@assemblyline-agents/cli 0.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (147) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +19 -0
  3. package/dist/add.d.ts +23 -0
  4. package/dist/add.d.ts.map +1 -0
  5. package/dist/add.js +352 -0
  6. package/dist/add.js.map +1 -0
  7. package/dist/args.d.ts +14 -0
  8. package/dist/args.d.ts.map +1 -0
  9. package/dist/args.js +95 -0
  10. package/dist/args.js.map +1 -0
  11. package/dist/auth.d.ts +8 -0
  12. package/dist/auth.d.ts.map +1 -0
  13. package/dist/auth.js +74 -0
  14. package/dist/auth.js.map +1 -0
  15. package/dist/channels.d.ts +25 -0
  16. package/dist/channels.d.ts.map +1 -0
  17. package/dist/channels.js +76 -0
  18. package/dist/channels.js.map +1 -0
  19. package/dist/checkpoints.d.ts +3 -0
  20. package/dist/checkpoints.d.ts.map +1 -0
  21. package/dist/checkpoints.js +60 -0
  22. package/dist/checkpoints.js.map +1 -0
  23. package/dist/deploy-definition.d.ts +14 -0
  24. package/dist/deploy-definition.d.ts.map +1 -0
  25. package/dist/deploy-definition.js +92 -0
  26. package/dist/deploy-definition.js.map +1 -0
  27. package/dist/deploy-lifecycle.d.ts +17 -0
  28. package/dist/deploy-lifecycle.d.ts.map +1 -0
  29. package/dist/deploy-lifecycle.js +25 -0
  30. package/dist/deploy-lifecycle.js.map +1 -0
  31. package/dist/deploy.d.ts +20 -0
  32. package/dist/deploy.d.ts.map +1 -0
  33. package/dist/deploy.js +462 -0
  34. package/dist/deploy.js.map +1 -0
  35. package/dist/dev.d.ts +13 -0
  36. package/dist/dev.d.ts.map +1 -0
  37. package/dist/dev.js +157 -0
  38. package/dist/dev.js.map +1 -0
  39. package/dist/eval-assertions.d.ts +5 -0
  40. package/dist/eval-assertions.d.ts.map +1 -0
  41. package/dist/eval-assertions.js +99 -0
  42. package/dist/eval-assertions.js.map +1 -0
  43. package/dist/eval-case-runner.d.ts +38 -0
  44. package/dist/eval-case-runner.d.ts.map +1 -0
  45. package/dist/eval-case-runner.js +247 -0
  46. package/dist/eval-case-runner.js.map +1 -0
  47. package/dist/eval-cases.d.ts +6 -0
  48. package/dist/eval-cases.d.ts.map +1 -0
  49. package/dist/eval-cases.js +151 -0
  50. package/dist/eval-cases.js.map +1 -0
  51. package/dist/eval-comparison.d.ts +8 -0
  52. package/dist/eval-comparison.d.ts.map +1 -0
  53. package/dist/eval-comparison.js +90 -0
  54. package/dist/eval-comparison.js.map +1 -0
  55. package/dist/eval-custom.d.ts +15 -0
  56. package/dist/eval-custom.d.ts.map +1 -0
  57. package/dist/eval-custom.js +118 -0
  58. package/dist/eval-custom.js.map +1 -0
  59. package/dist/eval-experiment.d.ts +19 -0
  60. package/dist/eval-experiment.d.ts.map +1 -0
  61. package/dist/eval-experiment.js +122 -0
  62. package/dist/eval-experiment.js.map +1 -0
  63. package/dist/eval-fixtures.d.ts +17 -0
  64. package/dist/eval-fixtures.d.ts.map +1 -0
  65. package/dist/eval-fixtures.js +59 -0
  66. package/dist/eval-fixtures.js.map +1 -0
  67. package/dist/eval-gateway-host.d.ts +34 -0
  68. package/dist/eval-gateway-host.d.ts.map +1 -0
  69. package/dist/eval-gateway-host.js +127 -0
  70. package/dist/eval-gateway-host.js.map +1 -0
  71. package/dist/eval-judge.d.ts +35 -0
  72. package/dist/eval-judge.d.ts.map +1 -0
  73. package/dist/eval-judge.js +105 -0
  74. package/dist/eval-judge.js.map +1 -0
  75. package/dist/eval-mocks.d.ts +10 -0
  76. package/dist/eval-mocks.d.ts.map +1 -0
  77. package/dist/eval-mocks.js +57 -0
  78. package/dist/eval-mocks.js.map +1 -0
  79. package/dist/eval-report.d.ts +7 -0
  80. package/dist/eval-report.d.ts.map +1 -0
  81. package/dist/eval-report.js +89 -0
  82. package/dist/eval-report.js.map +1 -0
  83. package/dist/eval-retry.d.ts +10 -0
  84. package/dist/eval-retry.d.ts.map +1 -0
  85. package/dist/eval-retry.js +37 -0
  86. package/dist/eval-retry.js.map +1 -0
  87. package/dist/eval-trajectory.d.ts +7 -0
  88. package/dist/eval-trajectory.d.ts.map +1 -0
  89. package/dist/eval-trajectory.js +92 -0
  90. package/dist/eval-trajectory.js.map +1 -0
  91. package/dist/eval-turns.d.ts +66 -0
  92. package/dist/eval-turns.d.ts.map +1 -0
  93. package/dist/eval-turns.js +187 -0
  94. package/dist/eval-turns.js.map +1 -0
  95. package/dist/eval-types.d.ts +271 -0
  96. package/dist/eval-types.d.ts.map +1 -0
  97. package/dist/eval-types.js +4 -0
  98. package/dist/eval-types.js.map +1 -0
  99. package/dist/eval.d.ts +26 -0
  100. package/dist/eval.d.ts.map +1 -0
  101. package/dist/eval.js +218 -0
  102. package/dist/eval.js.map +1 -0
  103. package/dist/help.d.ts +12 -0
  104. package/dist/help.d.ts.map +1 -0
  105. package/dist/help.js +323 -0
  106. package/dist/help.js.map +1 -0
  107. package/dist/hosts.d.ts +3 -0
  108. package/dist/hosts.d.ts.map +1 -0
  109. package/dist/hosts.js +51 -0
  110. package/dist/hosts.js.map +1 -0
  111. package/dist/index.d.ts +3 -0
  112. package/dist/index.d.ts.map +1 -0
  113. package/dist/index.js +318 -0
  114. package/dist/index.js.map +1 -0
  115. package/dist/init.d.ts +13 -0
  116. package/dist/init.d.ts.map +1 -0
  117. package/dist/init.js +79 -0
  118. package/dist/init.js.map +1 -0
  119. package/dist/legacy-bin.d.ts +3 -0
  120. package/dist/legacy-bin.d.ts.map +1 -0
  121. package/dist/legacy-bin.js +8 -0
  122. package/dist/legacy-bin.js.map +1 -0
  123. package/dist/models.d.ts +5 -0
  124. package/dist/models.d.ts.map +1 -0
  125. package/dist/models.js +20 -0
  126. package/dist/models.js.map +1 -0
  127. package/dist/operator-client.d.ts +7 -0
  128. package/dist/operator-client.d.ts.map +1 -0
  129. package/dist/operator-client.js +62 -0
  130. package/dist/operator-client.js.map +1 -0
  131. package/dist/pm.d.ts +20 -0
  132. package/dist/pm.d.ts.map +1 -0
  133. package/dist/pm.js +70 -0
  134. package/dist/pm.js.map +1 -0
  135. package/dist/project-env.d.ts +8 -0
  136. package/dist/project-env.d.ts.map +1 -0
  137. package/dist/project-env.js +51 -0
  138. package/dist/project-env.js.map +1 -0
  139. package/dist/secrets.d.ts +3 -0
  140. package/dist/secrets.d.ts.map +1 -0
  141. package/dist/secrets.js +48 -0
  142. package/dist/secrets.js.map +1 -0
  143. package/dist/state.d.ts +3 -0
  144. package/dist/state.d.ts.map +1 -0
  145. package/dist/state.js +85 -0
  146. package/dist/state.js.map +1 -0
  147. package/package.json +67 -0
package/dist/dev.js ADDED
@@ -0,0 +1,157 @@
1
+ import { watch } from "node:fs";
2
+ import { relative, resolve, sep } from "node:path";
3
+ import { buildAgent, validateAgentFolder } from "@assemblyline-agents/compiler";
4
+ import { listenNodeRuntime, withNodeModelHarnesses } from "@assemblyline-agents/node";
5
+ import { FileStateAdapter, LocalBlobAdapter, LocalSandboxAdapter, OpenEveRuntime, loadRuntimeBundle } from "@assemblyline-agents/runtime";
6
+ import { printIssues } from "./help.js";
7
+ const DEBOUNCE_MS = 200;
8
+ export async function runDevCommand(options) {
9
+ if (!options.watch) {
10
+ await runDevOnce(options);
11
+ return;
12
+ }
13
+ await runDevWatch(options);
14
+ }
15
+ async function runDevOnce(options) {
16
+ const artifact = buildAgent(devBuildOptions(options));
17
+ const runtimeOptions = loadRuntimeBundle(artifact.root);
18
+ const runtime = new OpenEveRuntime(withNodeModelHarnesses({
19
+ ...runtimeOptions,
20
+ state: new FileStateAdapter(resolve(artifact.root, "dev-state.json")),
21
+ blob: new LocalBlobAdapter(resolve(artifact.root, "blobs")),
22
+ sandbox: new LocalSandboxAdapter(resolve(artifact.root, "sandbox")),
23
+ devMode: true
24
+ }));
25
+ const result = await runtime.run({
26
+ message: options.message ?? "Assembly Line dev check",
27
+ toolName: options.toolName ?? "echo",
28
+ approve: options.approve ?? false,
29
+ channel: "local"
30
+ });
31
+ console.log(`Assembly Line dev runtime ready for ${options.agentRoot}`);
32
+ console.log(`Run ${result.run.id} ${result.run.status}`);
33
+ console.log(result.response ?? "");
34
+ }
35
+ async function runDevWatch(options) {
36
+ const agentRoot = options.agentRoot;
37
+ const validation = validateAgentFolder(agentRoot, catalogOptions(options));
38
+ printIssues(validation.issues);
39
+ if (!validation.valid)
40
+ process.exit(1);
41
+ let artifact = buildAgent(devBuildOptions(options));
42
+ let server = await listenDevServer(artifact.root, options.port ?? 0);
43
+ const port = boundPort(server.server) ?? options.port ?? 0;
44
+ let debounceTimer;
45
+ let rebuilding = false;
46
+ let rerunRequested = false;
47
+ let closed = false;
48
+ const watcher = watch(agentRoot, { recursive: true }, (_eventType, fileName) => {
49
+ if (closed)
50
+ return;
51
+ if (fileName && shouldIgnoreChange(agentRoot, artifact.root, String(fileName)))
52
+ return;
53
+ if (debounceTimer)
54
+ clearTimeout(debounceTimer);
55
+ debounceTimer = setTimeout(() => {
56
+ void handleChange();
57
+ }, DEBOUNCE_MS);
58
+ });
59
+ console.log(`Assembly Line dev server listening at ${server.url}`);
60
+ console.log(`Watching ${agentRoot} for changes. Press Ctrl+C to stop.`);
61
+ async function handleChange() {
62
+ if (closed)
63
+ return;
64
+ if (rebuilding) {
65
+ rerunRequested = true;
66
+ return;
67
+ }
68
+ rebuilding = true;
69
+ try {
70
+ const revalidation = validateAgentFolder(agentRoot, catalogOptions(options));
71
+ if (!revalidation.valid) {
72
+ printIssues(revalidation.issues);
73
+ console.log("Agent is invalid; waiting for a valid agent before restarting.");
74
+ return;
75
+ }
76
+ const nextArtifact = buildAgent(devBuildOptions(options));
77
+ if (nextArtifact.manifest.agentRevision === artifact.manifest.agentRevision) {
78
+ console.log("Rebuilt: no manifest change; keeping the running server.");
79
+ return;
80
+ }
81
+ artifact = nextArtifact;
82
+ await closeServer(server.server);
83
+ server = await listenDevServer(artifact.root, port);
84
+ console.log(`Restarted on ${server.url} (revision ${artifact.manifest.agentRevision}).`);
85
+ }
86
+ catch (error) {
87
+ console.error(`Dev rebuild failed: ${error instanceof Error ? error.message : String(error)}`);
88
+ }
89
+ finally {
90
+ rebuilding = false;
91
+ if (rerunRequested) {
92
+ rerunRequested = false;
93
+ void handleChange();
94
+ }
95
+ }
96
+ }
97
+ await new Promise((resolveExit) => {
98
+ process.once("SIGINT", () => {
99
+ closed = true;
100
+ if (debounceTimer)
101
+ clearTimeout(debounceTimer);
102
+ watcher.close();
103
+ void closeServer(server.server).finally(() => {
104
+ console.log("Stopped dev server.");
105
+ resolveExit();
106
+ });
107
+ });
108
+ });
109
+ process.exit(0);
110
+ }
111
+ async function listenDevServer(artifactRoot, port) {
112
+ const runtimeOptions = loadRuntimeBundle(artifactRoot);
113
+ return listenNodeRuntime({
114
+ ...runtimeOptions,
115
+ state: new FileStateAdapter(resolve(artifactRoot, "dev-state.json")),
116
+ blob: new LocalBlobAdapter(resolve(artifactRoot, "blobs")),
117
+ sandbox: new LocalSandboxAdapter(resolve(artifactRoot, "sandbox")),
118
+ devMode: true,
119
+ port
120
+ });
121
+ }
122
+ function devBuildOptions(options) {
123
+ const buildOptions = { root: options.agentRoot };
124
+ if (options.outDir)
125
+ buildOptions.outDir = options.outDir;
126
+ return buildOptions;
127
+ }
128
+ function catalogOptions(options) {
129
+ return options.modelCatalog ? { modelCatalog: options.modelCatalog } : {};
130
+ }
131
+ function shouldIgnoreChange(agentRoot, artifactRoot, fileName) {
132
+ const absolute = resolve(agentRoot, fileName);
133
+ const insideArtifact = relative(artifactRoot, absolute);
134
+ if (insideArtifact === "" || (!insideArtifact.startsWith("..") && !insideArtifact.includes(`..${sep}`))) {
135
+ return true;
136
+ }
137
+ const segments = fileName.split(/[\\/]/u);
138
+ return segments.some((segment) => segment === ".openeve" || segment === "node_modules" || segment.startsWith("."));
139
+ }
140
+ function boundPort(server) {
141
+ const address = server.address();
142
+ if (address && typeof address === "object")
143
+ return address.port;
144
+ return undefined;
145
+ }
146
+ function closeServer(server) {
147
+ return new Promise((resolveClose) => {
148
+ if (!server.listening) {
149
+ resolveClose();
150
+ return;
151
+ }
152
+ server.close(() => resolveClose());
153
+ // Dev-only: drop keep-alive sockets so restart is immediate.
154
+ server.closeAllConnections?.();
155
+ });
156
+ }
157
+ //# sourceMappingURL=dev.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"dev.js","sourceRoot":"","sources":["../src/dev.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAkB,MAAM,SAAS,CAAC;AAChD,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,mBAAmB,EAAmC,MAAM,+BAA+B,CAAC;AACjH,OAAO,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,2BAA2B,CAAC;AACtF,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,cAAc,EAAE,iBAAiB,EAAE,MAAM,8BAA8B,CAAC;AAE1I,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AAaxC,MAAM,WAAW,GAAG,GAAG,CAAC;AAExB,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,OAA0B;IAC5D,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,UAAU,CAAC,OAAO,CAAC,CAAC;QAC1B,OAAO;IACT,CAAC;IACD,MAAM,WAAW,CAAC,OAAO,CAAC,CAAC;AAC7B,CAAC;AAED,KAAK,UAAU,UAAU,CAAC,OAA0B;IAClD,MAAM,QAAQ,GAAG,UAAU,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IACtD,MAAM,cAAc,GAAG,iBAAiB,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IACxD,MAAM,OAAO,GAAG,IAAI,cAAc,CAAC,sBAAsB,CAAC;QACxD,GAAG,cAAc;QACjB,KAAK,EAAE,IAAI,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,gBAAgB,CAAC,CAAC;QACrE,IAAI,EAAE,IAAI,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC3D,OAAO,EAAE,IAAI,mBAAmB,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QACnE,OAAO,EAAE,IAAI;KACd,CAAC,CAAC,CAAC;IACJ,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAC/B,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,yBAAyB;QACrD,QAAQ,EAAE,OAAO,CAAC,QAAQ,IAAI,MAAM;QACpC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,KAAK;QACjC,OAAO,EAAE,OAAO;KACjB,CAAC,CAAC;IACH,OAAO,CAAC,GAAG,CAAC,uCAAuC,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC;IACxE,OAAO,CAAC,GAAG,CAAC,OAAO,MAAM,CAAC,GAAG,CAAC,EAAE,IAAI,MAAM,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;IACzD,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC;AACrC,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,OAA0B;IACnD,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC;IACpC,MAAM,UAAU,GAAG,mBAAmB,CAAC,SAAS,EAAE,cAAc,CAAC,OAAO,CAAC,CAAC,CAAC;IAC3E,WAAW,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;IAC/B,IAAI,CAAC,UAAU,CAAC,KAAK;QAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAEvC,IAAI,QAAQ,GAAG,UAAU,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;IACpD,IAAI,MAAM,GAAG,MAAM,eAAe,CAAC,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC;IACrE,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,OAAO,CAAC,IAAI,IAAI,CAAC,CAAC;IAE3D,IAAI,aAAyC,CAAC;IAC9C,IAAI,UAAU,GAAG,KAAK,CAAC;IACvB,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,IAAI,MAAM,GAAG,KAAK,CAAC;IAEnB,MAAM,OAAO,GAAc,KAAK,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,EAAE,CAAC,UAAU,EAAE,QAAQ,EAAE,EAAE;QACxF,IAAI,MAAM;YAAE,OAAO;QACnB,IAAI,QAAQ,IAAI,kBAAkB,CAAC,SAAS,EAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC;YAAE,OAAO;QACvF,IAAI,aAAa;YAAE,YAAY,CAAC,aAAa,CAAC,CAAC;QAC/C,aAAa,GAAG,UAAU,CAAC,GAAG,EAAE;YAC9B,KAAK,YAAY,EAAE,CAAC;QACtB,CAAC,EAAE,WAAW,CAAC,CAAC;IAClB,CAAC,CAAC,CAAC;IAEH,OAAO,CAAC,GAAG,CAAC,yCAAyC,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;IACnE,OAAO,CAAC,GAAG,CAAC,YAAY,SAAS,qCAAqC,CAAC,CAAC;IAExE,KAAK,UAAU,YAAY;QACzB,IAAI,MAAM;YAAE,OAAO;QACnB,IAAI,UAAU,EAAE,CAAC;YACf,cAAc,GAAG,IAAI,CAAC;YACtB,OAAO;QACT,CAAC;QACD,UAAU,GAAG,IAAI,CAAC;QAClB,IAAI,CAAC;YACH,MAAM,YAAY,GAAG,mBAAmB,CAAC,SAAS,EAAE,cAAc,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7E,IAAI,CAAC,YAAY,CAAC,KAAK,EAAE,CAAC;gBACxB,WAAW,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;gBACjC,OAAO,CAAC,GAAG,CAAC,gEAAgE,CAAC,CAAC;gBAC9E,OAAO;YACT,CAAC;YACD,MAAM,YAAY,GAAG,UAAU,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;YAC1D,IAAI,YAAY,CAAC,QAAQ,CAAC,aAAa,KAAK,QAAQ,CAAC,QAAQ,CAAC,aAAa,EAAE,CAAC;gBAC5E,OAAO,CAAC,GAAG,CAAC,0DAA0D,CAAC,CAAC;gBACxE,OAAO;YACT,CAAC;YACD,QAAQ,GAAG,YAAY,CAAC;YACxB,MAAM,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YACjC,MAAM,GAAG,MAAM,eAAe,CAAC,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACpD,OAAO,CAAC,GAAG,CAAC,gBAAgB,MAAM,CAAC,GAAG,cAAc,QAAQ,CAAC,QAAQ,CAAC,aAAa,IAAI,CAAC,CAAC;QAC3F,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,uBAAuB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACjG,CAAC;gBAAS,CAAC;YACT,UAAU,GAAG,KAAK,CAAC;YACnB,IAAI,cAAc,EAAE,CAAC;gBACnB,cAAc,GAAG,KAAK,CAAC;gBACvB,KAAK,YAAY,EAAE,CAAC;YACtB,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,IAAI,OAAO,CAAO,CAAC,WAAW,EAAE,EAAE;QACtC,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,EAAE;YAC1B,MAAM,GAAG,IAAI,CAAC;YACd,IAAI,aAAa;gBAAE,YAAY,CAAC,aAAa,CAAC,CAAC;YAC/C,OAAO,CAAC,KAAK,EAAE,CAAC;YAChB,KAAK,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE;gBAC3C,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC;gBACnC,WAAW,EAAE,CAAC;YAChB,CAAC,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IACH,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC;AAED,KAAK,UAAU,eAAe,CAAC,YAAoB,EAAE,IAAY;IAC/D,MAAM,cAAc,GAAG,iBAAiB,CAAC,YAAY,CAAC,CAAC;IACvD,OAAO,iBAAiB,CAAC;QACvB,GAAG,cAAc;QACjB,KAAK,EAAE,IAAI,gBAAgB,CAAC,OAAO,CAAC,YAAY,EAAE,gBAAgB,CAAC,CAAC;QACpE,IAAI,EAAE,IAAI,gBAAgB,CAAC,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;QAC1D,OAAO,EAAE,IAAI,mBAAmB,CAAC,OAAO,CAAC,YAAY,EAAE,SAAS,CAAC,CAAC;QAClE,OAAO,EAAE,IAAI;QACb,IAAI;KACL,CAAC,CAAC;AACL,CAAC;AAED,SAAS,eAAe,CAAC,OAA0B;IACjD,MAAM,YAAY,GAAsC,EAAE,IAAI,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC;IACpF,IAAI,OAAO,CAAC,MAAM;QAAE,YAAY,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IACzD,OAAO,YAAY,CAAC;AACtB,CAAC;AAED,SAAS,cAAc,CAAC,OAA0B;IAChD,OAAO,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,OAAO,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;AAC5E,CAAC;AAED,SAAS,kBAAkB,CAAC,SAAiB,EAAE,YAAoB,EAAE,QAAgB;IACnF,MAAM,QAAQ,GAAG,OAAO,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAC9C,MAAM,cAAc,GAAG,QAAQ,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;IACxD,IAAI,cAAc,KAAK,EAAE,IAAI,CAAC,CAAC,cAAc,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC;QACxG,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC1C,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,KAAK,UAAU,IAAI,OAAO,KAAK,cAAc,IAAI,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC;AACrH,CAAC;AAED,SAAS,SAAS,CAAC,MAAc;IAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,EAAE,CAAC;IACjC,IAAI,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAC,IAAI,CAAC;IAChE,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,WAAW,CAAC,MAAc;IACjC,OAAO,IAAI,OAAO,CAAC,CAAC,YAAY,EAAE,EAAE;QAClC,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;YACtB,YAAY,EAAE,CAAC;YACf,OAAO;QACT,CAAC;QACD,MAAM,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,YAAY,EAAE,CAAC,CAAC;QACnC,6DAA6D;QAC7D,MAAM,CAAC,mBAAmB,EAAE,EAAE,CAAC;IACjC,CAAC,CAAC,CAAC;AACL,CAAC"}
@@ -0,0 +1,5 @@
1
+ import type { EvalActual, EvalCase, EvalDeterministicExpectations } from "./eval-types.js";
2
+ export declare function evaluateCase(evalCase: EvalCase, actual: EvalActual): string[];
3
+ /** Deterministic assertion pass shared by whole-case and per-turn expectations. */
4
+ export declare function evaluateExpectations(expect: EvalDeterministicExpectations, actual: EvalActual): string[];
5
+ //# sourceMappingURL=eval-assertions.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-assertions.d.ts","sourceRoot":"","sources":["../src/eval-assertions.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,QAAQ,EAAE,6BAA6B,EAAE,MAAM,iBAAiB,CAAC;AAE3F,wBAAgB,YAAY,CAAC,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,UAAU,GAAG,MAAM,EAAE,CAE7E;AAED,mFAAmF;AACnF,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,6BAA6B,EAAE,MAAM,EAAE,UAAU,GAAG,MAAM,EAAE,CAqCxG"}
@@ -0,0 +1,99 @@
1
+ import { isDeepStrictEqual } from "node:util";
2
+ import { validateAgainstJsonSchema } from "@assemblyline-agents/core";
3
+ import { evaluateTrajectory } from "./eval-trajectory.js";
4
+ export function evaluateCase(evalCase, actual) {
5
+ return evaluateExpectations(evalCase.expect, actual);
6
+ }
7
+ /** Deterministic assertion pass shared by whole-case and per-turn expectations. */
8
+ export function evaluateExpectations(expect, actual) {
9
+ const failures = [];
10
+ if (expect.status && actual.status !== expect.status) {
11
+ failures.push(`expected status ${expect.status}, got ${actual.status}`);
12
+ }
13
+ const output = expect.output;
14
+ if (output?.contains) {
15
+ for (const fragment of output.contains) {
16
+ if (!actual.response.includes(fragment))
17
+ failures.push(`output did not contain ${JSON.stringify(fragment)}`);
18
+ }
19
+ }
20
+ if (output?.matches) {
21
+ try {
22
+ if (!new RegExp(output.matches, "u").test(actual.response))
23
+ failures.push(`output did not match /${output.matches}/u`);
24
+ }
25
+ catch (error) {
26
+ failures.push(`invalid output.matches regex: ${error instanceof Error ? error.message : String(error)}`);
27
+ }
28
+ }
29
+ if (output?.equals !== undefined && actual.response !== output.equals) {
30
+ failures.push(`expected exact output ${JSON.stringify(output.equals)}, got ${JSON.stringify(actual.response)}`);
31
+ }
32
+ if (output?.json)
33
+ evaluateJson(output.json, actual, failures);
34
+ for (const tool of expect.toolsCalled ?? []) {
35
+ if (!actual.toolsCalled.includes(tool))
36
+ failures.push(`expected tool ${tool} to be called`);
37
+ }
38
+ for (const tool of expect.toolsNotCalled ?? []) {
39
+ if (actual.toolsCalled.includes(tool))
40
+ failures.push(`expected tool ${tool} not to be called`);
41
+ }
42
+ if (expect.trajectory) {
43
+ failures.push(...evaluateTrajectory(expect.trajectory, actual.toolCalls ?? []));
44
+ }
45
+ if (expect.maxCostUsd !== undefined && !actual.costComplete) {
46
+ failures.push("maxCostUsd could not be evaluated because one or more provider cash receipts are unavailable");
47
+ }
48
+ else if (expect.maxCostUsd !== undefined && actual.costUsd > expect.maxCostUsd) {
49
+ failures.push(`cost $${actual.costUsd.toFixed(6)} exceeded $${expect.maxCostUsd.toFixed(6)}`);
50
+ }
51
+ return failures;
52
+ }
53
+ function evaluateJson(expectation, actual, failures) {
54
+ let parsed = actual.output;
55
+ let parseError;
56
+ if (parsed === undefined) {
57
+ try {
58
+ parsed = JSON.parse(actual.response);
59
+ }
60
+ catch (error) {
61
+ parseError = error instanceof Error ? error.message : String(error);
62
+ }
63
+ }
64
+ if (expectation.valid === true && parseError)
65
+ failures.push(`output was not valid JSON: ${parseError}`);
66
+ if (expectation.valid === false && !parseError)
67
+ failures.push("expected output not to be valid JSON");
68
+ if (parsed === undefined) {
69
+ if (expectation.jsonSchema !== undefined || expectation.fields !== undefined)
70
+ failures.push("JSON assertions require valid JSON output");
71
+ return;
72
+ }
73
+ if (expectation.jsonSchema !== undefined) {
74
+ const validation = validateAgainstJsonSchema(parsed, expectation.jsonSchema);
75
+ if (!validation.valid)
76
+ failures.push(`output schema validation failed: ${validation.errors.join("; ")}`);
77
+ }
78
+ for (const [path, expected] of Object.entries(expectation.fields ?? {})) {
79
+ const found = valueAtPath(parsed, path);
80
+ if (!found.exists)
81
+ failures.push(`JSON field ${path} was missing`);
82
+ else if (!isDeepStrictEqual(found.value, expected)) {
83
+ failures.push(`JSON field ${path} expected ${JSON.stringify(expected)}, got ${JSON.stringify(found.value)}`);
84
+ }
85
+ }
86
+ }
87
+ function valueAtPath(value, path) {
88
+ const segments = path.startsWith("/")
89
+ ? path.slice(1).split("/").filter(Boolean).map((segment) => segment.replace(/~1/gu, "/").replace(/~0/gu, "~"))
90
+ : path.split(".").filter(Boolean);
91
+ let current = value;
92
+ for (const segment of segments) {
93
+ if (!current || typeof current !== "object" || !(segment in current))
94
+ return { exists: false };
95
+ current = current[segment];
96
+ }
97
+ return { exists: true, value: current };
98
+ }
99
+ //# sourceMappingURL=eval-assertions.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-assertions.js","sourceRoot":"","sources":["../src/eval-assertions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAC9C,OAAO,EAAE,yBAAyB,EAAmC,MAAM,2BAA2B,CAAC;AACvG,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAG1D,MAAM,UAAU,YAAY,CAAC,QAAkB,EAAE,MAAkB;IACjE,OAAO,oBAAoB,CAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;AACvD,CAAC;AAED,mFAAmF;AACnF,MAAM,UAAU,oBAAoB,CAAC,MAAqC,EAAE,MAAkB;IAC5F,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,EAAE,CAAC;QACrD,QAAQ,CAAC,IAAI,CAAC,mBAAmB,MAAM,CAAC,MAAM,SAAS,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC1E,CAAC;IACD,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;IAC7B,IAAI,MAAM,EAAE,QAAQ,EAAE,CAAC;QACrB,KAAK,MAAM,QAAQ,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACvC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,CAAC;gBAAE,QAAQ,CAAC,IAAI,CAAC,0BAA0B,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;QAC/G,CAAC;IACH,CAAC;IACD,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,IAAI,CAAC;YACH,IAAI,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC;gBAAE,QAAQ,CAAC,IAAI,CAAC,yBAAyB,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC;QACzH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,QAAQ,CAAC,IAAI,CAAC,iCAAiC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAC3G,CAAC;IACH,CAAC;IACD,IAAI,MAAM,EAAE,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,KAAK,MAAM,CAAC,MAAM,EAAE,CAAC;QACtE,QAAQ,CAAC,IAAI,CAAC,yBAAyB,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;IAClH,CAAC;IACD,IAAI,MAAM,EAAE,IAAI;QAAE,YAAY,CAAC,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC;IAC9D,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,WAAW,IAAI,EAAE,EAAE,CAAC;QAC5C,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,IAAI,CAAC;YAAE,QAAQ,CAAC,IAAI,CAAC,iBAAiB,IAAI,eAAe,CAAC,CAAC;IAC9F,CAAC;IACD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,cAAc,IAAI,EAAE,EAAE,CAAC;QAC/C,IAAI,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,IAAI,CAAC;YAAE,QAAQ,CAAC,IAAI,CAAC,iBAAiB,IAAI,mBAAmB,CAAC,CAAC;IACjG,CAAC;IACD,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;QACtB,QAAQ,CAAC,IAAI,CAAC,GAAG,kBAAkB,CAAC,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,CAAC;IAClF,CAAC;IACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE,CAAC;QAC5D,QAAQ,CAAC,IAAI,CAAC,8FAA8F,CAAC,CAAC;IAChH,CAAC;SAAM,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,OAAO,GAAG,MAAM,CAAC,UAAU,EAAE,CAAC;QACjF,QAAQ,CAAC,IAAI,CAAC,SAAS,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,MAAM,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAChG,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,YAAY,CACnB,WAA2E,EAC3E,MAAkB,EAClB,QAAkB;IAElB,IAAI,MAAM,GAA0B,MAAM,CAAC,MAAM,CAAC;IAClD,IAAI,UAA8B,CAAC;IACnC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;QACzB,IAAI,CAAC;YACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAc,CAAC;QACpD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,UAAU,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACtE,CAAC;IACH,CAAC;IACD,IAAI,WAAW,CAAC,KAAK,KAAK,IAAI,IAAI,UAAU;QAAE,QAAQ,CAAC,IAAI,CAAC,8BAA8B,UAAU,EAAE,CAAC,CAAC;IACxG,IAAI,WAAW,CAAC,KAAK,KAAK,KAAK,IAAI,CAAC,UAAU;QAAE,QAAQ,CAAC,IAAI,CAAC,sCAAsC,CAAC,CAAC;IACtG,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;QACzB,IAAI,WAAW,CAAC,UAAU,KAAK,SAAS,IAAI,WAAW,CAAC,MAAM,KAAK,SAAS;YAAE,QAAQ,CAAC,IAAI,CAAC,2CAA2C,CAAC,CAAC;QACzI,OAAO;IACT,CAAC;IACD,IAAI,WAAW,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACzC,MAAM,UAAU,GAAG,yBAAyB,CAAC,MAAM,EAAE,WAAW,CAAC,UAAwB,CAAC,CAAC;QAC3F,IAAI,CAAC,UAAU,CAAC,KAAK;YAAE,QAAQ,CAAC,IAAI,CAAC,oCAAoC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC3G,CAAC;IACD,KAAK,MAAM,CAAC,IAAI,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,MAAM,IAAI,EAAE,CAAC,EAAE,CAAC;QACxE,MAAM,KAAK,GAAG,WAAW,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;QACxC,IAAI,CAAC,KAAK,CAAC,MAAM;YAAE,QAAQ,CAAC,IAAI,CAAC,cAAc,IAAI,cAAc,CAAC,CAAC;aAC9D,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,EAAE,CAAC;YACnD,QAAQ,CAAC,IAAI,CAAC,cAAc,IAAI,aAAa,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAC/G,CAAC;IACH,CAAC;AACH,CAAC;AAED,SAAS,WAAW,CAAC,KAAc,EAAE,IAAY;IAC/C,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC;QACnC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC9G,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACpC,IAAI,OAAO,GAAG,KAAK,CAAC;IACpB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,CAAC,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,CAAC,CAAC,OAAO,IAAI,OAAO,CAAC;YAAE,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;QAC/F,OAAO,GAAI,OAAmC,CAAC,OAAO,CAAC,CAAC;IAC1D,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AAC1C,CAAC"}
@@ -0,0 +1,38 @@
1
+ import { type RuntimeOptions } from "@assemblyline-agents/runtime";
2
+ import { EvalEvaluatorLoader } from "./eval-custom.js";
3
+ import { type EvalRuntimeHost } from "./eval-turns.js";
4
+ import type { DiscoveredEvalCase, EvalCaseResult } from "./eval-types.js";
5
+ export interface EvalCaseRuntimeDependencies {
6
+ createRuntime?: (input: {
7
+ caseRoot: string;
8
+ runtimeOptions: RuntimeOptions;
9
+ }) => EvalRuntimeHost;
10
+ judgeComplete?: (input: {
11
+ model: string;
12
+ systemPrompt: string;
13
+ prompt: string;
14
+ }) => Promise<{
15
+ text: string;
16
+ costUsd?: number;
17
+ }>;
18
+ sleep?: (ms: number) => Promise<void>;
19
+ }
20
+ export declare function runEvalCaseWithRetries(input: {
21
+ item: DiscoveredEvalCase & {
22
+ evalCase: NonNullable<DiscoveredEvalCase["evalCase"]>;
23
+ };
24
+ caseId: string;
25
+ repetition: number;
26
+ repetitionsPlanned: number;
27
+ caseRoot: string;
28
+ artifactRoot: string;
29
+ evaluatorLoader: EvalEvaluatorLoader;
30
+ approve: boolean;
31
+ timeoutMs: number;
32
+ judgeModel: string;
33
+ retryAttempts: number;
34
+ retryBackoffMs: number;
35
+ env: Record<string, string | undefined>;
36
+ dependencies: EvalCaseRuntimeDependencies;
37
+ }): Promise<EvalCaseResult>;
38
+ //# sourceMappingURL=eval-case-runner.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-case-runner.d.ts","sourceRoot":"","sources":["../src/eval-case-runner.ts"],"names":[],"mappings":"AAGA,OAAO,EASL,KAAK,cAAc,EACpB,MAAM,8BAA8B,CAAC;AAEtC,OAAO,EAA8B,mBAAmB,EAAE,MAAM,kBAAkB,CAAC;AAKnF,OAAO,EAA0C,KAAK,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAC/F,OAAO,KAAK,EAAE,kBAAkB,EAAE,cAAc,EAAqB,MAAM,iBAAiB,CAAC;AAE7F,MAAM,WAAW,2BAA2B;IAC1C,aAAa,CAAC,EAAE,CAAC,KAAK,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,cAAc,CAAA;KAAE,KAAK,eAAe,CAAC;IACjG,aAAa,CAAC,EAAE,CAAC,KAAK,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,KAAK,OAAO,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAChI,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CACvC;AAED,wBAAsB,sBAAsB,CAAC,KAAK,EAAE;IAClD,IAAI,EAAE,kBAAkB,GAAG;QAAE,QAAQ,EAAE,WAAW,CAAC,kBAAkB,CAAC,UAAU,CAAC,CAAC,CAAA;KAAE,CAAC;IACrF,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,mBAAmB,CAAC;IACrC,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAC;IACxC,YAAY,EAAE,2BAA2B,CAAC;CAC3C,GAAG,OAAO,CAAC,cAAc,CAAC,CAuB1B"}
@@ -0,0 +1,247 @@
1
+ import { mkdirSync } from "node:fs";
2
+ import { resolve } from "node:path";
3
+ import { withNodeModelHarnesses } from "@assemblyline-agents/node";
4
+ import { durableAgentScopeForManifest, FileStateAdapter, LocalBlobAdapter, LocalSandboxAdapter, OpenEveRuntime, StateMemoryStore, loadRuntimeBundle, noopLogger } from "@assemblyline-agents/runtime";
5
+ import { evaluateCase } from "./eval-assertions.js";
6
+ import { evaluateCustomExpectations } from "./eval-custom.js";
7
+ import { seedEvalFixtures } from "./eval-fixtures.js";
8
+ import { createDefaultJudgeCompleter, EvalJudgeResponseError, judgeCase } from "./eval-judge.js";
9
+ import { createToolStubs } from "./eval-mocks.js";
10
+ import { isTransientProviderError, retryTransient } from "./eval-retry.js";
11
+ import { conversationIdForCase, runConversation } from "./eval-turns.js";
12
+ export async function runEvalCaseWithRetries(input) {
13
+ const startedAt = Date.now();
14
+ let runCostUsd = 0;
15
+ let judgeCostUsd = 0;
16
+ let attempt = 1;
17
+ while (true) {
18
+ const attemptRoot = resolve(input.caseRoot, `attempt-${String(attempt).padStart(2, "0")}`);
19
+ mkdirSync(attemptRoot, { recursive: true });
20
+ const attempted = await runOneAttempt(input, attemptRoot, attempt);
21
+ runCostUsd += attempted.result.runCostUsd;
22
+ judgeCostUsd += attempted.result.judgeCostUsd;
23
+ if (!attempted.retryable || attempt > input.retryAttempts) {
24
+ return {
25
+ ...attempted.result,
26
+ attempts: attempt,
27
+ durationMs: Date.now() - startedAt,
28
+ runCostUsd,
29
+ judgeCostUsd
30
+ };
31
+ }
32
+ await (input.dependencies.sleep ?? defaultSleep)(input.retryBackoffMs * (2 ** (attempt - 1)));
33
+ attempt += 1;
34
+ }
35
+ }
36
+ async function runOneAttempt(input, caseRoot, attempt) {
37
+ const startedAt = Date.now();
38
+ const item = input.item;
39
+ const base = {
40
+ caseId: input.caseId,
41
+ name: item.evalCase.name,
42
+ path: item.path,
43
+ tags: item.evalCase.tags ?? [],
44
+ repetition: input.repetition,
45
+ repetitionsPlanned: input.repetitionsPlanned,
46
+ attempts: 1
47
+ };
48
+ let runCostUsd = 0;
49
+ let judgeCostUsd = 0;
50
+ let phase = "runtime";
51
+ try {
52
+ const bundle = loadRuntimeBundle(input.artifactRoot);
53
+ const state = new FileStateAdapter(resolve(caseRoot, "state.json"));
54
+ const sandbox = new LocalSandboxAdapter(resolve(caseRoot, "sandbox"));
55
+ const memoryStore = new StateMemoryStore(state);
56
+ const toolStubs = createToolStubs(item.evalCase.mocks);
57
+ const runtimeOptions = withNodeModelHarnesses({
58
+ ...bundle,
59
+ state,
60
+ blob: new LocalBlobAdapter(resolve(caseRoot, "blobs")),
61
+ sandbox,
62
+ memoryStore,
63
+ ...(toolStubs ? { toolStubs } : {}),
64
+ deliveryMode: "record-only",
65
+ logger: noopLogger(),
66
+ env: input.env,
67
+ devMode: true
68
+ });
69
+ const conversationId = conversationIdForCase(item.evalCase);
70
+ if (item.evalCase.fixtures) {
71
+ await seedEvalFixtures({
72
+ fixtures: item.evalCase.fixtures,
73
+ sandbox,
74
+ memoryStore,
75
+ agentScope: durableAgentScopeForManifest(bundle.manifest),
76
+ ...(conversationId ? { conversationId } : {})
77
+ });
78
+ }
79
+ const createdRuntime = input.dependencies.createRuntime?.({ caseRoot, runtimeOptions }) ?? new OpenEveRuntime(runtimeOptions);
80
+ const outcome = await runConversation({
81
+ runtime: createdRuntime,
82
+ evalCase: item.evalCase,
83
+ caseId: input.caseId,
84
+ repetition: input.repetition,
85
+ attempt,
86
+ approveDefault: input.approve,
87
+ timeoutMs: input.timeoutMs,
88
+ ...(conversationId ? { conversationId } : {})
89
+ });
90
+ if (outcome.kind === "error") {
91
+ runCostUsd += outcome.costUsd;
92
+ if (outcome.runId) {
93
+ try {
94
+ const timeline = await createdRuntime.replay(outcome.runId);
95
+ runCostUsd += timeline.usage.reduce((sum, usage) => sum + (usage.costUsd ?? 0), 0);
96
+ }
97
+ catch {
98
+ // Preserve the original case error when replay is unavailable.
99
+ }
100
+ }
101
+ const error = outcome.cause instanceof Error ? outcome.cause.message : String(outcome.cause);
102
+ return {
103
+ result: {
104
+ ...base,
105
+ outcome: "errored",
106
+ durationMs: Date.now() - startedAt,
107
+ runCostUsd,
108
+ judgeCostUsd: 0,
109
+ failures: [],
110
+ error,
111
+ ...(outcome.runId ? { runId: outcome.runId } : {})
112
+ },
113
+ retryable: isTransientProviderError(outcome.cause)
114
+ };
115
+ }
116
+ if (outcome.kind === "run_failed") {
117
+ runCostUsd += outcome.costUsd;
118
+ return {
119
+ result: {
120
+ ...base,
121
+ outcome: "errored",
122
+ durationMs: Date.now() - startedAt,
123
+ runCostUsd,
124
+ judgeCostUsd: 0,
125
+ failures: [],
126
+ error: outcome.error,
127
+ runId: outcome.runId,
128
+ status: outcome.status,
129
+ response: outcome.response
130
+ },
131
+ retryable: isTransientProviderError(outcome.error)
132
+ };
133
+ }
134
+ const actual = outcome.actual;
135
+ runCostUsd += actual.costUsd;
136
+ phase = "assertions";
137
+ const failures = [...outcome.turnFailures, ...evaluateCase(item.evalCase, actual)];
138
+ let judge;
139
+ if (item.evalCase.expect.judge) {
140
+ phase = "judge";
141
+ const model = item.evalCase.expect.judge.model ?? input.judgeModel;
142
+ const judged = await retryTransient(() => judgeCase(item.evalCase, actual, {
143
+ model,
144
+ env: input.env,
145
+ ...(input.dependencies.judgeComplete ? { complete: input.dependencies.judgeComplete } : {})
146
+ }), {
147
+ retryAttempts: input.retryAttempts,
148
+ retryBackoffMs: input.retryBackoffMs,
149
+ ...(input.dependencies.sleep ? { sleep: input.dependencies.sleep } : {})
150
+ });
151
+ judge = judged.value;
152
+ judgeCostUsd = judge.costUsd;
153
+ const threshold = item.evalCase.expect.judge.threshold ?? 0.7;
154
+ if (!judge.pass || judge.score < threshold) {
155
+ failures.push(`judge score ${judge.score.toFixed(3)} did not pass threshold ${threshold.toFixed(3)}: ${judge.reasoning}`);
156
+ }
157
+ // Durable verdict for local in-process runs (the case's state adapter
158
+ // owns the run); remote gateway runs skip — no append path over HTTP.
159
+ if (createdRuntime instanceof OpenEveRuntime && outcome.runId) {
160
+ try {
161
+ await state.appendEvent(outcome.runId, "eval.judged", {
162
+ caseId: input.caseId,
163
+ case: item.evalCase.name,
164
+ repetition: input.repetition,
165
+ attempt,
166
+ model,
167
+ score: judge.score,
168
+ pass: judge.pass,
169
+ threshold,
170
+ reasoning: judge.reasoning,
171
+ costUsd: judge.costUsd
172
+ });
173
+ }
174
+ catch {
175
+ // Grading must never fail on the verdict append.
176
+ }
177
+ }
178
+ }
179
+ phase = "custom";
180
+ const judgeComplete = input.dependencies.judgeComplete ?? createDefaultJudgeCompleter(input.env);
181
+ let evaluatorCostUsd = 0;
182
+ const complete = async (request) => {
183
+ const completion = await judgeComplete({
184
+ model: request.model ?? input.judgeModel,
185
+ systemPrompt: request.systemPrompt ?? "",
186
+ prompt: request.prompt
187
+ });
188
+ const costUsd = completion.costUsd ?? 0;
189
+ evaluatorCostUsd += costUsd;
190
+ return { text: completion.text, costUsd };
191
+ };
192
+ const finalTimeline = outcome.timelines[outcome.timelines.length - 1];
193
+ if (!finalTimeline)
194
+ throw new Error("conversation produced no timeline");
195
+ const custom = await evaluateCustomExpectations(item.evalCase.expect.evaluators ?? [], input.evaluatorLoader, {
196
+ evalCase: item.evalCase,
197
+ actual,
198
+ timeline: finalTimeline,
199
+ timelines: outcome.timelines,
200
+ repetition: input.repetition,
201
+ attempt,
202
+ complete
203
+ });
204
+ judgeCostUsd += evaluatorCostUsd;
205
+ failures.push(...custom.failures);
206
+ const evaluated = {
207
+ ...base,
208
+ outcome: failures.length === 0 ? "passed" : "failed",
209
+ durationMs: Date.now() - startedAt,
210
+ runCostUsd,
211
+ judgeCostUsd,
212
+ failures,
213
+ runId: outcome.runId,
214
+ status: actual.status,
215
+ response: actual.response
216
+ };
217
+ if (actual.output !== undefined)
218
+ evaluated.output = actual.output;
219
+ if (actual.turns)
220
+ evaluated.turns = actual.turns;
221
+ if (judge)
222
+ evaluated.judge = judge;
223
+ if (custom.metrics.length > 0)
224
+ evaluated.metrics = custom.metrics;
225
+ return { result: evaluated, retryable: false };
226
+ }
227
+ catch (error) {
228
+ if (error instanceof EvalJudgeResponseError)
229
+ judgeCostUsd += error.costUsd;
230
+ return {
231
+ result: {
232
+ ...base,
233
+ outcome: "errored",
234
+ durationMs: Date.now() - startedAt,
235
+ runCostUsd,
236
+ judgeCostUsd,
237
+ failures: [],
238
+ error: error instanceof Error ? error.message : String(error)
239
+ },
240
+ retryable: phase === "runtime" && isTransientProviderError(error)
241
+ };
242
+ }
243
+ }
244
+ function defaultSleep(ms) {
245
+ return new Promise((resolveSleep) => setTimeout(resolveSleep, ms));
246
+ }
247
+ //# sourceMappingURL=eval-case-runner.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-case-runner.js","sourceRoot":"","sources":["../src/eval-case-runner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,sBAAsB,EAAE,MAAM,2BAA2B,CAAC;AACnE,OAAO,EACL,4BAA4B,EAC5B,gBAAgB,EAChB,gBAAgB,EAChB,mBAAmB,EACnB,cAAc,EACd,gBAAgB,EAChB,iBAAiB,EACjB,UAAU,EAEX,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AACpD,OAAO,EAAE,0BAA0B,EAAuB,MAAM,kBAAkB,CAAC;AACnF,OAAO,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AACtD,OAAO,EAAE,2BAA2B,EAAE,sBAAsB,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AACjG,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,wBAAwB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAC3E,OAAO,EAAE,qBAAqB,EAAE,eAAe,EAAwB,MAAM,iBAAiB,CAAC;AAS/F,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,KAe5C;IACC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC7B,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,OAAO,IAAI,EAAE,CAAC;QACZ,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,QAAQ,EAAE,WAAW,MAAM,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAC3F,SAAS,CAAC,WAAW,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC5C,MAAM,SAAS,GAAG,MAAM,aAAa,CAAC,KAAK,EAAE,WAAW,EAAE,OAAO,CAAC,CAAC;QACnE,UAAU,IAAI,SAAS,CAAC,MAAM,CAAC,UAAU,CAAC;QAC1C,YAAY,IAAI,SAAS,CAAC,MAAM,CAAC,YAAY,CAAC;QAC9C,IAAI,CAAC,SAAS,CAAC,SAAS,IAAI,OAAO,GAAG,KAAK,CAAC,aAAa,EAAE,CAAC;YAC1D,OAAO;gBACL,GAAG,SAAS,CAAC,MAAM;gBACnB,QAAQ,EAAE,OAAO;gBACjB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBAClC,UAAU;gBACV,YAAY;aACb,CAAC;QACJ,CAAC;QACD,MAAM,CAAC,KAAK,CAAC,YAAY,CAAC,KAAK,IAAI,YAAY,CAAC,CAAC,KAAK,CAAC,cAAc,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QAC9F,OAAO,IAAI,CAAC,CAAC;IACf,CAAC;AACH,CAAC;AAED,KAAK,UAAU,aAAa,CAC1B,KAAmD,EACnD,QAAgB,EAChB,OAAe;IAEf,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC7B,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC;IACxB,MAAM,IAAI,GAAgG;QACxG,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;QACxB,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,IAAI,EAAE;QAC9B,UAAU,EAAE,KAAK,CAAC,UAAU;QAC5B,kBAAkB,EAAE,KAAK,CAAC,kBAAkB;QAC5C,QAAQ,EAAE,CAAC;KACZ,CAAC;IACF,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,KAAK,GAAkD,SAAS,CAAC;IACrE,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;QACrD,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,OAAO,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC,CAAC;QACpE,MAAM,OAAO,GAAG,IAAI,mBAAmB,CAAC,OAAO,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAC;QACtE,MAAM,WAAW,GAAG,IAAI,gBAAgB,CAAC,KAAK,CAAC,CAAC;QAChD,MAAM,SAAS,GAAG,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QACvD,MAAM,cAAc,GAAmB,sBAAsB,CAAC;YAC5D,GAAG,MAAM;YACT,KAAK;YACL,IAAI,EAAE,IAAI,gBAAgB,CAAC,OAAO,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACtD,OAAO;YACP,WAAW;YACX,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnC,YAAY,EAAE,aAAa;YAC3B,MAAM,EAAE,UAAU,EAAE;YACpB,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,OAAO,EAAE,IAAI;SACd,CAAC,CAAC;QACH,MAAM,cAAc,GAAG,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC5D,IAAI,IAAI,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC;YAC3B,MAAM,gBAAgB,CAAC;gBACrB,QAAQ,EAAE,IAAI,CAAC,QAAQ,CAAC,QAAQ;gBAChC,OAAO;gBACP,WAAW;gBACX,UAAU,EAAE,4BAA4B,CAAC,MAAM,CAAC,QAAQ,CAAC;gBACzD,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC9C,CAAC,CAAC;QACL,CAAC;QACD,MAAM,cAAc,GAAG,KAAK,CAAC,YAAY,CAAC,aAAa,EAAE,CAAC,EAAE,QAAQ,EAAE,cAAc,EAAE,CAAC,IAAI,IAAI,cAAc,CAAC,cAAc,CAAC,CAAC;QAC9H,MAAM,OAAO,GAAG,MAAM,eAAe,CAAC;YACpC,OAAO,EAAE,cAAc;YACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,OAAO;YACP,cAAc,EAAE,KAAK,CAAC,OAAO;YAC7B,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC9C,CAAC,CAAC;QACH,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YAC7B,UAAU,IAAI,OAAO,CAAC,OAAO,CAAC;YAC9B,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;gBAClB,IAAI,CAAC;oBACH,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;oBAC5D,UAAU,IAAI,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;gBACrF,CAAC;gBAAC,MAAM,CAAC;oBACP,+DAA+D;gBACjE,CAAC;YACH,CAAC;YACD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YAC7F,OAAO;gBACL,MAAM,EAAE;oBACN,GAAG,IAAI;oBACP,OAAO,EAAE,SAAS;oBAClB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAClC,UAAU;oBACV,YAAY,EAAE,CAAC;oBACf,QAAQ,EAAE,EAAE;oBACZ,KAAK;oBACL,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;iBACnD;gBACD,SAAS,EAAE,wBAAwB,CAAC,OAAO,CAAC,KAAK,CAAC;aACnD,CAAC;QACJ,CAAC;QACD,IAAI,OAAO,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;YAClC,UAAU,IAAI,OAAO,CAAC,OAAO,CAAC;YAC9B,OAAO;gBACL,MAAM,EAAE;oBACN,GAAG,IAAI;oBACP,OAAO,EAAE,SAAS;oBAClB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAClC,UAAU;oBACV,YAAY,EAAE,CAAC;oBACf,QAAQ,EAAE,EAAE;oBACZ,KAAK,EAAE,OAAO,CAAC,KAAK;oBACpB,KAAK,EAAE,OAAO,CAAC,KAAK;oBACpB,MAAM,EAAE,OAAO,CAAC,MAAM;oBACtB,QAAQ,EAAE,OAAO,CAAC,QAAQ;iBAC3B;gBACD,SAAS,EAAE,wBAAwB,CAAC,OAAO,CAAC,KAAK,CAAC;aACnD,CAAC;QACJ,CAAC;QACD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;QAC9B,UAAU,IAAI,MAAM,CAAC,OAAO,CAAC;QAC7B,KAAK,GAAG,YAAY,CAAC;QACrB,MAAM,QAAQ,GAAG,CAAC,GAAG,OAAO,CAAC,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;QACnF,IAAI,KAAK,CAAC;QACV,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;YAC/B,KAAK,GAAG,OAAO,CAAC;YAChB,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,UAAU,CAAC;YACnE,MAAM,MAAM,GAAG,MAAM,cAAc,CACjC,GAAG,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,EAAE;gBACrC,KAAK;gBACL,GAAG,EAAE,KAAK,CAAC,GAAG;gBACd,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,YAAY,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC5F,CAAC,EACF;gBACE,aAAa,EAAE,KAAK,CAAC,aAAa;gBAClC,cAAc,EAAE,KAAK,CAAC,cAAc;gBACpC,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,YAAY,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACzE,CACF,CAAC;YACF,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;YACrB,YAAY,GAAG,KAAK,CAAC,OAAO,CAAC;YAC7B,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,SAAS,IAAI,GAAG,CAAC;YAC9D,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,KAAK,GAAG,SAAS,EAAE,CAAC;gBAC3C,QAAQ,CAAC,IAAI,CAAC,eAAe,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,2BAA2B,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC;YAC5H,CAAC;YACD,sEAAsE;YACtE,sEAAsE;YACtE,IAAI,cAAc,YAAY,cAAc,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;gBAC9D,IAAI,CAAC;oBACH,MAAM,KAAK,CAAC,WAAW,CAAC,OAAO,CAAC,KAAK,EAAE,aAAa,EAAE;wBACpD,MAAM,EAAE,KAAK,CAAC,MAAM;wBACpB,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;wBACxB,UAAU,EAAE,KAAK,CAAC,UAAU;wBAC5B,OAAO;wBACP,KAAK;wBACL,KAAK,EAAE,KAAK,CAAC,KAAK;wBAClB,IAAI,EAAE,KAAK,CAAC,IAAI;wBAChB,SAAS;wBACT,SAAS,EAAE,KAAK,CAAC,SAAS;wBAC1B,OAAO,EAAE,KAAK,CAAC,OAAO;qBACvB,CAAC,CAAC;gBACL,CAAC;gBAAC,MAAM,CAAC;oBACP,iDAAiD;gBACnD,CAAC;YACH,CAAC;QACH,CAAC;QACD,KAAK,GAAG,QAAQ,CAAC;QACjB,MAAM,aAAa,GAAG,KAAK,CAAC,YAAY,CAAC,aAAa,IAAI,2BAA2B,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACjG,IAAI,gBAAgB,GAAG,CAAC,CAAC;QACzB,MAAM,QAAQ,GAAsB,KAAK,EAAE,OAAO,EAAE,EAAE;YACpD,MAAM,UAAU,GAAG,MAAM,aAAa,CAAC;gBACrC,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,KAAK,CAAC,UAAU;gBACxC,YAAY,EAAE,OAAO,CAAC,YAAY,IAAI,EAAE;gBACxC,MAAM,EAAE,OAAO,CAAC,MAAM;aACvB,CAAC,CAAC;YACH,MAAM,OAAO,GAAG,UAAU,CAAC,OAAO,IAAI,CAAC,CAAC;YACxC,gBAAgB,IAAI,OAAO,CAAC;YAC5B,OAAO,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC;QAC5C,CAAC,CAAC;QACF,MAAM,aAAa,GAAG,OAAO,CAAC,SAAS,CAAC,OAAO,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACtE,IAAI,CAAC,aAAa;YAAE,MAAM,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC;QACzE,MAAM,MAAM,GAAG,MAAM,0BAA0B,CAC7C,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,UAAU,IAAI,EAAE,EACrC,KAAK,CAAC,eAAe,EACrB;YACE,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,MAAM;YACN,QAAQ,EAAE,aAAa;YACvB,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,OAAO;YACP,QAAQ;SACT,CACF,CAAC;QACF,YAAY,IAAI,gBAAgB,CAAC;QACjC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC;QAClC,MAAM,SAAS,GAAmB;YAChC,GAAG,IAAI;YACP,OAAO,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ;YACpD,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAClC,UAAU;YACV,YAAY;YACZ,QAAQ;YACR,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,QAAQ,EAAE,MAAM,CAAC,QAAQ;SAC1B,CAAC;QACF,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS;YAAE,SAAS,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;QAClE,IAAI,MAAM,CAAC,KAAK;YAAE,SAAS,CAAC,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;QACjD,IAAI,KAAK;YAAE,SAAS,CAAC,KAAK,GAAG,KAAK,CAAC;QACnC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC;YAAE,SAAS,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAClE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;IACjD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,sBAAsB;YAAE,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC;QAC3E,OAAO;YACL,MAAM,EAAE;gBACN,GAAG,IAAI;gBACP,OAAO,EAAE,SAAS;gBAClB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBAClC,UAAU;gBACV,YAAY;gBACZ,QAAQ,EAAE,EAAE;gBACZ,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;aAC9D;YACD,SAAS,EAAE,KAAK,KAAK,SAAS,IAAI,wBAAwB,CAAC,KAAK,CAAC;SAClE,CAAC;IACJ,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CAAC,EAAU;IAC9B,OAAO,IAAI,OAAO,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,UAAU,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC,CAAC;AACrE,CAAC"}
@@ -0,0 +1,6 @@
1
+ import type { DiscoveredEvalCase } from "./eval-types.js";
2
+ export declare function discoverCases(agentRoot: string, options?: {
3
+ filter?: string;
4
+ tag?: string;
5
+ }): DiscoveredEvalCase[];
6
+ //# sourceMappingURL=eval-cases.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-cases.d.ts","sourceRoot":"","sources":["../src/eval-cases.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,kBAAkB,EAAY,MAAM,iBAAiB,CAAC;AAEpE,wBAAgB,aAAa,CAC3B,SAAS,EAAE,MAAM,EACjB,OAAO,GAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAA;CAAO,GAC9C,kBAAkB,EAAE,CA2BtB"}