@cursor/july 0.1.16 → 0.1.18

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 (273) hide show
  1. package/README.md +9 -4
  2. package/dist/bin/agent-serve.js +60 -2
  3. package/dist/channels/slack/api.d.ts.map +1 -1
  4. package/dist/channels/slack/api.js +13 -3
  5. package/dist/channels/slack/auth.d.ts +7 -1
  6. package/dist/channels/slack/auth.d.ts.map +1 -1
  7. package/dist/channels/slack/auth.js +14 -0
  8. package/dist/channels/slack/blocks.d.ts +61 -0
  9. package/dist/channels/slack/blocks.d.ts.map +1 -0
  10. package/dist/channels/slack/blocks.js +90 -0
  11. package/dist/channels/slack/cursor-account.d.ts.map +1 -1
  12. package/dist/channels/slack/cursor-account.js +1 -1
  13. package/dist/channels/slack/defaults.d.ts +2 -0
  14. package/dist/channels/slack/defaults.d.ts.map +1 -1
  15. package/dist/channels/slack/defaults.js +29 -3
  16. package/dist/channels/slack/dispatch.d.ts +3 -17
  17. package/dist/channels/slack/dispatch.d.ts.map +1 -1
  18. package/dist/channels/slack/dispatch.js +4 -1
  19. package/dist/channels/slack/inbound.d.ts.map +1 -1
  20. package/dist/channels/slack/inbound.js +6 -1
  21. package/dist/channels/slack/index.d.ts +2 -0
  22. package/dist/channels/slack/index.d.ts.map +1 -1
  23. package/dist/channels/slack/index.js +2 -0
  24. package/dist/channels/slack/init.d.ts.map +1 -1
  25. package/dist/channels/slack/init.js +4 -1
  26. package/dist/channels/slack/interactive.d.ts +39 -3
  27. package/dist/channels/slack/interactive.d.ts.map +1 -1
  28. package/dist/channels/slack/interactive.js +153 -7
  29. package/dist/channels/slack/manifest.d.ts +13 -0
  30. package/dist/channels/slack/manifest.d.ts.map +1 -1
  31. package/dist/channels/slack/manifest.js +36 -14
  32. package/dist/channels/slack/message-body.d.ts +13 -0
  33. package/dist/channels/slack/message-body.d.ts.map +1 -0
  34. package/dist/channels/slack/message-body.js +82 -0
  35. package/dist/channels/slack/setup.d.ts +4 -1
  36. package/dist/channels/slack/setup.d.ts.map +1 -1
  37. package/dist/channels/slack/setup.js +16 -2
  38. package/dist/channels/slack/slack-channel.d.ts.map +1 -1
  39. package/dist/channels/slack/slack-channel.js +24 -3
  40. package/dist/channels/slack/thread-context.d.ts +0 -2
  41. package/dist/channels/slack/thread-context.d.ts.map +1 -1
  42. package/dist/channels/slack/thread-context.js +1 -65
  43. package/dist/channels/slack/types.d.ts +93 -1
  44. package/dist/channels/slack/types.d.ts.map +1 -1
  45. package/dist/connections.d.ts +4 -1
  46. package/dist/connections.d.ts.map +1 -1
  47. package/dist/connections.js +4 -1
  48. package/dist/docs/404.html +2 -2
  49. package/dist/docs/ab.html +3 -3
  50. package/dist/docs/assets/{app.DjHC-Ggf.js → app.BR5EdVSq.js} +1 -1
  51. package/dist/docs/assets/chunks/@localSearchIndexroot.BOzkeUi8.js +1 -0
  52. package/dist/docs/assets/chunks/{VPLocalSearchBox.DwviwmIq.js → VPLocalSearchBox.CNCzfUKC.js} +1 -1
  53. package/dist/docs/assets/chunks/{theme.BIpeYSon.js → theme.CwKvMSWv.js} +2 -2
  54. package/dist/docs/assets/{example-agents_security-reviewer.md.BZSNycMf.js → example-agents_security-reviewer.md.Dhj_m7_B.js} +1 -1
  55. package/dist/docs/assets/{guides_mcp-oauth.md.DqmEOQY1.js → guides_mcp-oauth.md.Dd8EgSem.js} +3 -2
  56. package/dist/docs/assets/{guides_mcp-oauth.md.DqmEOQY1.lean.js → guides_mcp-oauth.md.Dd8EgSem.lean.js} +1 -1
  57. package/dist/docs/assets/{guides_slack.md.FWale6Ip.js → guides_slack.md.D_xFKlnF.js} +1 -1
  58. package/dist/docs/assets/{reference_connections.md.CsSr3rDz.js → reference_connections.md.B9Q3TOve.js} +12 -5
  59. package/dist/docs/assets/{reference_connections.md.CsSr3rDz.lean.js → reference_connections.md.B9Q3TOve.lean.js} +1 -1
  60. package/dist/docs/building-with-agents.html +3 -3
  61. package/dist/docs/concepts.html +3 -3
  62. package/dist/docs/deployment.html +3 -3
  63. package/dist/docs/evals.html +3 -3
  64. package/dist/docs/example-agents/approval-buddy.html +3 -3
  65. package/dist/docs/example-agents/benny.html +3 -3
  66. package/dist/docs/example-agents/bugbot.html +3 -3
  67. package/dist/docs/example-agents/codebase-wiki.html +3 -3
  68. package/dist/docs/example-agents/codeowners-review.html +3 -3
  69. package/dist/docs/example-agents/concierge.html +3 -3
  70. package/dist/docs/example-agents/fsd.html +3 -3
  71. package/dist/docs/example-agents/index.html +3 -3
  72. package/dist/docs/example-agents/knowledge-base.html +3 -3
  73. package/dist/docs/example-agents/oncall.html +3 -3
  74. package/dist/docs/example-agents/security-reviewer.html +5 -5
  75. package/dist/docs/example-agents/slack-agent.html +3 -3
  76. package/dist/docs/example-agents/weather-agent.html +3 -3
  77. package/dist/docs/guides/agent-to-agent.html +3 -3
  78. package/dist/docs/guides/cloud-runtime.html +3 -3
  79. package/dist/docs/guides/github.html +3 -3
  80. package/dist/docs/guides/human-in-the-loop.html +3 -3
  81. package/dist/docs/guides/mcp-oauth.html +6 -5
  82. package/dist/docs/guides/slack.html +5 -5
  83. package/dist/docs/guides/webhooks.html +3 -3
  84. package/dist/docs/hashmap.json +1 -1
  85. package/dist/docs/hillclimbing.html +3 -3
  86. package/dist/docs/index.html +3 -3
  87. package/dist/docs/quickstart.html +3 -3
  88. package/dist/docs/reference/agent-config.html +3 -3
  89. package/dist/docs/reference/channels.html +3 -3
  90. package/dist/docs/reference/cli.html +3 -3
  91. package/dist/docs/reference/connections.html +14 -7
  92. package/dist/docs/reference/hooks.html +3 -3
  93. package/dist/docs/reference/http-api.html +3 -3
  94. package/dist/docs/reference/instructions.html +3 -3
  95. package/dist/docs/reference/playground.html +3 -3
  96. package/dist/docs/reference/project-layout.html +3 -3
  97. package/dist/docs/reference/prompt.html +3 -3
  98. package/dist/docs/reference/schedules.html +3 -3
  99. package/dist/docs/reference/sessions.html +3 -3
  100. package/dist/docs/reference/skills.html +3 -3
  101. package/dist/docs/reference/subagents.html +3 -3
  102. package/dist/docs/reference/tools.html +3 -3
  103. package/dist/docs/scaffolding-agents.html +3 -3
  104. package/dist/docs/storage.html +3 -3
  105. package/dist/docs/troubleshooting.html +3 -3
  106. package/dist/evals/assertions.d.ts +190 -0
  107. package/dist/evals/assertions.d.ts.map +1 -0
  108. package/dist/evals/assertions.js +461 -0
  109. package/dist/evals/expect.d.ts +95 -0
  110. package/dist/evals/expect.d.ts.map +1 -0
  111. package/dist/evals/expect.js +194 -0
  112. package/dist/evals/judge.d.ts +133 -0
  113. package/dist/evals/judge.d.ts.map +1 -0
  114. package/dist/evals/judge.js +228 -0
  115. package/dist/evals/loaders.d.ts +51 -0
  116. package/dist/evals/loaders.d.ts.map +1 -0
  117. package/dist/evals/loaders.js +124 -0
  118. package/dist/evals/matchers.d.ts +95 -0
  119. package/dist/evals/matchers.d.ts.map +1 -0
  120. package/dist/evals/matchers.js +145 -0
  121. package/dist/evals/reporters.d.ts +70 -0
  122. package/dist/evals/reporters.d.ts.map +1 -0
  123. package/dist/evals/reporters.js +221 -0
  124. package/dist/evals/results.d.ts +103 -0
  125. package/dist/evals/results.d.ts.map +1 -0
  126. package/dist/evals/results.js +28 -0
  127. package/dist/evals/run-facts.d.ts +77 -0
  128. package/dist/evals/run-facts.d.ts.map +1 -0
  129. package/dist/evals/run-facts.js +187 -0
  130. package/dist/evals.d.ts +118 -31
  131. package/dist/evals.d.ts.map +1 -1
  132. package/dist/evals.js +46 -12
  133. package/dist/index.d.ts +1 -1
  134. package/dist/index.d.ts.map +1 -1
  135. package/dist/index.js +1 -1
  136. package/dist/internal/builtin-tools/index.d.ts +10 -0
  137. package/dist/internal/builtin-tools/index.d.ts.map +1 -0
  138. package/dist/internal/builtin-tools/index.js +27 -0
  139. package/dist/internal/builtin-tools/reminders.d.ts +9 -0
  140. package/dist/internal/builtin-tools/reminders.d.ts.map +1 -0
  141. package/dist/internal/builtin-tools/reminders.js +138 -0
  142. package/dist/internal/cli-ax.d.ts +27 -0
  143. package/dist/internal/cli-ax.d.ts.map +1 -1
  144. package/dist/internal/cli-ax.js +90 -12
  145. package/dist/internal/cli-mcp.d.ts +8 -0
  146. package/dist/internal/cli-mcp.d.ts.map +1 -1
  147. package/dist/internal/cli-mcp.js +98 -0
  148. package/dist/internal/cli-slack.d.ts +29 -1
  149. package/dist/internal/cli-slack.d.ts.map +1 -1
  150. package/dist/internal/cli-slack.js +799 -13
  151. package/dist/internal/cursor-event-relay.d.ts +30 -0
  152. package/dist/internal/cursor-event-relay.d.ts.map +1 -1
  153. package/dist/internal/cursor-event-relay.js +37 -0
  154. package/dist/internal/deploy-client.d.ts +7 -0
  155. package/dist/internal/deploy-client.d.ts.map +1 -1
  156. package/dist/internal/deploy-client.js +9 -0
  157. package/dist/internal/discovery.d.ts.map +1 -1
  158. package/dist/internal/discovery.js +70 -9
  159. package/dist/internal/distribution.d.ts.map +1 -1
  160. package/dist/internal/distribution.js +2 -0
  161. package/dist/internal/eval-judge-model.d.ts +59 -0
  162. package/dist/internal/eval-judge-model.d.ts.map +1 -0
  163. package/dist/internal/eval-judge-model.js +131 -0
  164. package/dist/internal/eval-run-store.d.ts.map +1 -1
  165. package/dist/internal/eval-run-store.js +12 -20
  166. package/dist/internal/eval-runner.d.ts +31 -25
  167. package/dist/internal/eval-runner.d.ts.map +1 -1
  168. package/dist/internal/eval-runner.js +329 -126
  169. package/dist/internal/evals-client.d.ts.map +1 -1
  170. package/dist/internal/evals-client.js +4 -1
  171. package/dist/internal/local-env.d.ts +8 -0
  172. package/dist/internal/local-env.d.ts.map +1 -0
  173. package/dist/internal/local-env.js +47 -0
  174. package/dist/internal/playground/static.d.ts +3 -0
  175. package/dist/internal/playground/static.d.ts.map +1 -1
  176. package/dist/internal/playground/static.js +3 -0
  177. package/dist/internal/reminder-runner.d.ts +7 -0
  178. package/dist/internal/reminder-runner.d.ts.map +1 -1
  179. package/dist/internal/reminder-runner.js +21 -0
  180. package/dist/internal/run-client.d.ts +15 -0
  181. package/dist/internal/run-client.d.ts.map +1 -1
  182. package/dist/internal/run-client.js +3 -2
  183. package/dist/internal/sdk-runner.d.ts +51 -1
  184. package/dist/internal/sdk-runner.d.ts.map +1 -1
  185. package/dist/internal/sdk-runner.js +176 -36
  186. package/dist/internal/server.d.ts.map +1 -1
  187. package/dist/internal/server.js +11 -1
  188. package/dist/internal/session-engine.d.ts +18 -5
  189. package/dist/internal/session-engine.d.ts.map +1 -1
  190. package/dist/internal/session-engine.js +81 -19
  191. package/dist/internal/slack-provision-client.d.ts +133 -0
  192. package/dist/internal/slack-provision-client.d.ts.map +1 -0
  193. package/dist/internal/slack-provision-client.js +286 -0
  194. package/dist/playground/assets/cursor-icons-16-f_W_ogc-.woff2 +0 -0
  195. package/dist/playground/assets/index-DRjS5klQ.js +85 -0
  196. package/dist/playground/assets/index-TCDs-UhZ.css +1 -0
  197. package/dist/playground/index.html +2 -2
  198. package/dist/types.d.ts +98 -4
  199. package/dist/types.d.ts.map +1 -1
  200. package/dist/types.js +26 -0
  201. package/docs/example-agents/security-reviewer.md +1 -1
  202. package/docs/guides/mcp-oauth.md +10 -5
  203. package/docs/guides/slack.md +3 -2
  204. package/docs/reference/connections.md +28 -18
  205. package/package.json +20 -2
  206. package/skills/create-agent/SKILL.md +15 -7
  207. package/skills/debug/SKILL.md +3 -1
  208. package/skills/mcp-auth/SKILL.md +1 -1
  209. package/skills/setup-slack/SKILL.md +177 -11
  210. package/src/bin/agent-serve.ts +63 -2
  211. package/src/channels/slack/api.ts +16 -3
  212. package/src/channels/slack/auth.ts +16 -1
  213. package/src/channels/slack/blocks.ts +139 -0
  214. package/src/channels/slack/cursor-account.ts +1 -0
  215. package/src/channels/slack/defaults.ts +32 -4
  216. package/src/channels/slack/dispatch.ts +6 -19
  217. package/src/channels/slack/inbound.ts +6 -1
  218. package/src/channels/slack/index.ts +2 -0
  219. package/src/channels/slack/init.ts +3 -0
  220. package/src/channels/slack/interactive.ts +184 -19
  221. package/src/channels/slack/manifest.ts +37 -6
  222. package/src/channels/slack/message-body.ts +90 -0
  223. package/src/channels/slack/setup.ts +16 -2
  224. package/src/channels/slack/slack-channel.ts +37 -6
  225. package/src/channels/slack/thread-context.ts +1 -69
  226. package/src/channels/slack/types.ts +105 -1
  227. package/src/connections.ts +4 -1
  228. package/src/evals/assertions.ts +704 -0
  229. package/src/evals/expect.ts +272 -0
  230. package/src/evals/judge.ts +346 -0
  231. package/src/evals/loaders.ts +113 -0
  232. package/src/evals/matchers.ts +246 -0
  233. package/src/evals/reporters.ts +274 -0
  234. package/src/evals/results.ts +132 -0
  235. package/src/evals/run-facts.ts +274 -0
  236. package/src/evals.ts +205 -42
  237. package/src/index.ts +2 -0
  238. package/src/internal/builtin-tools/index.ts +40 -0
  239. package/src/internal/builtin-tools/reminders.ts +158 -0
  240. package/src/internal/cli-ax.ts +134 -15
  241. package/src/internal/cli-mcp.ts +118 -5
  242. package/src/internal/cli-slack.ts +1017 -15
  243. package/src/internal/cursor-event-relay.ts +67 -0
  244. package/src/internal/deploy-client.ts +10 -0
  245. package/src/internal/discovery.ts +83 -5
  246. package/src/internal/distribution.ts +2 -0
  247. package/src/internal/eval-judge-model.ts +148 -0
  248. package/src/internal/eval-run-store.ts +10 -14
  249. package/src/internal/eval-runner.ts +460 -214
  250. package/src/internal/evals-client.ts +5 -0
  251. package/src/internal/local-env.ts +53 -0
  252. package/src/internal/playground/static.ts +3 -0
  253. package/src/internal/reminder-runner.ts +30 -0
  254. package/src/internal/run-client.ts +17 -1
  255. package/src/internal/sdk-runner.ts +234 -35
  256. package/src/internal/server.ts +12 -0
  257. package/src/internal/session-engine.ts +103 -21
  258. package/src/internal/slack-provision-client.ts +442 -0
  259. package/src/types.ts +125 -4
  260. package/dist/channels/slack/post-update-delivery.d.ts +0 -85
  261. package/dist/channels/slack/post-update-delivery.d.ts.map +0 -1
  262. package/dist/docs/assets/chunks/@localSearchIndexroot.DZtu0ZIc.js +0 -1
  263. package/dist/internal/json-dir-store.d.ts +0 -32
  264. package/dist/internal/json-dir-store.d.ts.map +0 -1
  265. package/dist/internal/persistence-coordinator.d.ts +0 -127
  266. package/dist/internal/persistence-coordinator.d.ts.map +0 -1
  267. package/dist/persistence.d.ts +0 -184
  268. package/dist/persistence.d.ts.map +0 -1
  269. package/dist/playground/assets/cursor-icons-16-CQ50JpfO.woff2 +0 -0
  270. package/dist/playground/assets/index-Bwl4Onx4.js +0 -85
  271. package/dist/playground/assets/index-ByV4nfch.css +0 -1
  272. /package/dist/docs/assets/{example-agents_security-reviewer.md.BZSNycMf.lean.js → example-agents_security-reviewer.md.Dhj_m7_B.lean.js} +0 -0
  273. /package/dist/docs/assets/{guides_slack.md.FWale6Ip.lean.js → guides_slack.md.D_xFKlnF.lean.js} +0 -0
package/dist/evals.d.ts CHANGED
@@ -44,33 +44,62 @@
44
44
  * ],
45
45
  * });
46
46
  * ```
47
+ *
48
+ * ## Severity: gates and scores
49
+ *
50
+ * Every assertion returns a chainable handle, so severity rides on the
51
+ * assertion instead of a separate thresholds map. Gates are hard (a miss fails
52
+ * the eval and `eval` exits non-zero); soft assertions are tracked scores that
53
+ * mark the eval `scored` and only fail the build under `--strict`:
54
+ *
55
+ * ```ts
56
+ * t.succeeded(); // gate (default)
57
+ * t.calledTool("get_weather").soft(); // tracked, no gate
58
+ * t.check(t.reply, similarity("Sunny, 72F")).atLeast(0.8);
59
+ * t.judge.closedQA("cites a source").atLeast(0.6); // LLM-graded
60
+ * ```
61
+ *
62
+ * Matchers narrow tool assertions (`{ input, output, status, count }` accept a
63
+ * literal, a RegExp, or a predicate), and each turn returned by `t.send(...)`
64
+ * carries the same vocabulary scoped to that turn.
47
65
  */
66
+ import type { EvalAssertionResult, EvalAssertions, EvalVerdict } from "./evals/assertions.js";
67
+ import type { Expectation } from "./evals/expect.js";
68
+ import type { EvalJudge, EvalJudgeConfig } from "./evals/judge.js";
69
+ import type { EvalReporter } from "./evals/results.js";
48
70
  import type { AgentCloudOptions, SessionEvent } from "./types.js";
49
- export type Expectation = {
50
- kind: "includes";
51
- value: string | RegExp;
52
- } | {
53
- kind: "equals";
54
- value: unknown;
55
- } | {
56
- kind: "satisfies";
57
- predicate: (value: unknown) => boolean;
58
- label: string;
59
- };
60
- export declare function includes(value: string | RegExp): Expectation;
61
- export declare function equals(value: unknown): Expectation;
62
- export declare function satisfies(predicate: (value: unknown) => boolean, label: string): Expectation;
63
- export interface EvalAssertionResult {
64
- name: string;
65
- passed: boolean;
66
- detail?: string;
67
- }
68
- export interface EvalTurnResult {
71
+ export type { EvalAssertionHandle, EvalAssertionResult, EvalAssertionSeverity, EvalAssertions, EvalVerdict, } from "./evals/assertions.js";
72
+ export { computeVerdict, verdictFailsRun } from "./evals/assertions.js";
73
+ export type { EvalSchemaLike, Expectation } from "./evals/expect.js";
74
+ export { equals, includes, matches, normalizedSimilarity, satisfies, similarity, } from "./evals/expect.js";
75
+ export type { EvalJudge, EvalJudgeCallOptions, EvalJudgeConfig, EvalJudgeGraderName, } from "./evals/judge.js";
76
+ export { EVAL_JUDGE_INJECTION_GUARD, fenceUntrusted, } from "./evals/judge.js";
77
+ export type { EvalCountMatcher, EvalEventMatcher, EvalEventOrderMatcher, EvalMatcher, EvalSubagentMatcher, EvalToolCallMatcher, EvalToolCallStatus, } from "./evals/matchers.js";
78
+ export type { EvalReporter, EvalReporterEvalInfo, EvalReporterTarget, EvalRunResult, EvalRunSummary, } from "./evals/results.js";
79
+ export type { EvalInputRequestFact, EvalRunFacts, EvalSubagentFact, EvalToolCallFact, } from "./evals/run-facts.js";
80
+ export { deriveRunFacts } from "./evals/run-facts.js";
81
+ /**
82
+ * One settled turn. Carries the same assertion vocabulary as `t`, narrowed to
83
+ * this turn's events, so an intermediate turn can be graded before the next
84
+ * `send` overwrites `t.reply`:
85
+ *
86
+ * ```ts
87
+ * const draft = await t.send("Draft the summary.");
88
+ * draft.calledTool("read_file");
89
+ * t.check(draft.message, includes("Summary"));
90
+ * ```
91
+ */
92
+ export interface EvalTurnResult extends EvalAssertions {
69
93
  message: string;
70
94
  sessionId: string;
71
95
  events: SessionEvent[];
96
+ /** Tool names requested during this turn, in order. */
72
97
  toolCalls: string[];
73
98
  ok: boolean;
99
+ /** 1-based turn index within the case. */
100
+ index: number;
101
+ /** Throws when this turn failed. Only needed when later steps depend on it. */
102
+ expectOk(): EvalTurnResult;
74
103
  }
75
104
  export interface EvalSendOptions {
76
105
  /** Absolute harness `cwd` for a new session (`SendMessageOptions.workspaceDir`). */
@@ -84,7 +113,15 @@ export interface EvalSendOptions {
84
113
  */
85
114
  cloud?: AgentCloudOptions;
86
115
  }
87
- export interface EvalTestContext {
116
+ /**
117
+ * The eval driver and assertion surface.
118
+ *
119
+ * `t` both drives the agent and grades it. Assertions on `t` read the whole run
120
+ * once `test` finishes; assertions on the turn returned by `t.send(...)` read
121
+ * only that turn. Recorded assertions never throw, so one run reports every
122
+ * failure rather than dying on the first.
123
+ */
124
+ export interface EvalTestContext extends EvalAssertions {
88
125
  /** Send a turn and wait until the session parks (or fails). */
89
126
  send(message: string, options?: EvalSendOptions): Promise<EvalTurnResult>;
90
127
  /** Last assistant text, or null before the first send. */
@@ -92,6 +129,8 @@ export interface EvalTestContext {
92
129
  readonly sessionId: string | undefined;
93
130
  /** All events captured so far across turns. */
94
131
  readonly events: SessionEvent[];
132
+ /** Turns settled so far, oldest first. */
133
+ readonly turns: readonly EvalTurnResult[];
95
134
  /**
96
135
  * 1-based index of this run when {@link EvalConfig.iterations} /
97
136
  * {@link EvalDatapoint.iterations} expands a datapoint. Always `1` when
@@ -100,16 +139,25 @@ export interface EvalTestContext {
100
139
  readonly iteration: number;
101
140
  /** Total iterations for this datapoint (default `1`). */
102
141
  readonly iterations: number;
103
- /** Gate: run did not fail. */
104
- succeeded(): void;
105
- /** Gate: at least one completed call to `name`. */
106
- calledTool(name: string): void;
107
- /** Gate: no call to `name`. */
108
- notCalledTool(name: string): void;
109
- /** Gate: joined assistant text matches. */
110
- messageIncludes(token: string | RegExp): void;
111
- /** Gate a value with an {@link Expectation} builder. */
112
- check(value: unknown, assertion: Expectation): void;
142
+ /** Aborts when the case hits its timeout. */
143
+ readonly signal: AbortSignal;
144
+ /**
145
+ * LLM-as-judge assertions. Soft by default (tracked, never fatal); add a bar
146
+ * with `.atLeast(0.7)` or promote with `.gate(0.8)`. Requires a judge model
147
+ * from `defineEvalConfig({ judge })`, `defineEval({ judge })`, or the call.
148
+ */
149
+ readonly judge: EvalJudge;
150
+ /**
151
+ * Record a gate that must pass before the script can safely continue.
152
+ * Returns `value` when it passes and stops the test body when it does not,
153
+ * without adding a duplicate execution error.
154
+ */
155
+ require<T>(value: T, assertion: Expectation): Promise<T>;
156
+ /**
157
+ * End the case as skipped. Skipped cases are reported separately and never
158
+ * change the exit code. Call it before sending messages.
159
+ */
160
+ skip(reason: string): never;
113
161
  /** Debug line (surfaced under `agent-serve eval --verbose`). */
114
162
  log(message: string): void;
115
163
  /**
@@ -139,6 +187,12 @@ export interface EvalDatapoint {
139
187
  * concurrent case.
140
188
  */
141
189
  iterations?: number;
190
+ /** Free-form data carried onto the result and every reporter. */
191
+ metadata?: Record<string, unknown>;
192
+ /** Judge model for this datapoint (overrides `evals.config.ts`). */
193
+ judge?: EvalJudgeConfig;
194
+ /** Destinations for just this datapoint, in addition to config reporters. */
195
+ reporters?: EvalReporter[];
142
196
  test: (t: EvalTestContext) => Promise<void>;
143
197
  }
144
198
  /**
@@ -154,6 +208,12 @@ export interface EvalConfig {
154
208
  * Case-level {@link EvalDatapoint.iterations} overrides this.
155
209
  */
156
210
  iterations?: number;
211
+ /** Free-form data carried onto every result in this file and to reporters. */
212
+ metadata?: Record<string, unknown>;
213
+ /** Judge model for this file (overrides `evals.config.ts`). */
214
+ judge?: EvalJudgeConfig;
215
+ /** Destinations for this file, in addition to config reporters. */
216
+ reporters?: EvalReporter[];
157
217
  /** Single-datapoint shorthand. Case id = file path under `evals/`. */
158
218
  test?: (t: EvalTestContext) => Promise<void>;
159
219
  /** Multiple datapoints in this file. Each id = `<fileId>/<case.id>`. */
@@ -168,6 +228,9 @@ export interface EvalCaseDefinition {
168
228
  iterations?: number;
169
229
  /** 1-based iteration index for this expanded runnable case. */
170
230
  iteration?: number;
231
+ metadata?: Record<string, unknown>;
232
+ judge?: EvalJudgeConfig;
233
+ reporters?: EvalReporter[];
171
234
  test: (t: EvalTestContext) => Promise<void>;
172
235
  readonly __agentServe: "eval";
173
236
  }
@@ -239,6 +302,10 @@ export interface EvalCaseSnapshot {
239
302
  tags?: string[];
240
303
  status: EvalCaseRunStatus;
241
304
  ok?: boolean;
305
+ /** Full grade, including `scored` (soft threshold miss) and `skipped`. */
306
+ verdict?: EvalVerdict;
307
+ /** Reason passed to `t.skip(...)`. */
308
+ skipReason?: string;
242
309
  assertions?: EvalAssertionResult[];
243
310
  sessionId?: string;
244
311
  durationMs?: number;
@@ -246,6 +313,8 @@ export interface EvalCaseSnapshot {
246
313
  logs?: string[];
247
314
  /** Structured scores from {@link EvalTestContext.metric}. */
248
315
  metrics?: Record<string, string | number>;
316
+ /** Author-supplied metadata from `defineEval({ metadata })`. */
317
+ metadata?: Record<string, unknown>;
249
318
  inputs?: string[];
250
319
  finalText?: string;
251
320
  tools?: string[];
@@ -266,6 +335,13 @@ export interface EvalRunSnapshot {
266
335
  summary: {
267
336
  passed: number;
268
337
  failed: number;
338
+ /**
339
+ * Cases that only missed a soft threshold (fatal only under `--strict`).
340
+ * Optional: a batch persisted by an older serve process has no such count.
341
+ */
342
+ scored?: number;
343
+ /** Cases ended by `t.skip(...)`; they never change the exit code. */
344
+ skipped?: number;
269
345
  total: number;
270
346
  done: number;
271
347
  };
@@ -326,6 +402,17 @@ export interface EvalConfigFile {
326
402
  * request limits.
327
403
  */
328
404
  maxConcurrency: number;
405
+ /**
406
+ * Default judge model for `t.judge.*` across this tree. Optional: a tree of
407
+ * fully deterministic evals never needs one.
408
+ */
409
+ judge?: EvalJudgeConfig;
410
+ /**
411
+ * Destinations that observe **every** case in the run — the right place for a
412
+ * shared one, so you do not repeat it in each eval file. `--skip-report`
413
+ * suppresses them for a local iteration loop.
414
+ */
415
+ reporters?: EvalReporter[];
329
416
  /**
330
417
  * Optional. Max batches kept in the **playground / `/v1/dev/evals*`**
331
418
  * history window (in-memory, and pruned from the `defineStorage` evals
@@ -1 +1 @@
1
- {"version":3,"file":"evals.d.ts","sourceRoot":"","sources":["../src/evals.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AAIH,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAElE,MAAM,MAAM,WAAW,GACnB;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,KAAK,EAAE,OAAO,CAAA;CAAE,GAClC;IACE,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,CAAC;IACvC,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AAEN,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,WAAW,CAE5D;AAED,wBAAgB,MAAM,CAAC,KAAK,EAAE,OAAO,GAAG,WAAW,CAElD;AAED,wBAAgB,SAAS,CACvB,SAAS,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,EACtC,KAAK,EAAE,MAAM,GACZ,WAAW,CAEb;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,YAAY,EAAE,CAAC;IACvB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,EAAE,EAAE,OAAO,CAAC;CACb;AAED,MAAM,WAAW,eAAe;IAC9B,oFAAoF;IACpF,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,qEAAqE;IACrE,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACxC;;;;OAIG;IACH,KAAK,CAAC,EAAE,iBAAiB,CAAC;CAC3B;AAED,MAAM,WAAW,eAAe;IAC9B,+DAA+D;IAC/D,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CAAC;IAC1E,0DAA0D;IAC1D,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,+CAA+C;IAC/C,QAAQ,CAAC,MAAM,EAAE,YAAY,EAAE,CAAC;IAChC;;;;OAIG;IACH,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,yDAAyD;IACzD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,8BAA8B;IAC9B,SAAS,IAAI,IAAI,CAAC;IAClB,mDAAmD;IACnD,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,+BAA+B;IAC/B,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,2CAA2C;IAC3C,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAAC;IAC9C,wDAAwD;IACxD,KAAK,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,WAAW,GAAG,IAAI,CAAC;IACpD,gEAAgE;IAChE,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B;;;;OAIG;IACH,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAAC;CACpD;AAED,0EAA0E;AAC1E,eAAO,MAAM,mBAAmB,MAAM,CAAC;AAEvC,0EAA0E;AAC1E,MAAM,WAAW,aAAa;IAC5B;;;;OAIG;IACH,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED;;;GAGG;AACH,MAAM,WAAW,UAAU;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,sEAAsE;IACtE,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7C,wEAAwE;IACxE,KAAK,CAAC,EAAE,aAAa,EAAE,CAAC;CACzB;AAED,sEAAsE;AACtE,MAAM,WAAW,kBAAkB;IACjC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,8DAA8D;IAC9D,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,+DAA+D;IAC/D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5C,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAED,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG;IAAE,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAAC;AAE5E;;;;GAIG;AACH,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAK9D;AAED,wBAAgB,UAAU,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAoC7D;AAaD;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAClC,MAAM,EAAE,MAAM,EACd,UAAU,EAAE,cAAc,GACzB,KAAK,CAAC;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,kBAAkB,CAAA;CAAE,CAAC,CA8BvD;AAiCD;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAE3E;AAED;;;;;GAKG;AACH,eAAO,MAAM,gBAAgB,YAAY,CAAC;AAE1C,sDAAsD;AACtD,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,GAClC,OAAO,CAET;AAED;;;;GAIG;AACH,wBAAgB,2BAA2B,CACzC,CAAC,SAAS;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EAExC,GAAG,EAAE,SAAS,CAAC,EAAE,EACjB,QAAQ,EAAE,SAAS,CAAC,EAAE,EACtB,SAAS,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,OAAO,GAC9B,CAAC,EAAE,CAgBL;AAED;;;;;GAKG;AACH,eAAO,MAAM,0BAA0B,MAAM,CAAC;AAE9C,gGAAgG;AAChG,eAAO,MAAM,gCAAgC,KAAK,CAAC;AAEnD,iEAAiE;AACjE,eAAO,MAAM,wBAAwB,MAAM,CAAC;AAE5C,MAAM,MAAM,iBAAiB,GAAG,SAAS,GAAG,SAAS,GAAG,MAAM,CAAC;AAC/D,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,WAAW,GAAG,QAAQ,GAAG,WAAW,CAAC;AAE/E,+DAA+D;AAC/D,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,iBAAiB,CAAC;IAC1B,EAAE,CAAC,EAAE,OAAO,CAAC;IACb,UAAU,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,6DAA6D;IAC7D,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAAC,CAAC;IAC1C,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,KAAK,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,CAAC,EAAE,OAAO,CAAC;QACf,MAAM,CAAC,EAAE,OAAO,CAAC;QACjB,OAAO,EAAE,OAAO,CAAC;KAClB,CAAC,CAAC;CACJ;AAED,gFAAgF;AAChF,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,eAAe,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACzE,KAAK,EAAE,gBAAgB,EAAE,CAAC;IAC1B,MAAM,EAAE;QACN,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,WAAW,CAAC,EAAE,OAAO,CAAC;KACvB,CAAC;CACH;AAED,8EAA8E;AAC9E,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,4DAA4D;AAC5D,MAAM,WAAW,kBAAkB;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,8DAA8D;IAC9D,WAAW,EAAE,OAAO,CAAC;CACtB;AAED,wFAAwF;AACxF,MAAM,WAAW,qBAAqB;IACpC,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,wEAAwE;IACxE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,IAAI,IAAI,OAAO,CAAC;QACd,KAAK,EAAE,gBAAgB,EAAE,CAAC;QAC1B,MAAM,EAAE,kBAAkB,CAAC;KAC5B,CAAC,CAAC;IACH,GAAG,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,GAAG,SAAS,CAAC,CAAC;IACzD,kBAAkB,IAAI,OAAO,CAAC;QAC5B,IAAI,EAAE,eAAe,EAAE,CAAC;QACxB,WAAW,CAAC,EAAE,MAAM,CAAC;KACtB,CAAC,CAAC;IACH,KAAK,CAAC,KAAK,CAAC,EAAE,qBAAqB,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;IAC/D,mDAAmD;IACnD,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,cAAc;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,cAAc,EAAE,MAAM,CAAC;IACvB;;;;;;;;OAQG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED;;;;;;GAMG;AACH,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,cAAc,GAAG,cAAc,CAQvE;AAED;;GAEG;AACH,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAMlE;AAED,yDAAyD;AACzD,wBAAgB,8BAA8B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAKrE;AAED;;;GAGG;AACH,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CASnE"}
1
+ {"version":3,"file":"evals.d.ts","sourceRoot":"","sources":["../src/evals.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgEG;AAEH,OAAO,KAAK,EACV,mBAAmB,EACnB,cAAc,EACd,WAAW,EACZ,MAAM,uBAAuB,CAAC;AAC/B,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AACnE,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAGvD,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAKlE,YAAY,EACV,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACrB,cAAc,EACd,WAAW,GACZ,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxE,YAAY,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrE,OAAO,EACL,MAAM,EACN,QAAQ,EACR,OAAO,EACP,oBAAoB,EACpB,SAAS,EACT,UAAU,GACX,MAAM,mBAAmB,CAAC;AAC3B,YAAY,EACV,SAAS,EACT,oBAAoB,EACpB,eAAe,EACf,mBAAmB,GACpB,MAAM,kBAAkB,CAAC;AAG1B,OAAO,EACL,0BAA0B,EAC1B,cAAc,GACf,MAAM,kBAAkB,CAAC;AAC1B,YAAY,EACV,gBAAgB,EAChB,gBAAgB,EAChB,qBAAqB,EACrB,WAAW,EACX,mBAAmB,EACnB,mBAAmB,EACnB,kBAAkB,GACnB,MAAM,qBAAqB,CAAC;AAC7B,YAAY,EACV,YAAY,EACZ,oBAAoB,EACpB,kBAAkB,EAClB,aAAa,EACb,cAAc,GACf,MAAM,oBAAoB,CAAC;AAC5B,YAAY,EACV,oBAAoB,EACpB,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,GACjB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAEtD;;;;;;;;;;GAUG;AACH,MAAM,WAAW,cAAe,SAAQ,cAAc;IACpD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,YAAY,EAAE,CAAC;IACvB,uDAAuD;IACvD,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,EAAE,EAAE,OAAO,CAAC;IACZ,0CAA0C;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,+EAA+E;IAC/E,QAAQ,IAAI,cAAc,CAAC;CAC5B;AAED,MAAM,WAAW,eAAe;IAC9B,oFAAoF;IACpF,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,qEAAqE;IACrE,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACxC;;;;OAIG;IACH,KAAK,CAAC,EAAE,iBAAiB,CAAC;CAC3B;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,eAAgB,SAAQ,cAAc;IACrD,+DAA+D;IAC/D,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CAAC;IAC1E,0DAA0D;IAC1D,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,+CAA+C;IAC/C,QAAQ,CAAC,MAAM,EAAE,YAAY,EAAE,CAAC;IAChC,0CAA0C;IAC1C,QAAQ,CAAC,KAAK,EAAE,SAAS,cAAc,EAAE,CAAC;IAC1C;;;;OAIG;IACH,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,yDAAyD;IACzD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,6CAA6C;IAC7C,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAC7B;;;;OAIG;IACH,QAAQ,CAAC,KAAK,EAAE,SAAS,CAAC;IAC1B;;;;OAIG;IACH,OAAO,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,SAAS,EAAE,WAAW,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IACzD;;;OAGG;IACH,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,KAAK,CAAC;IAC5B,gEAAgE;IAChE,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B;;;;OAIG;IACH,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAAC;CACpD;AAED,0EAA0E;AAC1E,eAAO,MAAM,mBAAmB,MAAM,CAAC;AAEvC,0EAA0E;AAC1E,MAAM,WAAW,aAAa;IAC5B;;;;OAIG;IACH,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,iEAAiE;IACjE,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,oEAAoE;IACpE,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,6EAA6E;IAC7E,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B,IAAI,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED;;;GAGG;AACH,MAAM,WAAW,UAAU;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8EAA8E;IAC9E,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,+DAA+D;IAC/D,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,mEAAmE;IACnE,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B,sEAAsE;IACtE,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7C,wEAAwE;IACxE,KAAK,CAAC,EAAE,aAAa,EAAE,CAAC;CACzB;AAED,sEAAsE;AACtE,MAAM,WAAW,kBAAkB;IACjC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,8DAA8D;IAC9D,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,+DAA+D;IAC/D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B,IAAI,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5C,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAED,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG;IAAE,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAAC;AAE5E;;;;GAIG;AACH,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAK9D;AAED,wBAAgB,UAAU,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAoC7D;AAaD;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAClC,MAAM,EAAE,MAAM,EACd,UAAU,EAAE,cAAc,GACzB,KAAK,CAAC;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,kBAAkB,CAAA;CAAE,CAAC,CAwCvD;AAqDD;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAE3E;AAED;;;;;GAKG;AACH,eAAO,MAAM,gBAAgB,YAAY,CAAC;AAE1C,sDAAsD;AACtD,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,GAClC,OAAO,CAET;AAED;;;;GAIG;AACH,wBAAgB,2BAA2B,CACzC,CAAC,SAAS;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EAExC,GAAG,EAAE,SAAS,CAAC,EAAE,EACjB,QAAQ,EAAE,SAAS,CAAC,EAAE,EACtB,SAAS,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,OAAO,GAC9B,CAAC,EAAE,CAgBL;AAED;;;;;GAKG;AACH,eAAO,MAAM,0BAA0B,MAAM,CAAC;AAE9C,gGAAgG;AAChG,eAAO,MAAM,gCAAgC,KAAK,CAAC;AAEnD,iEAAiE;AACjE,eAAO,MAAM,wBAAwB,MAAM,CAAC;AAE5C,MAAM,MAAM,iBAAiB,GAAG,SAAS,GAAG,SAAS,GAAG,MAAM,CAAC;AAC/D,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,WAAW,GAAG,QAAQ,GAAG,WAAW,CAAC;AAE/E,+DAA+D;AAC/D,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,iBAAiB,CAAC;IAC1B,EAAE,CAAC,EAAE,OAAO,CAAC;IACb,0EAA0E;IAC1E,OAAO,CAAC,EAAE,WAAW,CAAC;IACtB,sCAAsC;IACtC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,6DAA6D;IAC7D,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAAC,CAAC;IAC1C,gEAAgE;IAChE,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,KAAK,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,CAAC,EAAE,OAAO,CAAC;QACf,MAAM,CAAC,EAAE,OAAO,CAAC;QACjB,OAAO,EAAE,OAAO,CAAC;KAClB,CAAC,CAAC;CACJ;AAED,gFAAgF;AAChF,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,eAAe,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,EAAE;QACP,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf;;;WAGG;QACH,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,qEAAqE;QACrE,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,KAAK,EAAE,MAAM,CAAC;QACd,IAAI,EAAE,MAAM,CAAC;KACd,CAAC;IACF,KAAK,EAAE,gBAAgB,EAAE,CAAC;IAC1B,MAAM,EAAE;QACN,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,WAAW,CAAC,EAAE,OAAO,CAAC;KACvB,CAAC;CACH;AAED,8EAA8E;AAC9E,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,4DAA4D;AAC5D,MAAM,WAAW,kBAAkB;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,8DAA8D;IAC9D,WAAW,EAAE,OAAO,CAAC;CACtB;AAED,wFAAwF;AACxF,MAAM,WAAW,qBAAqB;IACpC,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,wEAAwE;IACxE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,IAAI,IAAI,OAAO,CAAC;QACd,KAAK,EAAE,gBAAgB,EAAE,CAAC;QAC1B,MAAM,EAAE,kBAAkB,CAAC;KAC5B,CAAC,CAAC;IACH,GAAG,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,GAAG,SAAS,CAAC,CAAC;IACzD,kBAAkB,IAAI,OAAO,CAAC;QAC5B,IAAI,EAAE,eAAe,EAAE,CAAC;QACxB,WAAW,CAAC,EAAE,MAAM,CAAC;KACtB,CAAC,CAAC;IACH,KAAK,CAAC,KAAK,CAAC,EAAE,qBAAqB,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;IAC/D,mDAAmD;IACnD,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,cAAc;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,cAAc,EAAE,MAAM,CAAC;IACvB;;;OAGG;IACH,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB;;;;OAIG;IACH,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B;;;;;;;;OAQG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED;;;;;;GAMG;AACH,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,cAAc,GAAG,cAAc,CAUvE;AAED;;GAEG;AACH,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAMlE;AAED,yDAAyD;AACzD,wBAAgB,8BAA8B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAKrE;AAED;;;GAGG;AACH,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CASnE"}
package/dist/evals.js CHANGED
@@ -44,18 +44,33 @@
44
44
  * ],
45
45
  * });
46
46
  * ```
47
+ *
48
+ * ## Severity: gates and scores
49
+ *
50
+ * Every assertion returns a chainable handle, so severity rides on the
51
+ * assertion instead of a separate thresholds map. Gates are hard (a miss fails
52
+ * the eval and `eval` exits non-zero); soft assertions are tracked scores that
53
+ * mark the eval `scored` and only fail the build under `--strict`:
54
+ *
55
+ * ```ts
56
+ * t.succeeded(); // gate (default)
57
+ * t.calledTool("get_weather").soft(); // tracked, no gate
58
+ * t.check(t.reply, similarity("Sunny, 72F")).atLeast(0.8);
59
+ * t.judge.closedQA("cites a source").atLeast(0.6); // LLM-graded
60
+ * ```
61
+ *
62
+ * Matchers narrow tool assertions (`{ input, output, status, count }` accept a
63
+ * literal, a RegExp, or a predicate), and each turn returned by `t.send(...)`
64
+ * carries the same vocabulary scoped to that turn.
47
65
  */
48
66
  import { normalizePositiveInt } from "./internal/bounded-int.js";
49
67
  import { brandDefinition } from "./internal/brand.js";
50
- export function includes(value) {
51
- return { kind: "includes", value };
52
- }
53
- export function equals(value) {
54
- return { kind: "equals", value };
55
- }
56
- export function satisfies(predicate, label) {
57
- return { kind: "satisfies", predicate, label };
58
- }
68
+ export { computeVerdict, verdictFailsRun } from "./evals/assertions.js";
69
+ export { equals, includes, matches, normalizedSimilarity, satisfies, similarity, } from "./evals/expect.js";
70
+ // A custom judge built on `t.judge.model` embeds the same attacker-controlled
71
+ // agent output the built-in graders do, so it needs the same two protections.
72
+ export { EVAL_JUDGE_INJECTION_GUARD, fenceUntrusted, } from "./evals/judge.js";
73
+ export { deriveRunFacts } from "./evals/run-facts.js";
59
74
  /** Hard ceiling for {@link EvalConfig.iterations} / case `iterations`. */
60
75
  export const EVAL_MAX_ITERATIONS = 100;
61
76
  /**
@@ -122,16 +137,22 @@ export function expandEvalDefinition(fileId, definition) {
122
137
  var _a;
123
138
  if (definition.cases !== undefined) {
124
139
  return definition.cases.flatMap((datapoint) => {
125
- var _a, _b, _c, _d, _e;
140
+ var _a, _b, _c, _d, _e, _f;
126
141
  const description = (_a = datapoint.description) !== null && _a !== void 0 ? _a : definition.description;
127
142
  const timeoutMs = (_b = datapoint.timeoutMs) !== null && _b !== void 0 ? _b : definition.timeoutMs;
128
143
  const iterations = normalizeEvalIterations((_d = (_c = datapoint.iterations) !== null && _c !== void 0 ? _c : definition.iterations) !== null && _d !== void 0 ? _d : 1);
144
+ const metadata = datapoint.metadata === undefined && definition.metadata === undefined
145
+ ? undefined
146
+ : Object.assign(Object.assign({}, definition.metadata), datapoint.metadata);
129
147
  return expandIterations({
130
148
  baseId: `${fileId}/${datapoint.id}`,
131
149
  description,
132
150
  tags: (_e = datapoint.tags) !== null && _e !== void 0 ? _e : definition.tags,
133
151
  timeoutMs,
134
152
  iterations,
153
+ metadata,
154
+ judge: (_f = datapoint.judge) !== null && _f !== void 0 ? _f : definition.judge,
155
+ reporters: mergeReporters(definition.reporters, datapoint.reporters),
135
156
  test: datapoint.test,
136
157
  });
137
158
  });
@@ -146,9 +167,22 @@ export function expandEvalDefinition(fileId, definition) {
146
167
  tags: definition.tags,
147
168
  timeoutMs: definition.timeoutMs,
148
169
  iterations,
170
+ metadata: definition.metadata,
171
+ judge: definition.judge,
172
+ reporters: definition.reporters,
149
173
  test: definition.test,
150
174
  });
151
175
  }
176
+ /** Union file-level and case-level reporters, keeping each instance once. */
177
+ function mergeReporters(fileLevel, caseLevel) {
178
+ if (fileLevel === undefined) {
179
+ return caseLevel;
180
+ }
181
+ if (caseLevel === undefined) {
182
+ return fileLevel;
183
+ }
184
+ return [...new Set([...fileLevel, ...caseLevel])];
185
+ }
152
186
  function expandIterations(args) {
153
187
  const out = [];
154
188
  for (let iteration = 1; iteration <= args.iterations; iteration++) {
@@ -158,7 +192,7 @@ function expandIterations(args) {
158
192
  : `${args.description} (iteration ${iteration}/${args.iterations})`;
159
193
  out.push({
160
194
  id,
161
- definition: brandDefinition("eval", Object.assign(Object.assign(Object.assign(Object.assign({}, (description !== undefined ? { description } : {})), (args.tags !== undefined ? { tags: args.tags } : {})), (args.timeoutMs !== undefined ? { timeoutMs: args.timeoutMs } : {})), { iterations: args.iterations, iteration, test: args.test })),
195
+ definition: brandDefinition("eval", Object.assign(Object.assign(Object.assign(Object.assign({}, (description !== undefined ? { description } : {})), (args.tags !== undefined ? { tags: args.tags } : {})), (args.timeoutMs !== undefined ? { timeoutMs: args.timeoutMs } : {})), { metadata: args.metadata, judge: args.judge, reporters: args.reporters, iterations: args.iterations, iteration, test: args.test })),
162
196
  });
163
197
  }
164
198
  return out;
@@ -222,7 +256,7 @@ export const EVAL_MAX_PLAYGROUND_RUNS = 500;
222
256
  */
223
257
  export function defineEvalConfig(config) {
224
258
  var _a;
225
- return Object.assign(Object.assign({}, (config.timeoutMs === undefined ? {} : { timeoutMs: config.timeoutMs })), { maxConcurrency: normalizeEvalMaxConcurrency(config.maxConcurrency), maxPlaygroundRuns: normalizeEvalMaxPlaygroundRuns((_a = config.maxPlaygroundRuns) !== null && _a !== void 0 ? _a : EVAL_DEFAULT_MAX_PLAYGROUND_RUNS) });
259
+ return Object.assign(Object.assign({}, (config.timeoutMs === undefined ? {} : { timeoutMs: config.timeoutMs })), { maxConcurrency: normalizeEvalMaxConcurrency(config.maxConcurrency), maxPlaygroundRuns: normalizeEvalMaxPlaygroundRuns((_a = config.maxPlaygroundRuns) !== null && _a !== void 0 ? _a : EVAL_DEFAULT_MAX_PLAYGROUND_RUNS), judge: config.judge, reporters: config.reporters });
226
260
  }
227
261
  /**
228
262
  * Validate and enforce the hard concurrency ceiling.
package/dist/index.d.ts CHANGED
@@ -35,7 +35,7 @@ export type { StorageABTable, StorageConfig, StorageContext, StorageDefinition,
35
35
  export { defineStorage, STORAGE_KEY_ROOT, storageKeys, } from "./storage.js";
36
36
  export { defineTool } from "./tools.js";
37
37
  export type * from "./types.js";
38
- export { DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, } from "./types.js";
38
+ export { cursorAccountServersHasMixedWildcard, DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, normalizeCursorAccountServers, } from "./types.js";
39
39
  /**
40
40
  * Runtime config for an agent — authored at `agent/agent.ts` (the root
41
41
  * agent) or `agent/subagents/<id>/agent.ts` (a subagent).
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAKH,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,yBAAyB,CAAC;AAEjC,OAAO,KAAK,EACV,WAAW,EACX,qBAAqB,EACrB,YAAY,EACZ,gBAAgB,EAChB,kBAAkB,EAClB,sBAAsB,EACtB,YAAY,EACb,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAEnC,OAAO,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,aAAa,EACb,GAAG,EACH,WAAW,EACX,QAAQ,EACR,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,KAAK,EACL,IAAI,EACJ,GAAG,GACJ,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,YAAY,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AACvE,OAAO,EACL,iBAAiB,EACjB,mBAAmB,EACnB,iBAAiB,EACjB,gBAAgB,EAChB,iBAAiB,EACjB,WAAW,EACX,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,YAAY,EACV,4BAA4B,EAC5B,6BAA6B,EAC7B,sBAAsB,EACtB,uBAAuB,EACvB,6BAA6B,GAC9B,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,2BAA2B,EAC3B,uBAAuB,GACxB,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,oBAAoB,GACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,uBAAuB,EACvB,oBAAoB,GACrB,MAAM,+BAA+B,CAAC;AACvC,YAAY,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EACL,gBAAgB,EAChB,cAAc,EACd,4BAA4B,EAC5B,mBAAmB,EACnB,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,8BAA8B,CAAC;AACtC,YAAY,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,YAAY,EACV,kBAAkB,EAClB,mBAAmB,EACnB,mBAAmB,EACnB,kBAAkB,EAClB,eAAe,EACf,YAAY,EACZ,kBAAkB,EAClB,oBAAoB,EACpB,cAAc,EACd,cAAc,EACd,oBAAoB,GACrB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC1C,YAAY,EACV,cAAc,EACd,aAAa,EACb,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,iBAAiB,EACjB,aAAa,EACb,oBAAoB,GACrB,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,aAAa,EACb,gBAAgB,EAChB,WAAW,GACZ,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,mBAAmB,YAAY,CAAC;AAChC,OAAO,EACL,aAAa,EACb,2BAA2B,EAC3B,wBAAwB,EACxB,yBAAyB,EACzB,6BAA6B,EAC7B,uBAAuB,GACxB,MAAM,YAAY,CAAC;AAEpB;;;GAGG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,WAAW,GAAG,qBAAqB,CAEtE;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,kBAAkB,GACzB,sBAAsB,CAExB;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAsB,KAAK,CACzB,YAAY,EAAE,MAAM,GAAG,YAAY,EACnC,OAAO,GAAE,YAAY,GAAG;IACtB,SAAS,CAAC,EAAE,uBAAuB,CAAC;IACpC;;;;OAIG;IACH,IAAI,CAAC,EAAE,QAAQ,GAAG,OAAO,CAAC;CACtB,GACL,OAAO,CAAC,gBAAgB,CAAC,CAqB3B"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAKH,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,yBAAyB,CAAC;AAEjC,OAAO,KAAK,EACV,WAAW,EACX,qBAAqB,EACrB,YAAY,EACZ,gBAAgB,EAChB,kBAAkB,EAClB,sBAAsB,EACtB,YAAY,EACb,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAEnC,OAAO,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,aAAa,EACb,GAAG,EACH,WAAW,EACX,QAAQ,EACR,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,KAAK,EACL,IAAI,EACJ,GAAG,GACJ,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,YAAY,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AACvE,OAAO,EACL,iBAAiB,EACjB,mBAAmB,EACnB,iBAAiB,EACjB,gBAAgB,EAChB,iBAAiB,EACjB,WAAW,EACX,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,YAAY,EACV,4BAA4B,EAC5B,6BAA6B,EAC7B,sBAAsB,EACtB,uBAAuB,EACvB,6BAA6B,GAC9B,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,2BAA2B,EAC3B,uBAAuB,GACxB,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,oBAAoB,GACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,uBAAuB,EACvB,oBAAoB,GACrB,MAAM,+BAA+B,CAAC;AACvC,YAAY,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EACL,gBAAgB,EAChB,cAAc,EACd,4BAA4B,EAC5B,mBAAmB,EACnB,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,8BAA8B,CAAC;AACtC,YAAY,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,YAAY,EACV,kBAAkB,EAClB,mBAAmB,EACnB,mBAAmB,EACnB,kBAAkB,EAClB,eAAe,EACf,YAAY,EACZ,kBAAkB,EAClB,oBAAoB,EACpB,cAAc,EACd,cAAc,EACd,oBAAoB,GACrB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC1C,YAAY,EACV,cAAc,EACd,aAAa,EACb,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,iBAAiB,EACjB,aAAa,EACb,oBAAoB,GACrB,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,aAAa,EACb,gBAAgB,EAChB,WAAW,GACZ,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,mBAAmB,YAAY,CAAC;AAChC,OAAO,EACL,oCAAoC,EACpC,aAAa,EACb,2BAA2B,EAC3B,wBAAwB,EACxB,yBAAyB,EACzB,6BAA6B,EAC7B,uBAAuB,EACvB,6BAA6B,GAC9B,MAAM,YAAY,CAAC;AAEpB;;;GAGG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,WAAW,GAAG,qBAAqB,CAEtE;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,kBAAkB,GACzB,sBAAsB,CAExB;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAsB,KAAK,CACzB,YAAY,EAAE,MAAM,GAAG,YAAY,EACnC,OAAO,GAAE,YAAY,GAAG;IACtB,SAAS,CAAC,EAAE,uBAAuB,CAAC;IACpC;;;;OAIG;IACH,IAAI,CAAC,EAAE,QAAQ,GAAG,OAAO,CAAC;CACtB,GACL,OAAO,CAAC,gBAAgB,CAAC,CAqB3B"}
package/dist/index.js CHANGED
@@ -52,7 +52,7 @@ export { defineSchedule } from "./schedules.js";
52
52
  export { defineSkill } from "./skills.js";
53
53
  export { defineStorage, STORAGE_KEY_ROOT, storageKeys, } from "./storage.js";
54
54
  export { defineTool } from "./tools.js";
55
- export { DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, } from "./types.js";
55
+ export { cursorAccountServersHasMixedWildcard, DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, normalizeCursorAccountServers, } from "./types.js";
56
56
  /**
57
57
  * Runtime config for an agent — authored at `agent/agent.ts` (the root
58
58
  * agent) or `agent/subagents/<id>/agent.ts` (a subagent).
@@ -0,0 +1,10 @@
1
+ /**
2
+ * Built-in tools (`AgentConfig.builtinTools`): framework-provided
3
+ * model-facing tools, materialized as ordinary server tools at discovery
4
+ * time so turns, direct calls, `info`, and the playground treat them
5
+ * exactly like authored tools.
6
+ */
7
+ import type { BuiltinToolsConfig, DiscoveredTool } from "../../types.js";
8
+ /** Resolve the enabled built-in tools, in stable catalog order. */
9
+ export declare function builtinTools(config: BuiltinToolsConfig | undefined): DiscoveredTool[];
10
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/internal/builtin-tools/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EACV,kBAAkB,EAClB,cAAc,EAEf,MAAM,gBAAgB,CAAC;AAIxB,mEAAmE;AACnE,wBAAgB,YAAY,CAC1B,MAAM,EAAE,kBAAkB,GAAG,SAAS,GACrC,cAAc,EAAE,CAMlB"}
@@ -0,0 +1,27 @@
1
+ /**
2
+ * Built-in tools (`AgentConfig.builtinTools`): framework-provided
3
+ * model-facing tools, materialized as ordinary server tools at discovery
4
+ * time so turns, direct calls, `info`, and the playground treat them
5
+ * exactly like authored tools.
6
+ */
7
+ import { toolInputJsonSchema } from "../json-schema.js";
8
+ import { reminderTools } from "./reminders.js";
9
+ /** Resolve the enabled built-in tools, in stable catalog order. */
10
+ export function builtinTools(config) {
11
+ const tools = [];
12
+ if ((config === null || config === void 0 ? void 0 : config.reminders) === true) {
13
+ tools.push(...toDiscovered(reminderTools));
14
+ }
15
+ return tools;
16
+ }
17
+ function toDiscovered(byName) {
18
+ return Object.entries(byName).map(([name, definition]) => ({
19
+ name,
20
+ description: definition.description,
21
+ inputSchema: toolInputJsonSchema(definition.inputSchema),
22
+ execution: "server",
23
+ needsApproval: definition.needsApproval === true ||
24
+ typeof definition.needsApproval === "function",
25
+ definition,
26
+ }));
27
+ }
@@ -0,0 +1,9 @@
1
+ /**
2
+ * Built-in reminder tools (`builtinTools: { reminders: true }`): durable
3
+ * wakes bound to the current conversation over `host.reminders`. The
4
+ * framework's reminder fire prompt tells the model to call
5
+ * `reminders_cancel` by name, so these tool names must not change.
6
+ */
7
+ import type { ToolDefinition } from "../../types.js";
8
+ export declare const reminderTools: Record<string, ToolDefinition>;
9
+ //# sourceMappingURL=reminders.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"reminders.d.ts","sourceRoot":"","sources":["../../../src/internal/builtin-tools/reminders.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAIH,OAAO,KAAK,EAIV,cAAc,EACf,MAAM,gBAAgB,CAAC;AA2IxB,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAIxD,CAAC"}
@@ -0,0 +1,138 @@
1
+ /**
2
+ * Built-in reminder tools (`builtinTools: { reminders: true }`): durable
3
+ * wakes bound to the current conversation over `host.reminders`. The
4
+ * framework's reminder fire prompt tells the model to call
5
+ * `reminders_cancel` by name, so these tool names must not change.
6
+ */
7
+ var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
8
+ function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
9
+ return new (P || (P = Promise))(function (resolve, reject) {
10
+ function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
11
+ function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
12
+ function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
13
+ step((generator = generator.apply(thisArg, _arguments || [])).next());
14
+ });
15
+ };
16
+ import { z } from "zod";
17
+ import { defineTool } from "../../tools.js";
18
+ import { parseDuration } from "../duration.js";
19
+ /** Floor for model-armed recurring wakes: each fire spends a model turn. */
20
+ const MIN_EVERY_MS = 60000;
21
+ function api(ctx) {
22
+ const reminders = ctx.host.reminders;
23
+ if (reminders === undefined) {
24
+ throw new Error("Reminders are disabled on this host.");
25
+ }
26
+ return reminders;
27
+ }
28
+ function conversation(ctx) {
29
+ const token = ctx.session.continuationKey;
30
+ if (token == null) {
31
+ throw new Error("This session is not bound to a followable conversation; reminders cannot wake it.");
32
+ }
33
+ return { channelId: ctx.session.channelId, continuationToken: token };
34
+ }
35
+ /** Reminders visible to this conversation (tokens are channel-local). */
36
+ function listMine(ctx, status) {
37
+ return __awaiter(this, void 0, void 0, function* () {
38
+ const conv = conversation(ctx);
39
+ const all = yield api(ctx).list({
40
+ status,
41
+ continuationToken: conv.continuationToken,
42
+ });
43
+ return all.filter((r) => r.channelId === conv.channelId);
44
+ });
45
+ }
46
+ function toResult(r) {
47
+ return {
48
+ reminderId: r.id,
49
+ purpose: r.purpose,
50
+ schedule: `${r.scheduleKind} ${r.scheduleValue}`,
51
+ nextFireAt: r.nextFireAt,
52
+ until: r.untilPrompt,
53
+ };
54
+ }
55
+ const create = defineTool({
56
+ description: "Set a reminder for yourself in this conversation. When it fires, your prompt returns here as a follow-up turn. Use it to re-check state that needs time to settle. One-shot (delay) or recurring (every).",
57
+ inputSchema: z.object({
58
+ prompt: z
59
+ .string()
60
+ .min(1)
61
+ .describe("What to do when you wake. Keep it generic ('re-check CI and report only if the status changed') and re-read live state on wake instead of replaying stale details."),
62
+ delay: z
63
+ .string()
64
+ .optional()
65
+ .describe('One-shot: wake once after this duration, e.g. "20m" or "2h". Provide exactly one of delay or every.'),
66
+ every: z
67
+ .string()
68
+ .optional()
69
+ .describe('Recurring: wake at this interval, e.g. "30m", until cancelled or the stop condition is met. Provide exactly one of delay or every.'),
70
+ purpose: z
71
+ .string()
72
+ .min(1)
73
+ .max(64)
74
+ .optional()
75
+ .describe('Short slug naming the reminder, e.g. "ci_recheck". Defaults to "follow_up".'),
76
+ until: z
77
+ .string()
78
+ .optional()
79
+ .describe("Recurring only: plain-language stop condition you will honor by calling reminders_cancel, e.g. 'Cancel once CI is green or the PR is merged.'"),
80
+ }),
81
+ execute(input, ctx) {
82
+ return __awaiter(this, void 0, void 0, function* () {
83
+ var _a;
84
+ if ((input.delay !== undefined) === (input.every !== undefined)) {
85
+ throw new Error("Provide exactly one of delay or every.");
86
+ }
87
+ if (input.every !== undefined &&
88
+ parseDuration(input.every) < MIN_EVERY_MS) {
89
+ throw new Error('Recurring reminders must use every >= "1m".');
90
+ }
91
+ const conv = conversation(ctx);
92
+ const reminder = yield api(ctx).create({
93
+ purpose: (_a = input.purpose) !== null && _a !== void 0 ? _a : "follow_up",
94
+ channelId: conv.channelId,
95
+ continuationToken: conv.continuationToken,
96
+ prompt: input.prompt,
97
+ delay: input.delay,
98
+ every: input.every,
99
+ until: input.until,
100
+ });
101
+ return toResult(reminder);
102
+ });
103
+ },
104
+ });
105
+ const list = defineTool({
106
+ description: "List the armed reminders bound to this conversation. Check before arming duplicates.",
107
+ inputSchema: z.object({}),
108
+ execute(_input, ctx) {
109
+ return __awaiter(this, void 0, void 0, function* () {
110
+ return { reminders: (yield listMine(ctx, "armed")).map(toResult) };
111
+ });
112
+ },
113
+ });
114
+ const cancel = defineTool({
115
+ description: "Cancel one of this conversation's reminders by id, the moment its stop condition is met or it is no longer needed.",
116
+ inputSchema: z.object({
117
+ id: z
118
+ .string()
119
+ .min(1)
120
+ .describe("Reminder id from reminders_create or reminders_list."),
121
+ }),
122
+ execute(input, ctx) {
123
+ return __awaiter(this, void 0, void 0, function* () {
124
+ var _a;
125
+ const mine = (yield listMine(ctx)).find((r) => r.id === input.id);
126
+ if (mine === undefined) {
127
+ throw new Error(`No reminder "${input.id}" in this conversation.`);
128
+ }
129
+ const cancelled = yield api(ctx).cancel(input.id);
130
+ return { reminderId: input.id, status: (_a = cancelled === null || cancelled === void 0 ? void 0 : cancelled.status) !== null && _a !== void 0 ? _a : "cancelled" };
131
+ });
132
+ },
133
+ });
134
+ export const reminderTools = {
135
+ reminders_create: create,
136
+ reminders_list: list,
137
+ reminders_cancel: cancel,
138
+ };