0
  58. package/dist/builder/node-test-discovery.js.map +1 -0
  59. package/dist/builder/sandbox.d.ts +172 -0
  60. package/dist/builder/sandbox.d.ts.map +1 -0
  61. package/dist/builder/sandbox.js +294 -0
  62. package/dist/builder/sandbox.js.map +1 -0
  63. package/dist/builder/workspace-files.d.ts +63 -0
  64. package/dist/builder/workspace-files.d.ts.map +1 -0
  65. package/dist/builder/workspace-files.js +190 -0
  66. package/dist/builder/workspace-files.js.map +1 -0
  67. package/dist/core/actions.d.ts +55 -0
  68. package/dist/core/actions.d.ts.map +1 -0
  69. package/dist/core/actions.js +311 -0
  70. package/dist/core/actions.js.map +1 -0
  71. package/dist/core/attachment-notes.d.ts +7 -0
  72. package/dist/core/attachment-notes.d.ts.map +1 -0
  73. package/dist/core/attachment-notes.js +38 -0
  74. package/dist/core/attachment-notes.js.map +1 -0
  75. package/dist/core/context.d.ts +10 -0
  76. package/dist/core/context.d.ts.map +1 -0
  77. package/dist/core/context.js +108 -0
  78. package/dist/core/context.js.map +1 -0
  79. package/dist/core/crud-prompt.d.ts +16 -0
  80. package/dist/core/crud-prompt.d.ts.map +1 -0
  81. package/dist/core/crud-prompt.js +268 -0
  82. package/dist/core/crud-prompt.js.map +1 -0
  83. package/dist/core/crud-schema.d.ts +12 -0
  84. package/dist/core/crud-schema.d.ts.map +1 -0
  85. package/dist/core/crud-schema.js +42 -0
  86. package/dist/core/crud-schema.js.map +1 -0
  87. package/dist/core/effective-config.d.ts +13 -0
  88. package/dist/core/effective-config.d.ts.map +1 -0
  89. package/dist/core/effective-config.js +33 -0
  90. package/dist/core/effective-config.js.map +1 -0
  91. package/dist/core/entities.d.ts +82 -0
  92. package/dist/core/entities.d.ts.map +1 -0
  93. package/dist/core/entities.js +116 -0
  94. package/dist/core/entities.js.map +1 -0
  95. package/dist/core/entity-helpers.d.ts +47 -0
  96. package/dist/core/entity-helpers.d.ts.map +1 -0
  97. package/dist/core/entity-helpers.js +122 -0
  98. package/dist/core/entity-helpers.js.map +1 -0
  99. package/dist/core/entity-registry.d.ts +47 -0
  100. package/dist/core/entity-registry.d.ts.map +1 -0
  101. package/dist/core/entity-registry.js +54 -0
  102. package/dist/core/entity-registry.js.map +1 -0
  103. package/dist/core/eq.d.ts +13 -0
  104. package/dist/core/eq.d.ts.map +1 -0
  105. package/dist/core/eq.js +41 -0
  106. package/dist/core/eq.js.map +1 -0
  107. package/dist/core/focus-context.d.ts +19 -0
  108. package/dist/core/focus-context.d.ts.map +1 -0
  109. package/dist/core/focus-context.js +46 -0
  110. package/dist/core/focus-context.js.map +1 -0
  111. package/dist/core/focus-mode-actions.d.ts +23 -0
  112. package/dist/core/focus-mode-actions.d.ts.map +1 -0
  113. package/dist/core/focus-mode-actions.js +74 -0
  114. package/dist/core/focus-mode-actions.js.map +1 -0
  115. package/dist/core/greeting.d.ts +10 -0
  116. package/dist/core/greeting.d.ts.map +1 -0
  117. package/dist/core/greeting.js +41 -0
  118. package/dist/core/greeting.js.map +1 -0
  119. package/dist/core/identity.d.ts +13 -0
  120. package/dist/core/identity.d.ts.map +1 -0
  121. package/dist/core/identity.js +54 -0
  122. package/dist/core/identity.js.map +1 -0
  123. package/dist/core/knowledge.d.ts +10 -0
  124. package/dist/core/knowledge.d.ts.map +1 -0
  125. package/dist/core/knowledge.js +40 -0
  126. package/dist/core/knowledge.js.map +1 -0
  127. package/dist/core/memory-actions.d.ts +38 -0
  128. package/dist/core/memory-actions.d.ts.map +1 -0
  129. package/dist/core/memory-actions.js +181 -0
  130. package/dist/core/memory-actions.js.map +1 -0
  131. package/dist/core/memory.d.ts +35 -0
  132. package/dist/core/memory.d.ts.map +1 -0
  133. package/dist/core/memory.js +168 -0
  134. package/dist/core/memory.js.map +1 -0
  135. package/dist/core/peer-actions.d.ts +15 -0
  136. package/dist/core/peer-actions.d.ts.map +1 -0
  137. package/dist/core/peer-actions.js +33 -0
  138. package/dist/core/peer-actions.js.map +1 -0
  139. package/dist/core/prompt-builder.d.ts +46 -0
  140. package/dist/core/prompt-builder.d.ts.map +1 -0
  141. package/dist/core/prompt-builder.js +543 -0
  142. package/dist/core/prompt-builder.js.map +1 -0
  143. package/dist/core/prompt-mode.d.ts +3 -0
  144. package/dist/core/prompt-mode.d.ts.map +1 -0
  145. package/dist/core/prompt-mode.js +6 -0
  146. package/dist/core/prompt-mode.js.map +1 -0
  147. package/dist/core/prompted-turn.d.ts +6 -0
  148. package/dist/core/prompted-turn.d.ts.map +1 -0
  149. package/dist/core/prompted-turn.js +48 -0
  150. package/dist/core/prompted-turn.js.map +1 -0
  151. package/dist/core/request-builder.d.ts +14 -0
  152. package/dist/core/request-builder.d.ts.map +1 -0
  153. package/dist/core/request-builder.js +64 -0
  154. package/dist/core/request-builder.js.map +1 -0
  155. package/dist/core/session-routing.d.ts +23 -0
  156. package/dist/core/session-routing.d.ts.map +1 -0
  157. package/dist/core/session-routing.js +59 -0
  158. package/dist/core/session-routing.js.map +1 -0
  159. package/dist/core/voice.d.ts +6 -0
  160. package/dist/core/voice.d.ts.map +1 -0
  161. package/dist/core/voice.js +30 -0
  162. package/dist/core/voice.js.map +1 -0
  163. package/dist/engine/chat.d.ts +45 -0
  164. package/dist/engine/chat.d.ts.map +1 -0
  165. package/dist/engine/chat.js +308 -0
  166. package/dist/engine/chat.js.map +1 -0
  167. package/dist/engine/continuity.d.ts +107 -0
  168. package/dist/engine/continuity.d.ts.map +1 -0
  169. package/dist/engine/continuity.js +320 -0
  170. package/dist/engine/continuity.js.map +1 -0
  171. package/dist/engine/crud.d.ts +62 -0
  172. package/dist/engine/crud.d.ts.map +1 -0
  173. package/dist/engine/crud.js +260 -0
  174. package/dist/engine/crud.js.map +1 -0
  175. package/dist/engine/side-effects.d.ts +93 -0
  176. package/dist/engine/side-effects.d.ts.map +1 -0
  177. package/dist/engine/side-effects.js +271 -0
  178. package/dist/engine/side-effects.js.map +1 -0
  179. package/dist/engine/staging.d.ts +29 -0
  180. package/dist/engine/staging.d.ts.map +1 -0
  181. package/dist/engine/staging.js +159 -0
  182. package/dist/engine/staging.js.map +1 -0
  183. package/dist/engine/working-set.d.ts +18 -0
  184. package/dist/engine/working-set.d.ts.map +1 -0
  185. package/dist/engine/working-set.js +246 -0
  186. package/dist/engine/working-set.js.map +1 -0
  187. package/dist/evals/action-contracts.d.ts +40 -0
  188. package/dist/evals/action-contracts.d.ts.map +1 -0
  189. package/dist/evals/action-contracts.js +208 -0
  190. package/dist/evals/action-contracts.js.map +1 -0
  191. package/dist/evals/brain-bloat.d.ts +39 -0
  192. package/dist/evals/brain-bloat.d.ts.map +1 -0
  193. package/dist/evals/brain-bloat.js +167 -0
  194. package/dist/evals/brain-bloat.js.map +1 -0
  195. package/dist/evals/brain-prescriptions.d.ts +30 -0
  196. package/dist/evals/brain-prescriptions.d.ts.map +1 -0
  197. package/dist/evals/brain-prescriptions.js +148 -0
  198. package/dist/evals/brain-prescriptions.js.map +1 -0
  199. package/dist/evals/cross-layer-duplicates.d.ts +49 -0
  200. package/dist/evals/cross-layer-duplicates.d.ts.map +1 -0
  201. package/dist/evals/cross-layer-duplicates.js +289 -0
  202. package/dist/evals/cross-layer-duplicates.js.map +1 -0
  203. package/dist/evals/entity-visibility.d.ts +28 -0
  204. package/dist/evals/entity-visibility.d.ts.map +1 -0
  205. package/dist/evals/entity-visibility.js +216 -0
  206. package/dist/evals/entity-visibility.js.map +1 -0
  207. package/dist/evals/index.d.ts +19 -0
  208. package/dist/evals/index.d.ts.map +1 -0
  209. package/dist/evals/index.js +11 -0
  210. package/dist/evals/index.js.map +1 -0
  211. package/dist/evals/judge.d.ts +22 -0
  212. package/dist/evals/judge.d.ts.map +1 -0
  213. package/dist/evals/judge.js +337 -0
  214. package/dist/evals/judge.js.map +1 -0
  215. package/dist/evals/operational-contract.d.ts +40 -0
  216. package/dist/evals/operational-contract.d.ts.map +1 -0
  217. package/dist/evals/operational-contract.js +115 -0
  218. package/dist/evals/operational-contract.js.map +1 -0
  219. package/dist/evals/prompt-content.d.ts +14 -0
  220. package/dist/evals/prompt-content.d.ts.map +1 -0
  221. package/dist/evals/prompt-content.js +104 -0
  222. package/dist/evals/prompt-content.js.map +1 -0
  223. package/dist/evals/runtime.d.ts +4 -0
  224. package/dist/evals/runtime.d.ts.map +1 -0
  225. package/dist/evals/runtime.js +197 -0
  226. package/dist/evals/runtime.js.map +1 -0
  227. package/dist/evals/sample-projects.d.ts +143 -0
  228. package/dist/evals/sample-projects.d.ts.map +1 -0
  229. package/dist/evals/sample-projects.js +644 -0
  230. package/dist/evals/sample-projects.js.map +1 -0
  231. package/dist/evals/types.d.ts +88 -0
  232. package/dist/evals/types.d.ts.map +1 -0
  233. package/dist/evals/types.js +2 -0
  234. package/dist/evals/types.js.map +1 -0
  235. package/dist/foundation/index.d.ts +158 -0
  236. package/dist/foundation/index.d.ts.map +1 -0
  237. package/dist/foundation/index.js +256 -0
  238. package/dist/foundation/index.js.map +1 -0
  239. package/dist/index.d.ts +223 -0
  240. package/dist/index.d.ts.map +1 -0
  241. package/dist/index.js +998 -0
  242. package/dist/index.js.map +1 -0
  243. package/dist/managed/autonomous-loop.d.ts +199 -0
  244. package/dist/managed/autonomous-loop.d.ts.map +1 -0
  245. package/dist/managed/autonomous-loop.js +451 -0
  246. package/dist/managed/autonomous-loop.js.map +1 -0
  247. package/dist/managed/conversation.d.ts +20 -0
  248. package/dist/managed/conversation.d.ts.map +1 -0
  249. package/dist/managed/conversation.js +40 -0
  250. package/dist/managed/conversation.js.map +1 -0
  251. package/dist/managed/knowledge.d.ts +7 -0
  252. package/dist/managed/knowledge.d.ts.map +1 -0
  253. package/dist/managed/knowledge.js +174 -0
  254. package/dist/managed/knowledge.js.map +1 -0
  255. package/dist/managed/memory-manager.d.ts +7 -0
  256. package/dist/managed/memory-manager.d.ts.map +1 -0
  257. package/dist/managed/memory-manager.js +18 -0
  258. package/dist/managed/memory-manager.js.map +1 -0
  259. package/dist/managed/memory-review.d.ts +45 -0
  260. package/dist/managed/memory-review.d.ts.map +1 -0
  261. package/dist/managed/memory-review.js +130 -0
  262. package/dist/managed/memory-review.js.map +1 -0
  263. package/dist/managed/storage.d.ts +2 -0
  264. package/dist/managed/storage.d.ts.map +1 -0
  265. package/dist/managed/storage.js +2 -0
  266. package/dist/managed/storage.js.map +1 -0
  267. package/dist/managed/work-history.d.ts +23 -0
  268. package/dist/managed/work-history.d.ts.map +1 -0
  269. package/dist/managed/work-history.js +31 -0
  270. package/dist/managed/work-history.js.map +1 -0
  271. package/dist/observability/index.d.ts +15 -0
  272. package/dist/observability/index.d.ts.map +1 -0
  273. package/dist/observability/index.js +15 -0
  274. package/dist/observability/index.js.map +1 -0
  275. package/dist/observability/render-run-markdown.d.ts +90 -0
  276. package/dist/observability/render-run-markdown.d.ts.map +1 -0
  277. package/dist/observability/render-run-markdown.js +231 -0
  278. package/dist/observability/render-run-markdown.js.map +1 -0
  279. package/dist/observability/turn-reporter.d.ts +20 -0
  280. package/dist/observability/turn-reporter.d.ts.map +1 -0
  281. package/dist/observability/turn-reporter.js +106 -0
  282. package/dist/observability/turn-reporter.js.map +1 -0
  283. package/dist/persona.d.ts +49 -0
  284. package/dist/persona.d.ts.map +1 -0
  285. package/dist/persona.js +287 -0
  286. package/dist/persona.js.map +1 -0
  287. package/dist/playbook/defaults.d.ts +25 -0
  288. package/dist/playbook/defaults.d.ts.map +1 -0
  289. package/dist/playbook/defaults.js +108 -0
  290. package/dist/playbook/defaults.js.map +1 -0
  291. package/dist/playbook/invariants.d.ts +244 -0
  292. package/dist/playbook/invariants.d.ts.map +1 -0
  293. package/dist/playbook/invariants.js +259 -0
  294. package/dist/playbook/invariants.js.map +1 -0
  295. package/dist/playbook/templates.d.ts +7 -0
  296. package/dist/playbook/templates.d.ts.map +1 -0
  297. package/dist/playbook/templates.js +437 -0
  298. package/dist/playbook/templates.js.map +1 -0
  299. package/dist/providers/gemini.d.ts +73 -0
  300. package/dist/providers/gemini.d.ts.map +1 -0
  301. package/dist/providers/gemini.js +536 -0
  302. package/dist/providers/gemini.js.map +1 -0
  303. package/dist/providers/types.d.ts +2 -0
  304. package/dist/providers/types.d.ts.map +1 -0
  305. package/dist/providers/types.js +2 -0
  306. package/dist/providers/types.js.map +1 -0
  307. package/dist/providers/zod-to-gemini.d.ts +8 -0
  308. package/dist/providers/zod-to-gemini.d.ts.map +1 -0
  309. package/dist/providers/zod-to-gemini.js +148 -0
  310. package/dist/providers/zod-to-gemini.js.map +1 -0
  311. package/dist/samples/pm-spec-agent.d.ts +22 -0
  312. package/dist/samples/pm-spec-agent.d.ts.map +1 -0
  313. package/dist/samples/pm-spec-agent.js +53 -0
  314. package/dist/samples/pm-spec-agent.js.map +1 -0
  315. package/dist/types.d.ts +920 -0
  316. package/dist/types.d.ts.map +1 -0
  317. package/dist/types.js +2 -0
  318. package/dist/types.js.map +1 -0
  319. package/package.json +68 -0
@@ -0,0 +1,104 @@
1
+ const BANNED_DRIFT_MARKERS = [
2
+ 'Recovery yogurt bowl',
3
+ 'A lighter yogurt bowl makes sense here.',
4
+ '"entity":"meal"',
5
+ '"entity": "meal"',
6
+ 'meal_2',
7
+ ];
8
+ // Markers that reference a specific entity name. When that entity is
9
+ // declared in the persona's config, the marker is legitimate domain
10
+ // vocabulary, not cross-domain drift.
11
+ const MARKER_ENTITY_NAMES = {
12
+ 'Recovery yogurt bowl': 'meal',
13
+ 'A lighter yogurt bowl makes sense here.': 'meal',
14
+ '"entity":"meal"': 'meal',
15
+ '"entity": "meal"': 'meal',
16
+ 'meal_2': 'meal',
17
+ };
18
+ const BOXING_SMELL_MARKERS = [
19
+ 'Don\'t pad, don\'t truncate.',
20
+ 'Before each response, silently review the full conversation.',
21
+ 'Frequency Rule:',
22
+ "you're not a dashboard reading it out.",
23
+ ];
24
+ const DUPLICATE_SCAFFOLD_MARKERS = [
25
+ 'single most impactful thing',
26
+ 'come back and think with me again',
27
+ 'trusted thinking partner',
28
+ ];
29
+ export function auditPromptContent(input) {
30
+ const issues = [];
31
+ const prompt = input.prompt;
32
+ const declaredEntities = new Set(input.declaredEntities ?? []);
33
+ for (const marker of BANNED_DRIFT_MARKERS) {
34
+ if (!prompt.includes(marker))
35
+ continue;
36
+ // Skip markers whose associated entity is legitimately declared on
37
+ // this persona — the vocabulary is native, not drift.
38
+ const entityName = MARKER_ENTITY_NAMES[marker];
39
+ if (entityName && declaredEntities.has(entityName))
40
+ continue;
41
+ issues.push({
42
+ severity: 'error',
43
+ message: `Prompt contains banned cross-domain example drift: ${marker}`,
44
+ });
45
+ }
46
+ for (const marker of BOXING_SMELL_MARKERS) {
47
+ if (prompt.includes(marker)) {
48
+ issues.push({
49
+ severity: 'warn',
50
+ message: `Prompt contains likely boxing or instruction-theater language: ${marker}`,
51
+ });
52
+ }
53
+ }
54
+ for (const marker of DUPLICATE_SCAFFOLD_MARKERS) {
55
+ const occurrences = prompt.toLowerCase().split(marker.toLowerCase()).length - 1;
56
+ if (occurrences > 1) {
57
+ issues.push({
58
+ severity: 'warn',
59
+ message: `Prompt repeats scaffold phrase "${marker}" ${occurrences} times`,
60
+ });
61
+ }
62
+ }
63
+ if (prompt.includes('--- FILES ---') && prompt.includes('--- SPEC BUNDLE ---')) {
64
+ issues.push({
65
+ severity: 'warn',
66
+ message: 'Prompt splits file context across FILES and SPEC BUNDLE; use one FILES surface with virtual roots such as spec/... and artifact/....',
67
+ });
68
+ }
69
+ for (const marker of ['spec/spec/', 'artifact/artifact/']) {
70
+ if (prompt.includes(marker)) {
71
+ issues.push({
72
+ severity: 'error',
73
+ message: `Prompt contains nested virtual workspace mount path "${marker}"`,
74
+ });
75
+ }
76
+ }
77
+ if (prompt.includes('default for unprefixed file actions')) {
78
+ issues.push({
79
+ severity: 'error',
80
+ message: 'Prompt exposes a hidden default workspace path alias; mounted workspaces must use one canonical visible path per file.',
81
+ });
82
+ }
83
+ if (prompt.includes('- editFile:') && prompt.includes('How the tools behave here:') && prompt.includes("editFile: each entry's oldText")) {
84
+ issues.push({
85
+ severity: 'warn',
86
+ message: 'Prompt documents editFile in both the action list and tool-behavior section; keep the contract in one place.',
87
+ });
88
+ }
89
+ const exampleEntityMatches = [...prompt.matchAll(/example response item:\s*\{.*?"entity":"([^"]+)"/g)];
90
+ for (const match of exampleEntityMatches) {
91
+ const entityName = match[1];
92
+ if (declaredEntities.size > 0 && !declaredEntities.has(entityName)) {
93
+ issues.push({
94
+ severity: 'error',
95
+ message: `Prompt example references undeclared entity "${entityName}"`,
96
+ });
97
+ }
98
+ }
99
+ return {
100
+ pass: issues.every(issue => issue.severity !== 'error'),
101
+ issues,
102
+ };
103
+ }
104
+ //# sourceMappingURL=prompt-content.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"prompt-content.js","sourceRoot":"","sources":["../../src/evals/prompt-content.ts"],"names":[],"mappings":"AAeA,MAAM,oBAAoB,GAAG;IAC3B,sBAAsB;IACtB,yCAAyC;IACzC,iBAAiB;IACjB,kBAAkB;IAClB,QAAQ;CACT,CAAA;AAED,qEAAqE;AACrE,oEAAoE;AACpE,sCAAsC;AACtC,MAAM,mBAAmB,GAA2B;IAClD,sBAAsB,EAAE,MAAM;IAC9B,yCAAyC,EAAE,MAAM;IACjD,iBAAiB,EAAE,MAAM;IACzB,kBAAkB,EAAE,MAAM;IAC1B,QAAQ,EAAE,MAAM;CACjB,CAAA;AAED,MAAM,oBAAoB,GAAG;IAC3B,8BAA8B;IAC9B,8DAA8D;IAC9D,iBAAiB;IACjB,wCAAwC;CACzC,CAAA;AAED,MAAM,0BAA0B,GAAG;IACjC,6BAA6B;IAC7B,mCAAmC;IACnC,0BAA0B;CAClB,CAAA;AAEV,MAAM,UAAU,kBAAkB,CAAC,KAA8B;IAC/D,MAAM,MAAM,GAA8B,EAAE,CAAA;IAC5C,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAA;IAC3B,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,gBAAgB,IAAI,EAAE,CAAC,CAAA;IAE9D,KAAK,MAAM,MAAM,IAAI,oBAAoB,EAAE,CAAC;QAC1C,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,SAAQ;QACtC,mEAAmE;QACnE,sDAAsD;QACtD,MAAM,UAAU,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAA;QAC9C,IAAI,UAAU,IAAI,gBAAgB,CAAC,GAAG,CAAC,UAAU,CAAC;YAAE,SAAQ;QAC5D,MAAM,CAAC,IAAI,CAAC;YACV,QAAQ,EAAE,OAAO;YACjB,OAAO,EAAE,sDAAsD,MAAM,EAAE;SACxE,CAAC,CAAA;IACJ,CAAC;IAED,KAAK,MAAM,MAAM,IAAI,oBAAoB,EAAE,CAAC;QAC1C,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;YAC5B,MAAM,CAAC,IAAI,CAAC;gBACV,QAAQ,EAAE,MAAM;gBAChB,OAAO,EAAE,kEAAkE,MAAM,EAAE;aACpF,CAAC,CAAA;QACJ,CAAC;IACH,CAAC;IAED,KAAK,MAAM,MAAM,IAAI,0BAA0B,EAAE,CAAC;QAChD,MAAM,WAAW,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,MAAM,GAAG,CAAC,CAAA;QAC/E,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC;gBACV,QAAQ,EAAE,MAAM;gBAChB,OAAO,EAAE,mCAAmC,MAAM,KAAK,WAAW,QAAQ;aAC3E,CAAC,CAAA;QACJ,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,qBAAqB,CAAC,EAAE,CAAC;QAC/E,MAAM,CAAC,IAAI,CAAC;YACV,QAAQ,EAAE,MAAM;YAChB,OAAO,EAAE,sIAAsI;SAChJ,CAAC,CAAA;IACJ,CAAC;IAED,KAAK,MAAM,MAAM,IAAI,CAAC,YAAY,EAAE,oBAAoB,CAAC,EAAE,CAAC;QAC1D,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;YAC5B,MAAM,CAAC,IAAI,CAAC;gBACV,QAAQ,EAAE,OAAO;gBACjB,OAAO,EAAE,wDAAwD,MAAM,GAAG;aAC3E,CAAC,CAAA;QACJ,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,QAAQ,CAAC,qCAAqC,CAAC,EAAE,CAAC;QAC3D,MAAM,CAAC,IAAI,CAAC;YACV,QAAQ,EAAE,OAAO;YACjB,OAAO,EAAE,wHAAwH;SAClI,CAAC,CAAA;IACJ,CAAC;IAED,IAAI,MAAM,CAAC,QAAQ,CAAC,aAAa,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,4BAA4B,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,gCAAgC,CAAC,EAAE,CAAC;QACzI,MAAM,CAAC,IAAI,CAAC;YACV,QAAQ,EAAE,MAAM;YAChB,OAAO,EAAE,8GAA8G;SACxH,CAAC,CAAA;IACJ,CAAC;IAED,MAAM,oBAAoB,GAAG,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,mDAAmD,CAAC,CAAC,CAAA;IACtG,KAAK,MAAM,KAAK,IAAI,oBAAoB,EAAE,CAAC;QACzC,MAAM,UAAU,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;QAC3B,IAAI,gBAAgB,CAAC,IAAI,GAAG,CAAC,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,CAAC;YACnE,MAAM,CAAC,IAAI,CAAC;gBACV,QAAQ,EAAE,OAAO;gBACjB,OAAO,EAAE,gDAAgD,UAAU,GAAG;aACvE,CAAC,CAAA;QACJ,CAAC;IACH,CAAC;IAED,OAAO;QACL,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,OAAO,CAAC;QACvD,MAAM;KACP,CAAA;AACH,CAAC"}
@@ -0,0 +1,4 @@
1
+ import type { LLMProvider } from '../types.js';
2
+ import type { EvalConversationResult, EvalProject, EvalState, EvalTurn } from './types.js';
3
+ export declare function runEvalConversation<State extends EvalState>(project: EvalProject<State>, provider: LLMProvider, turns: EvalTurn[]): Promise<EvalConversationResult<State>>;
4
+ //# sourceMappingURL=runtime.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"runtime.d.ts","sourceRoot":"","sources":["../../src/evals/runtime.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,WAAW,EAAsB,MAAM,aAAa,CAAA;AASlE,OAAO,KAAK,EAEV,sBAAsB,EACtB,WAAW,EACX,SAAS,EACT,QAAQ,EAET,MAAM,YAAY,CAAA;AAEnB,wBAAsB,mBAAmB,CAAC,KAAK,SAAS,SAAS,EAC/D,OAAO,EAAE,WAAW,CAAC,KAAK,CAAC,EAC3B,QAAQ,EAAE,WAAW,EACrB,KAAK,EAAE,QAAQ,EAAE,GAChB,OAAO,CAAC,sBAAsB,CAAC,KAAK,CAAC,CAAC,CA6GxC"}
@@ -0,0 +1,197 @@
1
+ import { PersonaEngine } from '../persona.js';
2
+ import { resolveActions, resolveEntities } from '../core/effective-config.js';
3
+ import { executeSideEffects, getExecutedAnnotations, getProposedActions, } from '../engine/side-effects.js';
4
+ import { buildAssistantContinuityMessage } from '../engine/continuity.js';
5
+ import { crudActionToAnnotation, resolveTempIds } from '../engine/crud.js';
6
+ export async function runEvalConversation(project, provider, turns) {
7
+ const engine = new PersonaEngine({ ...project.persona, provider });
8
+ const state = project.initialState();
9
+ const history = [];
10
+ const results = [];
11
+ let memoryCounter = 1;
12
+ for (const turn of turns) {
13
+ const stateBefore = project.summarizeState(state);
14
+ const chatResult = await engine.chat({
15
+ message: turn.userMessage,
16
+ history,
17
+ context: project.buildContext(state),
18
+ memories: state.memories,
19
+ timezone: project.timezone,
20
+ directives: turn.directives,
21
+ extraSystemSections: turn.extraSystemSections,
22
+ userIdentity: project.userIdentity,
23
+ });
24
+ const actionDefs = resolveActions(engine.config) ?? {};
25
+ const handlers = project.buildHandlers(state);
26
+ const sideEffectResults = chatResult.actions.length > 0
27
+ ? await executeSideEffects(chatResult.actions, handlers, actionDefs, {
28
+ approval: engine.config.approval,
29
+ })
30
+ : [];
31
+ // Execute CRUD actions through entity-specific handlers (including memory entities)
32
+ if (chatResult.crudActions && chatResult.crudActions.length > 0) {
33
+ // Resolve temp IDs (_w1 → UUID) so handlers see real IDs with cross-references resolved
34
+ const resolvedCrud = resolveTempIds(chatResult.crudActions);
35
+ const appCrudHandlers = project.buildCrudHandlers ? project.buildCrudHandlers(state) : {};
36
+ const memoryCrudHandlers = buildMemoryCrudHandlers(state, () => `mem-auto-${memoryCounter++}`);
37
+ const allCrudHandlers = { ...memoryCrudHandlers, ...appCrudHandlers };
38
+ for (const crudAction of resolvedCrud) {
39
+ const handler = allCrudHandlers[crudAction.entity];
40
+ if (handler) {
41
+ await handler(crudAction);
42
+ }
43
+ }
44
+ }
45
+ const proposedActions = getProposedActions(sideEffectResults);
46
+ const actionRecords = sideEffectResults.map(result => ({
47
+ action: result.action,
48
+ annotation: result.annotation,
49
+ proposed: result.status === 'proposed',
50
+ success: result.success,
51
+ error: result.error,
52
+ }));
53
+ history.push({ role: 'user', content: turn.userMessage });
54
+ // Store assistant continuity through the same invariant used by managed
55
+ // chat and autonomous loops: narrative + outcome notes are model-visible;
56
+ // raw action annotations are debug/display-only and stripped before future
57
+ // turns.
58
+ const actionAnnotations = getExecutedAnnotations(sideEffectResults);
59
+ const actionOutcomes = sideEffectResults.map(result => ({
60
+ action: result.action,
61
+ status: result.status,
62
+ success: result.success,
63
+ error: result.error,
64
+ annotation: result.annotation,
65
+ }));
66
+ // Append CRUD action annotations (use effective entities to include memory entities)
67
+ const crudAnnotations = [];
68
+ if (chatResult.crudActions && chatResult.crudActions.length > 0) {
69
+ const effectiveEntities = resolveEntities(engine.config) ?? engine.config.entities ?? {};
70
+ crudAnnotations.push(...chatResult.crudActions.map(a => crudActionToAnnotation(a, effectiveEntities)));
71
+ }
72
+ const storedAssistantMessage = buildAssistantContinuityMessage({
73
+ message: chatResult.message,
74
+ modelOutcomeNotes: chatResult.outcomeNotes,
75
+ actionOutcomes: [
76
+ ...actionOutcomes,
77
+ ...crudAnnotations.map(annotation => ({ outcomeNote: `${annotation} executed.` })),
78
+ ],
79
+ actionAnnotations: [...actionAnnotations, ...crudAnnotations],
80
+ });
81
+ history.push({ role: 'assistant', content: storedAssistantMessage });
82
+ results.push({
83
+ userMessage: turn.userMessage,
84
+ assistantMessage: chatResult.message,
85
+ storedAssistantMessage,
86
+ raw: chatResult.raw,
87
+ actions: chatResult.actions,
88
+ actionRecords,
89
+ proposedActions,
90
+ followUps: chatResult.followUps ?? [],
91
+ outcomeNotes: chatResult.outcomeNotes,
92
+ stateBefore,
93
+ stateAfter: project.summarizeState(state),
94
+ stateSnapshot: cloneState(state),
95
+ sideEffectResults,
96
+ trace: chatResult.trace,
97
+ });
98
+ }
99
+ return {
100
+ project,
101
+ provider,
102
+ turns: results,
103
+ finalState: state,
104
+ history,
105
+ };
106
+ }
107
+ /**
108
+ * CRUD handlers for memory and craftMemory entities in evals.
109
+ */
110
+ function buildMemoryCrudHandlers(state, nextId) {
111
+ return {
112
+ memory: async (action) => {
113
+ const params = action.params ?? {};
114
+ if (action.operation === 'create') {
115
+ state.memories.push({
116
+ id: nextId(),
117
+ content: String(params.content),
118
+ category: String(params.category),
119
+ source: params.source === 'user' || params.source === 'inferred' || params.source === 'suggested'
120
+ ? params.source
121
+ : undefined,
122
+ stability: params.stability === 'durable' || params.stability === 'tentative' || params.stability === 'temporary'
123
+ ? params.stability
124
+ : undefined,
125
+ contextHint: typeof params.contextHint === 'string' ? params.contextHint : undefined,
126
+ pinned: false,
127
+ createdAt: new Date().toISOString(),
128
+ });
129
+ }
130
+ else if (action.operation === 'update' && action.id) {
131
+ const memory = state.memories.find(item => item.id === action.id);
132
+ if (!memory)
133
+ return;
134
+ if (params.content != null)
135
+ memory.content = String(params.content);
136
+ if (params.category != null)
137
+ memory.category = String(params.category);
138
+ if (params.source === 'user' || params.source === 'inferred' || params.source === 'suggested')
139
+ memory.source = params.source;
140
+ if (params.stability === 'durable' || params.stability === 'tentative' || params.stability === 'temporary')
141
+ memory.stability = params.stability;
142
+ if (typeof params.contextHint === 'string')
143
+ memory.contextHint = params.contextHint;
144
+ }
145
+ else if (action.operation === 'delete' && action.id) {
146
+ const index = state.memories.findIndex(item => item.id === action.id);
147
+ if (index !== -1)
148
+ state.memories.splice(index, 1);
149
+ }
150
+ },
151
+ craftMemory: async (action) => {
152
+ // Craft memories also live in state.memories with scope: 'craft'
153
+ const params = action.params ?? {};
154
+ if (action.operation === 'create') {
155
+ state.memories.push({
156
+ id: nextId(),
157
+ content: String(params.content),
158
+ category: String(params.category),
159
+ scope: 'craft',
160
+ source: params.source === 'user' || params.source === 'inferred' || params.source === 'suggested'
161
+ ? params.source
162
+ : undefined,
163
+ stability: params.stability === 'durable' || params.stability === 'tentative' || params.stability === 'temporary'
164
+ ? params.stability
165
+ : undefined,
166
+ contextHint: typeof params.contextHint === 'string' ? params.contextHint : undefined,
167
+ pinned: false,
168
+ createdAt: new Date().toISOString(),
169
+ });
170
+ }
171
+ else if (action.operation === 'update' && action.id) {
172
+ const memory = state.memories.find(item => item.id === action.id);
173
+ if (!memory)
174
+ return;
175
+ if (params.content != null)
176
+ memory.content = String(params.content);
177
+ if (params.category != null)
178
+ memory.category = String(params.category);
179
+ if (params.source === 'user' || params.source === 'inferred' || params.source === 'suggested')
180
+ memory.source = params.source;
181
+ if (params.stability === 'durable' || params.stability === 'tentative' || params.stability === 'temporary')
182
+ memory.stability = params.stability;
183
+ if (typeof params.contextHint === 'string')
184
+ memory.contextHint = params.contextHint;
185
+ }
186
+ else if (action.operation === 'delete' && action.id) {
187
+ const index = state.memories.findIndex(item => item.id === action.id);
188
+ if (index !== -1)
189
+ state.memories.splice(index, 1);
190
+ }
191
+ },
192
+ };
193
+ }
194
+ function cloneState(state) {
195
+ return JSON.parse(JSON.stringify(state));
196
+ }
197
+ //# sourceMappingURL=runtime.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"runtime.js","sourceRoot":"","sources":["../../src/evals/runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAA;AAE7C,OAAO,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,6BAA6B,CAAA;AAC7E,OAAO,EACL,kBAAkB,EAClB,sBAAsB,EACtB,kBAAkB,GACnB,MAAM,2BAA2B,CAAA;AAClC,OAAO,EAAE,+BAA+B,EAAE,MAAM,yBAAyB,CAAA;AACzE,OAAO,EAAE,sBAAsB,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAA;AAU1E,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,OAA2B,EAC3B,QAAqB,EACrB,KAAiB;IAEjB,MAAM,MAAM,GAAG,IAAI,aAAa,CAAC,EAAE,GAAG,OAAO,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,CAAA;IAClE,MAAM,KAAK,GAAG,OAAO,CAAC,YAAY,EAAE,CAAA;IACpC,MAAM,OAAO,GAA2D,EAAE,CAAA;IAC1E,MAAM,OAAO,GAA4B,EAAE,CAAA;IAC3C,IAAI,aAAa,GAAG,CAAC,CAAA;IAErB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,WAAW,GAAG,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC,CAAA;QACjD,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC;YACnC,OAAO,EAAE,IAAI,CAAC,WAAW;YACzB,OAAO;YACP,OAAO,EAAE,OAAO,CAAC,YAAY,CAAC,KAAK,CAAC;YACpC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,mBAAmB,EAAE,IAAI,CAAC,mBAAmB;YAC7C,YAAY,EAAE,OAAO,CAAC,YAAY;SACnC,CAAC,CAAA;QAEF,MAAM,UAAU,GAAG,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAA;QACtD,MAAM,QAAQ,GAAG,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,CAAA;QAE7C,MAAM,iBAAiB,GAAG,UAAU,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC;YACrD,CAAC,CAAC,MAAM,kBAAkB,CAAC,UAAU,CAAC,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE;gBACjE,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,QAAQ;aACjC,CAAC;YACJ,CAAC,CAAC,EAAE,CAAA;QAEN,oFAAoF;QACpF,IAAI,UAAU,CAAC,WAAW,IAAI,UAAU,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChE,wFAAwF;YACxF,MAAM,YAAY,GAAG,cAAc,CAAC,UAAU,CAAC,WAAW,CAAC,CAAA;YAC3D,MAAM,eAAe,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAAC,CAAC,OAAO,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAA;YACzF,MAAM,kBAAkB,GAAG,uBAAuB,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC,YAAY,aAAa,EAAE,EAAE,CAAC,CAAA;YAC9F,MAAM,eAAe,GAAG,EAAE,GAAG,kBAAkB,EAAE,GAAG,eAAe,EAAE,CAAA;YACrE,KAAK,MAAM,UAAU,IAAI,YAAY,EAAE,CAAC;gBACtC,MAAM,OAAO,GAAG,eAAe,CAAC,UAAU,CAAC,MAAM,CAAC,CAAA;gBAClD,IAAI,OAAO,EAAE,CAAC;oBACZ,MAAM,OAAO,CAAC,UAAU,CAAC,CAAA;gBAC3B,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,eAAe,GAAG,kBAAkB,CAAC,iBAAiB,CAAC,CAAA;QAC7D,MAAM,aAAa,GAAuB,iBAAiB,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;YACzE,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,QAAQ,EAAE,MAAM,CAAC,MAAM,KAAK,UAAU;YACtC,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,KAAK,EAAE,MAAM,CAAC,KAAK;SACpB,CAAC,CAAC,CAAA;QAEH,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC,CAAA;QAEzD,wEAAwE;QACxE,0EAA0E;QAC1E,2EAA2E;QAC3E,SAAS;QACT,MAAM,iBAAiB,GAAG,sBAAsB,CAAC,iBAAiB,CAAC,CAAA;QACnE,MAAM,cAAc,GAAG,iBAAiB,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;YACtD,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,UAAU,EAAE,MAAM,CAAC,UAAU;SAC9B,CAAC,CAAC,CAAA;QACH,qFAAqF;QACrF,MAAM,eAAe,GAAa,EAAE,CAAA;QACpC,IAAI,UAAU,CAAC,WAAW,IAAI,UAAU,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChE,MAAM,iBAAiB,GAAG,eAAe,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,IAAI,EAAE,CAAA;YACxF,eAAe,CAAC,IAAI,CAAC,GAAG,UAAU,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,sBAAsB,CAAC,CAAC,EAAE,iBAAiB,CAAC,CAAC,CAAC,CAAA;QACxG,CAAC;QACD,MAAM,sBAAsB,GAAG,+BAA+B,CAAC;YAC7D,OAAO,EAAE,UAAU,CAAC,OAAO;YAC3B,iBAAiB,EAAE,UAAU,CAAC,YAAY;YAC1C,cAAc,EAAE;gBACd,GAAG,cAAc;gBACjB,GAAG,eAAe,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC,EAAE,WAAW,EAAE,GAAG,UAAU,YAAY,EAAE,CAAC,CAAC;aACnF;YACD,iBAAiB,EAAE,CAAC,GAAG,iBAAiB,EAAE,GAAG,eAAe,CAAC;SAC9D,CAAC,CAAA;QACF,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,sBAAsB,EAAE,CAAC,CAAA;QAEpE,OAAO,CAAC,IAAI,CAAC;YACX,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,gBAAgB,EAAE,UAAU,CAAC,OAAO;YACpC,sBAAsB;YACtB,GAAG,EAAE,UAAU,CAAC,GAAG;YACnB,OAAO,EAAE,UAAU,CAAC,OAAO;YAC3B,aAAa;YACb,eAAe;YACf,SAAS,EAAE,UAAU,CAAC,SAAS,IAAI,EAAE;YACrC,YAAY,EAAE,UAAU,CAAC,YAAY;YACrC,WAAW;YACX,UAAU,EAAE,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC;YACzC,aAAa,EAAE,UAAU,CAAC,KAAK,CAAC;YAChC,iBAAiB;YACjB,KAAK,EAAE,UAAU,CAAC,KAAK;SACxB,CAAC,CAAA;IACJ,CAAC;IAED,OAAO;QACL,OAAO;QACP,QAAQ;QACR,KAAK,EAAE,OAAO;QACd,UAAU,EAAE,KAAK;QACjB,OAAO;KACR,CAAA;AACH,CAAC;AAED;;GAEG;AACH,SAAS,uBAAuB,CAC9B,KAAY,EACZ,MAAoB;IAEpB,OAAO;QACL,MAAM,EAAE,KAAK,EAAE,MAAkB,EAAE,EAAE;YACnC,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,EAAE,CAAA;YAClC,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ,EAAE,CAAC;gBAClC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC;oBAClB,EAAE,EAAE,MAAM,EAAE;oBACZ,OAAO,EAAE,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC;oBAC/B,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC;oBACjC,MAAM,EAAE,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;wBAC/F,CAAC,CAAC,MAAM,CAAC,MAAM;wBACf,CAAC,CAAC,SAAS;oBACb,SAAS,EAAE,MAAM,CAAC,SAAS,KAAK,SAAS,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW;wBAC/G,CAAC,CAAC,MAAM,CAAC,SAAS;wBAClB,CAAC,CAAC,SAAS;oBACb,WAAW,EAAE,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS;oBACpF,MAAM,EAAE,KAAK;oBACb,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;iBACpC,CAAC,CAAA;YACJ,CAAC;iBAAM,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;gBACtD,MAAM,MAAM,GAAG,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAA;gBACjE,IAAI,CAAC,MAAM;oBAAE,OAAM;gBACnB,IAAI,MAAM,CAAC,OAAO,IAAI,IAAI;oBAAE,MAAM,CAAC,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;gBACnE,IAAI,MAAM,CAAC,QAAQ,IAAI,IAAI;oBAAE,MAAM,CAAC,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAA;gBACtE,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;oBAAE,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAA;gBAC5H,IAAI,MAAM,CAAC,SAAS,KAAK,SAAS,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW;oBAAE,MAAM,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAA;gBAC/I,IAAI,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ;oBAAE,MAAM,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,CAAA;YACrF,CAAC;iBAAM,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;gBACtD,MAAM,KAAK,GAAG,KAAK,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAA;gBACrE,IAAI,KAAK,KAAK,CAAC,CAAC;oBAAE,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAA;YACnD,CAAC;QACH,CAAC;QACD,WAAW,EAAE,KAAK,EAAE,MAAkB,EAAE,EAAE;YACxC,iEAAiE;YACjE,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,EAAE,CAAA;YAClC,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ,EAAE,CAAC;gBAClC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC;oBAClB,EAAE,EAAE,MAAM,EAAE;oBACZ,OAAO,EAAE,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC;oBAC/B,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC;oBACjC,KAAK,EAAE,OAAO;oBACd,MAAM,EAAE,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;wBAC/F,CAAC,CAAC,MAAM,CAAC,MAAM;wBACf,CAAC,CAAC,SAAS;oBACb,SAAS,EAAE,MAAM,CAAC,SAAS,KAAK,SAAS,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW;wBAC/G,CAAC,CAAC,MAAM,CAAC,SAAS;wBAClB,CAAC,CAAC,SAAS;oBACb,WAAW,EAAE,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS;oBACpF,MAAM,EAAE,KAAK;oBACb,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;iBACpC,CAAC,CAAA;YACJ,CAAC;iBAAM,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;gBACtD,MAAM,MAAM,GAAG,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAA;gBACjE,IAAI,CAAC,MAAM;oBAAE,OAAM;gBACnB,IAAI,MAAM,CAAC,OAAO,IAAI,IAAI;oBAAE,MAAM,CAAC,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;gBACnE,IAAI,MAAM,CAAC,QAAQ,IAAI,IAAI;oBAAE,MAAM,CAAC,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAA;gBACtE,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;oBAAE,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAA;gBAC5H,IAAI,MAAM,CAAC,SAAS,KAAK,SAAS,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW,IAAI,MAAM,CAAC,SAAS,KAAK,WAAW;oBAAE,MAAM,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAA;gBAC/I,IAAI,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ;oBAAE,MAAM,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,CAAA;YACrF,CAAC;iBAAM,IAAI,MAAM,CAAC,SAAS,KAAK,QAAQ,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;gBACtD,MAAM,KAAK,GAAG,KAAK,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAA;gBACrE,IAAI,KAAK,KAAK,CAAC,CAAC;oBAAE,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAA;YACnD,CAAC;QACH,CAAC;KACF,CAAA;AACH,CAAC;AAED,SAAS,UAAU,CAAQ,KAAY;IACrC,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAU,CAAA;AACnD,CAAC"}
@@ -0,0 +1,143 @@
1
+ import type { EvalProject, EvalState } from './types.js';
2
+ export interface CoachEvalState extends EvalState {
3
+ threads: Array<{
4
+ id: string;
5
+ title: string;
6
+ status: string;
7
+ owner: string;
8
+ description?: string;
9
+ }>;
10
+ forcingFunctions: Array<{
11
+ id: string;
12
+ title: string;
13
+ owner: string;
14
+ due: string;
15
+ status: string;
16
+ threadId?: string;
17
+ }>;
18
+ coachingNotes: Array<{
19
+ id: string;
20
+ type: string;
21
+ text: string;
22
+ }>;
23
+ profile: {
24
+ name: string;
25
+ role: string;
26
+ company: string;
27
+ };
28
+ }
29
+ export interface NutritionEvalState extends EvalState {
30
+ meals: Array<{
31
+ id: string;
32
+ name: string;
33
+ calories?: number;
34
+ protein?: number;
35
+ }>;
36
+ recentDays: string[];
37
+ profile: {
38
+ name: string;
39
+ goal: string;
40
+ targetProtein: string;
41
+ };
42
+ }
43
+ export interface FitnessEvalState extends EvalState {
44
+ injuries: string[];
45
+ equipment: string[];
46
+ recentWorkouts: Array<{
47
+ id: string;
48
+ title: string;
49
+ result: string;
50
+ }>;
51
+ generatedWorkouts: Array<{
52
+ id: string;
53
+ exercises: Array<{
54
+ name: string;
55
+ sets: number;
56
+ reps: string;
57
+ notes?: string;
58
+ }>;
59
+ warmup?: string;
60
+ cooldown?: string;
61
+ }>;
62
+ profile: {
63
+ name: string;
64
+ focus: string;
65
+ };
66
+ }
67
+ export interface LanguageTutorEvalState extends EvalState {
68
+ currentGoal: {
69
+ language: string;
70
+ reason: string;
71
+ correctionPreference: string;
72
+ };
73
+ recentMistakes: Array<{
74
+ id: string;
75
+ pattern: string;
76
+ correction: string;
77
+ example?: string;
78
+ }>;
79
+ phrasebook: Array<{
80
+ text: string;
81
+ translation: string;
82
+ }>;
83
+ practiceQueue: Array<{
84
+ id: string;
85
+ title: string;
86
+ focus: string;
87
+ prompt: string;
88
+ durationMin?: number;
89
+ }>;
90
+ profile: {
91
+ name: string;
92
+ level: string;
93
+ };
94
+ }
95
+ export interface ChiefOfStaffEvalState extends EvalState {
96
+ openTasks: Array<{
97
+ id: string;
98
+ title: string;
99
+ owner?: string;
100
+ due?: string;
101
+ notes?: string;
102
+ definitionOfDone?: string;
103
+ priority?: 'low' | 'medium' | 'high';
104
+ status: 'open' | 'done' | 'canceled';
105
+ }>;
106
+ constraints: string[];
107
+ profile: {
108
+ name: string;
109
+ role: string;
110
+ };
111
+ }
112
+ export declare const coachProject: EvalProject<CoachEvalState>;
113
+ export declare const nutritionProject: EvalProject<NutritionEvalState>;
114
+ export declare const fitnessProject: EvalProject<FitnessEvalState>;
115
+ export declare const languageTutorProject: EvalProject<LanguageTutorEvalState>;
116
+ export declare const chiefOfStaffProject: EvalProject<ChiefOfStaffEvalState>;
117
+ /**
118
+ * Savor-specific eval: tests a production nutrition coaching persona
119
+ * that has the user's full food diary, body composition data, and
120
+ * accumulated relationship history. The key failure surfaces are:
121
+ * 1. Ignoring established routines visible in recent days + memories
122
+ * 2. Encyclopedic lecturing instead of grounded coaching
123
+ * 3. Double-logging / arithmetic hallucinations
124
+ */
125
+ export interface SavorEvalState extends EvalState {
126
+ meals: Array<{
127
+ id: string;
128
+ description: string;
129
+ calories: number;
130
+ proteinG: number;
131
+ carbsG: number;
132
+ fatG: number;
133
+ }>;
134
+ todayConsumed: {
135
+ calories: number;
136
+ proteinG: number;
137
+ carbsG: number;
138
+ fatG: number;
139
+ };
140
+ }
141
+ export declare const savorProject: EvalProject<SavorEvalState>;
142
+ export declare const SAMPLE_PROJECTS: readonly [EvalProject<CoachEvalState>, EvalProject<NutritionEvalState>, EvalProject<FitnessEvalState>, EvalProject<LanguageTutorEvalState>, EvalProject<ChiefOfStaffEvalState>, EvalProject<SavorEvalState>];
143
+ //# sourceMappingURL=sample-projects.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"sample-projects.d.ts","sourceRoot":"","sources":["../../src/evals/sample-projects.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAe,WAAW,EAAE,SAAS,EAAE,MAAM,YAAY,CAAA;AAUrE,MAAM,WAAW,cAAe,SAAQ,SAAS;IAC/C,OAAO,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAClG,gBAAgB,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,GAAG,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACrH,aAAa,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAChE,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAA;CACzD;AAED,MAAM,WAAW,kBAAmB,SAAQ,SAAS;IACnD,KAAK,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC/E,UAAU,EAAE,MAAM,EAAE,CAAA;IACpB,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,aAAa,EAAE,MAAM,CAAA;KAAE,CAAA;CAC/D;AAED,MAAM,WAAW,gBAAiB,SAAQ,SAAS;IACjD,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,SAAS,EAAE,MAAM,EAAE,CAAA;IACnB,cAAc,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACpE,iBAAiB,EAAE,KAAK,CAAC;QACvB,EAAE,EAAE,MAAM,CAAA;QACV,SAAS,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC,CAAA;QAC9E,MAAM,CAAC,EAAE,MAAM,CAAA;QACf,QAAQ,CAAC,EAAE,MAAM,CAAA;KAClB,CAAC,CAAA;IACF,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;CACzC;AAED,MAAM,WAAW,sBAAuB,SAAQ,SAAS;IACvD,WAAW,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,oBAAoB,EAAE,MAAM,CAAA;KAAE,CAAA;IAC/E,cAAc,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IAC5F,UAAU,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACxD,aAAa,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACxG,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;CACzC;AAED,MAAM,WAAW,qBAAsB,SAAQ,SAAS;IACtD,SAAS,EAAE,KAAK,CAAC;QACf,EAAE,EAAE,MAAM,CAAA;QACV,KAAK,EAAE,MAAM,CAAA;QACb,KAAK,CAAC,EAAE,MAAM,CAAA;QACd,GAAG,CAAC,EAAE,MAAM,CAAA;QACZ,KAAK,CAAC,EAAE,MAAM,CAAA;QACd,gBAAgB,CAAC,EAAE,MAAM,CAAA;QACzB,QAAQ,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAA;QACpC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,UAAU,CAAA;KACrC,CAAC,CAAA;IACF,WAAW,EAAE,MAAM,EAAE,CAAA;IACrB,OAAO,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;CACxC;AAKD,eAAO,MAAM,YAAY,EAAE,WAAW,CAAC,cAAc,CAkGpD,CAAA;AAED,eAAO,MAAM,gBAAgB,EAAE,WAAW,CAAC,kBAAkB,CAgE5D,CAAA;AAED,eAAO,MAAM,cAAc,EAAE,WAAW,CAAC,gBAAgB,CAsExD,CAAA;AAED,eAAO,MAAM,oBAAoB,EAAE,WAAW,CAAC,sBAAsB,CAwFpE,CAAA;AAED,eAAO,MAAM,mBAAmB,EAAE,WAAW,CAAC,qBAAqB,CAgFlE,CAAA;AAID;;;;;;;GAOG;AAEH,MAAM,WAAW,cAAe,SAAQ,SAAS;IAC/C,KAAK,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACnH,aAAa,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;CACpF;AAiCD,eAAO,MAAM,YAAY,EAAE,WAAW,CAAC,cAAc,CA6IpD,CAAA;AAED,eAAO,MAAM,eAAe,8MAOlB,CAAA"}