@post-print/agent-test 1.1.9 → 2.0.0

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 (238) hide show
  1. package/README.md +18 -66
  2. package/dist/cli.d.ts +2 -64
  3. package/dist/cli.d.ts.map +1 -1
  4. package/dist/cli.js +26 -691
  5. package/dist/cli.js.map +1 -1
  6. package/dist/compare-scenario.d.ts +1 -1
  7. package/dist/compare-scenario.d.ts.map +1 -1
  8. package/dist/compare-scenario.js +72 -75
  9. package/dist/compare-scenario.js.map +1 -1
  10. package/dist/failures.d.ts +5 -1
  11. package/dist/failures.d.ts.map +1 -1
  12. package/dist/failures.js +1 -1
  13. package/dist/failures.js.map +1 -1
  14. package/dist/host-login.d.ts +1 -26
  15. package/dist/host-login.d.ts.map +1 -1
  16. package/dist/host-login.js +1 -92
  17. package/dist/host-login.js.map +1 -1
  18. package/dist/index.d.ts +6 -28
  19. package/dist/index.d.ts.map +1 -1
  20. package/dist/index.js +4 -26
  21. package/dist/index.js.map +1 -1
  22. package/dist/scenario-story.d.ts.map +1 -1
  23. package/dist/scenario-story.js +57 -39
  24. package/dist/scenario-story.js.map +1 -1
  25. package/dist/sdk/cli.d.ts +3 -0
  26. package/dist/sdk/cli.d.ts.map +1 -0
  27. package/dist/sdk/cli.js +20 -0
  28. package/dist/sdk/cli.js.map +1 -0
  29. package/dist/sdk/definitions.d.ts +22 -0
  30. package/dist/sdk/definitions.d.ts.map +1 -0
  31. package/dist/sdk/definitions.js +58 -0
  32. package/dist/sdk/definitions.js.map +1 -0
  33. package/dist/sdk/expect.d.ts +31 -0
  34. package/dist/sdk/expect.d.ts.map +1 -0
  35. package/dist/sdk/expect.js +68 -0
  36. package/dist/sdk/expect.js.map +1 -0
  37. package/dist/sdk/judge.d.ts +16 -0
  38. package/dist/sdk/judge.d.ts.map +1 -0
  39. package/dist/sdk/judge.js +82 -0
  40. package/dist/sdk/judge.js.map +1 -0
  41. package/dist/sdk/metrics.d.ts +3 -0
  42. package/dist/sdk/metrics.d.ts.map +1 -0
  43. package/dist/sdk/metrics.js +24 -0
  44. package/dist/sdk/metrics.js.map +1 -0
  45. package/dist/sdk/reporter.d.ts +12 -0
  46. package/dist/sdk/reporter.d.ts.map +1 -0
  47. package/dist/sdk/reporter.js +58 -0
  48. package/dist/sdk/reporter.js.map +1 -0
  49. package/dist/sdk/runtime.d.ts +33 -0
  50. package/dist/sdk/runtime.d.ts.map +1 -0
  51. package/dist/sdk/runtime.js +188 -0
  52. package/dist/sdk/runtime.js.map +1 -0
  53. package/dist/sdk/test.d.ts +27 -0
  54. package/dist/sdk/test.d.ts.map +1 -0
  55. package/dist/sdk/test.js +92 -0
  56. package/dist/sdk/test.js.map +1 -0
  57. package/dist/sdk/types.d.ts +75 -0
  58. package/dist/sdk/types.d.ts.map +1 -0
  59. package/dist/sdk/types.js +2 -0
  60. package/dist/sdk/types.js.map +1 -0
  61. package/dist/sdk/usage.d.ts +10 -0
  62. package/dist/sdk/usage.d.ts.map +1 -0
  63. package/dist/sdk/usage.js +13 -0
  64. package/dist/sdk/usage.js.map +1 -0
  65. package/dist/sdk/viewer-results.d.ts +43 -0
  66. package/dist/sdk/viewer-results.d.ts.map +1 -0
  67. package/dist/sdk/viewer-results.js +126 -0
  68. package/dist/sdk/viewer-results.js.map +1 -0
  69. package/dist/sdk/viewer.d.ts +11 -0
  70. package/dist/sdk/viewer.d.ts.map +1 -0
  71. package/dist/sdk/viewer.js +160 -0
  72. package/dist/sdk/viewer.js.map +1 -0
  73. package/dist/sdk/workspace.d.ts +33 -0
  74. package/dist/sdk/workspace.d.ts.map +1 -0
  75. package/dist/sdk/workspace.js +100 -0
  76. package/dist/sdk/workspace.js.map +1 -0
  77. package/dist/types.d.ts +4 -40
  78. package/dist/types.d.ts.map +1 -1
  79. package/dist/viewer/catalog.d.ts +2 -8
  80. package/dist/viewer/catalog.d.ts.map +1 -1
  81. package/dist/viewer/catalog.js +23 -128
  82. package/dist/viewer/catalog.js.map +1 -1
  83. package/dist/viewer/client.d.ts.map +1 -1
  84. package/dist/viewer/client.js +12 -12
  85. package/dist/viewer/client.js.map +1 -1
  86. package/dist/viewer/run-controller.d.ts +4 -2
  87. package/dist/viewer/run-controller.d.ts.map +1 -1
  88. package/dist/viewer/run-controller.js +297 -248
  89. package/dist/viewer/run-controller.js.map +1 -1
  90. package/dist/viewer/server.d.ts +4 -3
  91. package/dist/viewer/server.d.ts.map +1 -1
  92. package/dist/viewer/server.js +149 -112
  93. package/dist/viewer/server.js.map +1 -1
  94. package/dist/viewer/styles.d.ts +2 -0
  95. package/dist/viewer/styles.d.ts.map +1 -0
  96. package/dist/viewer/styles.js +444 -0
  97. package/dist/viewer/styles.js.map +1 -0
  98. package/dist/worker-pool.d.ts +5 -1
  99. package/dist/worker-pool.d.ts.map +1 -1
  100. package/dist/worker-pool.js +1 -1
  101. package/dist/worker-pool.js.map +1 -1
  102. package/package.json +7 -7
  103. package/dist/check.d.ts +0 -38
  104. package/dist/check.d.ts.map +0 -1
  105. package/dist/check.js +0 -169
  106. package/dist/check.js.map +0 -1
  107. package/dist/command-allowlist.d.ts +0 -9
  108. package/dist/command-allowlist.d.ts.map +0 -1
  109. package/dist/command-allowlist.js +0 -29
  110. package/dist/command-allowlist.js.map +0 -1
  111. package/dist/debug-bundle.d.ts +0 -81
  112. package/dist/debug-bundle.d.ts.map +0 -1
  113. package/dist/debug-bundle.js +0 -392
  114. package/dist/debug-bundle.js.map +0 -1
  115. package/dist/discover-suites.d.ts +0 -2
  116. package/dist/discover-suites.d.ts.map +0 -1
  117. package/dist/discover-suites.js +0 -19
  118. package/dist/discover-suites.js.map +0 -1
  119. package/dist/doctor.d.ts +0 -24
  120. package/dist/doctor.d.ts.map +0 -1
  121. package/dist/doctor.js +0 -272
  122. package/dist/doctor.js.map +0 -1
  123. package/dist/expect.d.ts +0 -59
  124. package/dist/expect.d.ts.map +0 -1
  125. package/dist/expect.js +0 -518
  126. package/dist/expect.js.map +0 -1
  127. package/dist/host-log.d.ts +0 -17
  128. package/dist/host-log.d.ts.map +0 -1
  129. package/dist/host-log.js +0 -154
  130. package/dist/host-log.js.map +0 -1
  131. package/dist/hosts.d.ts +0 -22
  132. package/dist/hosts.d.ts.map +0 -1
  133. package/dist/hosts.js +0 -84
  134. package/dist/hosts.js.map +0 -1
  135. package/dist/html-report.d.ts +0 -18
  136. package/dist/html-report.d.ts.map +0 -1
  137. package/dist/html-report.js +0 -1276
  138. package/dist/html-report.js.map +0 -1
  139. package/dist/live-isolation.d.ts +0 -82
  140. package/dist/live-isolation.d.ts.map +0 -1
  141. package/dist/live-isolation.js +0 -399
  142. package/dist/live-isolation.js.map +0 -1
  143. package/dist/live-timeout.d.ts +0 -11
  144. package/dist/live-timeout.d.ts.map +0 -1
  145. package/dist/live-timeout.js +0 -29
  146. package/dist/live-timeout.js.map +0 -1
  147. package/dist/load-adapters.d.ts +0 -13
  148. package/dist/load-adapters.d.ts.map +0 -1
  149. package/dist/load-adapters.js +0 -76
  150. package/dist/load-adapters.js.map +0 -1
  151. package/dist/load-suite.d.ts +0 -22
  152. package/dist/load-suite.d.ts.map +0 -1
  153. package/dist/load-suite.js +0 -154
  154. package/dist/load-suite.js.map +0 -1
  155. package/dist/mcp-config.d.ts +0 -13
  156. package/dist/mcp-config.d.ts.map +0 -1
  157. package/dist/mcp-config.js +0 -121
  158. package/dist/mcp-config.js.map +0 -1
  159. package/dist/preflight.d.ts +0 -3
  160. package/dist/preflight.d.ts.map +0 -1
  161. package/dist/preflight.js +0 -17
  162. package/dist/preflight.js.map +0 -1
  163. package/dist/progress.d.ts +0 -24
  164. package/dist/progress.d.ts.map +0 -1
  165. package/dist/progress.js +0 -111
  166. package/dist/progress.js.map +0 -1
  167. package/dist/record-trace.d.ts +0 -64
  168. package/dist/record-trace.d.ts.map +0 -1
  169. package/dist/record-trace.js +0 -124
  170. package/dist/record-trace.js.map +0 -1
  171. package/dist/report-preview-server.d.ts +0 -2
  172. package/dist/report-preview-server.d.ts.map +0 -1
  173. package/dist/report-preview-server.js +0 -24
  174. package/dist/report-preview-server.js.map +0 -1
  175. package/dist/report-preview.d.ts +0 -35
  176. package/dist/report-preview.d.ts.map +0 -1
  177. package/dist/report-preview.js +0 -205
  178. package/dist/report-preview.js.map +0 -1
  179. package/dist/run-suite.d.ts +0 -139
  180. package/dist/run-suite.d.ts.map +0 -1
  181. package/dist/run-suite.js +0 -1688
  182. package/dist/run-suite.js.map +0 -1
  183. package/dist/scenario-finalizer.d.ts +0 -27
  184. package/dist/scenario-finalizer.d.ts.map +0 -1
  185. package/dist/scenario-finalizer.js +0 -39
  186. package/dist/scenario-finalizer.js.map +0 -1
  187. package/dist/scenario-retry.d.ts +0 -15
  188. package/dist/scenario-retry.d.ts.map +0 -1
  189. package/dist/scenario-retry.js +0 -52
  190. package/dist/scenario-retry.js.map +0 -1
  191. package/dist/scenario-seed.d.ts +0 -13
  192. package/dist/scenario-seed.d.ts.map +0 -1
  193. package/dist/scenario-seed.js +0 -72
  194. package/dist/scenario-seed.js.map +0 -1
  195. package/dist/suite-summary.d.ts +0 -16
  196. package/dist/suite-summary.d.ts.map +0 -1
  197. package/dist/suite-summary.js +0 -218
  198. package/dist/suite-summary.js.map +0 -1
  199. package/dist/theme.d.ts +0 -96
  200. package/dist/theme.d.ts.map +0 -1
  201. package/dist/theme.js +0 -446
  202. package/dist/theme.js.map +0 -1
  203. package/dist/validate-seeds.d.ts +0 -20
  204. package/dist/validate-seeds.d.ts.map +0 -1
  205. package/dist/validate-seeds.js +0 -72
  206. package/dist/validate-seeds.js.map +0 -1
  207. package/dist/validate-suite.d.ts +0 -24
  208. package/dist/validate-suite.d.ts.map +0 -1
  209. package/dist/validate-suite.js +0 -542
  210. package/dist/validate-suite.js.map +0 -1
  211. package/dist/viewer/cli.d.ts +0 -17
  212. package/dist/viewer/cli.d.ts.map +0 -1
  213. package/dist/viewer/cli.js +0 -38
  214. package/dist/viewer/cli.js.map +0 -1
  215. package/dist/viewer/compare-metrics.d.ts +0 -30
  216. package/dist/viewer/compare-metrics.d.ts.map +0 -1
  217. package/dist/viewer/compare-metrics.js +0 -101
  218. package/dist/viewer/compare-metrics.js.map +0 -1
  219. package/dist/viewer/context-files.d.ts +0 -8
  220. package/dist/viewer/context-files.d.ts.map +0 -1
  221. package/dist/viewer/context-files.js +0 -57
  222. package/dist/viewer/context-files.js.map +0 -1
  223. package/dist/viewer/emit.d.ts +0 -7
  224. package/dist/viewer/emit.d.ts.map +0 -1
  225. package/dist/viewer/emit.js +0 -25
  226. package/dist/viewer/emit.js.map +0 -1
  227. package/dist/viewer/live-runner.d.ts +0 -27
  228. package/dist/viewer/live-runner.d.ts.map +0 -1
  229. package/dist/viewer/live-runner.js +0 -166
  230. package/dist/viewer/live-runner.js.map +0 -1
  231. package/dist/viewer/session-server.d.ts +0 -2
  232. package/dist/viewer/session-server.d.ts.map +0 -1
  233. package/dist/viewer/session-server.js +0 -71
  234. package/dist/viewer/session-server.js.map +0 -1
  235. package/dist/warnings.d.ts +0 -3
  236. package/dist/warnings.d.ts.map +0 -1
  237. package/dist/warnings.js +0 -13
  238. package/dist/warnings.js.map +0 -1
@@ -1,392 +0,0 @@
1
- import { access, chmod, mkdir, writeFile } from "node:fs/promises";
2
- import { join } from "node:path";
3
- import { scenarioArtifactSlug } from "./record-trace.js";
4
- const AGENT_TEST_ENV_KEYS = [
5
- "AGENT_TEST_VERBOSE",
6
- "AGENT_TEST_VERBOSE_PATHS",
7
- "AGENT_TEST_HOST_LOGS",
8
- "AGENT_TEST_NO_REPORT_PREVIEW",
9
- "AGENT_TEST_QUIET",
10
- "AGENT_TEST_TIMEOUT_MS",
11
- "AGENT_TEST_ALLOW_IN_PLACE",
12
- "AGENT_TEST_NO_WORKTREE",
13
- "AGENT_TEST_NO_ISOLATE",
14
- "AGENT_TEST_SCENARIO_SETTLE_MS",
15
- "AGENT_TEST_SCENARIO_RETRIES",
16
- "AGENT_TEST_DEBUG",
17
- "AGENT_TEST_CHILD",
18
- ];
19
- const TOOL_ARG_COMPACT_CHARS = 500;
20
- const TOOL_RESULT_MAX_CHARS = 4_000;
21
- /** Shell-quote a single argument for POSIX sh. */
22
- export function shellQuote(value) {
23
- if (value.length === 0) {
24
- return "''";
25
- }
26
- if (/^[A-Za-z0-9_./:@%+=,-]+$/.test(value)) {
27
- return value;
28
- }
29
- return `'${value.replace(/'/g, `'\\''`)}'`;
30
- }
31
- /** Flatten text for a single `# …` shell comment (no newline breakout). */
32
- export function shellCommentText(value) {
33
- return value.replace(/[\r\n]+/g, " ").trim();
34
- }
35
- export function buildRerunCommand(options) {
36
- const args = [options.cliPath];
37
- args.push("--suites-dir", options.suitesDir, "--suite", options.suite, "--scenario", options.scenario);
38
- if (options.rubricsDir) {
39
- args.push("--rubrics-dir", options.rubricsDir);
40
- }
41
- if (options.host) {
42
- args.push("--host", options.host);
43
- }
44
- for (const modulePath of options.adapterModules ?? []) {
45
- args.push("--adapter", modulePath);
46
- }
47
- if (options.judge === false) {
48
- args.push("--no-judge");
49
- }
50
- if (options.worktree === false) {
51
- args.push("--no-worktree");
52
- }
53
- if (options.noTimeout) {
54
- args.push("--no-timeout");
55
- }
56
- else if (options.timeoutMs !== undefined) {
57
- args.push("--timeout-ms", String(options.timeoutMs));
58
- }
59
- if (options.allowUserInput) {
60
- args.push("--allow-user-input");
61
- }
62
- if (options.authMode) {
63
- args.push("--auth-mode", options.authMode);
64
- }
65
- args.push("--debug");
66
- if (options.debugDir) {
67
- args.push("--debug-dir", options.debugDir);
68
- }
69
- if (options.keepRecordings !== false) {
70
- args.push("--keep-recordings");
71
- }
72
- return [shellQuote(process.execPath), ...args.map(shellQuote)].join(" ");
73
- }
74
- export function collectDebugEnvironment(options) {
75
- const agentTestEnv = {};
76
- for (const key of AGENT_TEST_ENV_KEYS) {
77
- const value = process.env[key];
78
- if (value !== undefined) {
79
- agentTestEnv[key] = value;
80
- }
81
- }
82
- return {
83
- nodeVersion: process.versions.node,
84
- bunVersion: process.versions.bun,
85
- packageVersion: options.packageVersion,
86
- host: options.host,
87
- suite: options.suite,
88
- scenario: options.scenario,
89
- cursorApiKeySet: Boolean(process.env.CURSOR_API_KEY?.trim()),
90
- cursorAgentModel: process.env.CURSOR_AGENT_MODEL,
91
- cursorJudgeModel: process.env.CURSOR_JUDGE_MODEL,
92
- cursorJudgeTemperature: process.env.CURSOR_JUDGE_TEMPERATURE,
93
- anthropicApiKeySet: Boolean(process.env.ANTHROPIC_API_KEY?.trim()),
94
- claudeAgentModel: process.env.CLAUDE_AGENT_MODEL,
95
- claudeCodeBin: process.env.CLAUDE_CODE_BIN,
96
- timeoutMs: options.timeoutMs,
97
- worktree: options.worktree,
98
- isolateLive: options.isolateLive,
99
- agentTestEnv,
100
- };
101
- }
102
- function compactText(input, maxChars) {
103
- if (input === undefined) {
104
- return "";
105
- }
106
- const raw = typeof input === "string" ? input : JSON.stringify(input);
107
- if (raw.length <= maxChars) {
108
- return raw;
109
- }
110
- return `${raw.slice(0, maxChars)}…`;
111
- }
112
- function whyHint(category) {
113
- switch (category) {
114
- case "agent_runtime":
115
- return "SDK/runtime did not finish cleanly — check raw status, sdkError, and the last tools/messages in the transcript.";
116
- case "worktree_leak":
117
- return "Caller checkout changed outside the scenario worktree — inspect porcelain lines and whether Shell ignored worktree cwd.";
118
- case "judge_infra":
119
- return "Judge SDK/API failure (not a criterion miss) — retry or check CURSOR_API_KEY / rate limits.";
120
- case "judge_parse":
121
- return "Judge returned unparseable JSON — see judge-debug.json and rationale.";
122
- case "recording_error":
123
- return "Failed to persist a staging/fixture trace — check disk permissions under the session root.";
124
- default:
125
- return "Rubric/judge criterion miss — compare expected rubric in scenario.json to transcript output.";
126
- }
127
- }
128
- export function formatDebugWhySection(options) {
129
- const lines = ["## Why", ""];
130
- if (options.result.passed) {
131
- lines.push("_Scenario passed._", "");
132
- return lines;
133
- }
134
- if (options.result.failures.length === 0) {
135
- lines.push("_Failed with no recorded failures._", "");
136
- return lines;
137
- }
138
- for (const [index, failure] of options.result.failures.entries()) {
139
- lines.push(`### ${index + 1}. \`${failure.category}\` — ${failure.matcher}`, "");
140
- lines.push(failure.message, "");
141
- if (failure.evidence) {
142
- lines.push("```", failure.evidence, "```", "");
143
- }
144
- lines.push(`_${whyHint(failure.category)}_`, "");
145
- }
146
- const trace = options.trace ?? options.result.trace;
147
- if (trace) {
148
- const skills = trace.skillsInvoked?.length ? trace.skillsInvoked.join(", ") : "(none)";
149
- const breakdown = options.result.usage;
150
- const flat = breakdown?.total ?? trace.usage;
151
- const usageLines = [];
152
- if (breakdown?.total?.totalTokens !== undefined) {
153
- const t = breakdown.total;
154
- usageLines.push(`- **tokens (total):** ${t.totalTokens?.toLocaleString() ?? "?"} (${t.inputTokens?.toLocaleString() ?? "?"} in / ${t.outputTokens?.toLocaleString() ?? "?"} out)`);
155
- }
156
- else if (flat) {
157
- usageLines.push(flat.totalTokens !== undefined
158
- ? `- totalTokens: ${flat.totalTokens}`
159
- : `- usage: in=${flat.inputTokens ?? "?"} out=${flat.outputTokens ?? "?"}`);
160
- }
161
- else {
162
- usageLines.push("- totalTokens: (none)");
163
- }
164
- if (breakdown?.agent) {
165
- usageLines.push(`- **tokens (agent):** ${breakdown.agent.totalTokens?.toLocaleString() ?? "?"} total`);
166
- }
167
- if (breakdown?.judge) {
168
- usageLines.push(`- **tokens (judge):** ${breakdown.judge.totalTokens?.toLocaleString() ?? "?"} total`);
169
- }
170
- lines.push("**Trace stats:**", "", `- messages: ${trace.messages.length}`, `- toolCalls: ${trace.toolCalls.length}`, `- skillsInvoked: ${skills}`, `- routing.tier: ${trace.routing?.tier ?? "(none)"}`, ...usageLines, "");
171
- }
172
- return lines;
173
- }
174
- export function formatDebugSummaryMarkdown(options) {
175
- const verdict = options.result.skipped ? "SKIPPED" : options.result.passed ? "PASSED" : "FAILED";
176
- const lines = [
177
- `# ${options.scenario.name}`,
178
- "",
179
- `- **verdict:** ${verdict}`,
180
- `- **durationMs:** ${options.result.durationMs}`,
181
- `- **failures:** ${options.result.failures.length}`,
182
- "",
183
- ...formatDebugWhySection({ result: options.result, trace: options.trace }),
184
- ];
185
- return `${lines.join("\n")}\n`;
186
- }
187
- export function formatTranscriptMarkdown(options) {
188
- const result = options.result ?? {
189
- suite: "",
190
- scenario: options.scenario.name,
191
- passed: options.failures.length === 0,
192
- failures: options.failures,
193
- durationMs: 0,
194
- judgeVerdicts: options.judgeVerdicts,
195
- trace: options.trace,
196
- };
197
- const lines = [
198
- `# ${options.scenario.name}`,
199
- "",
200
- ...formatDebugWhySection({ result, trace: options.trace }),
201
- "## Prompt",
202
- "",
203
- "```",
204
- options.scenario.prompt,
205
- "```",
206
- "",
207
- "## Rubric",
208
- "",
209
- "```json",
210
- JSON.stringify(options.scenario.rubric, null, 2),
211
- "```",
212
- "",
213
- "## Transcript",
214
- "",
215
- ];
216
- const trace = options.trace;
217
- if (!trace) {
218
- lines.push("_No trace available._", "");
219
- }
220
- else {
221
- if (trace.routing) {
222
- lines.push("### Routing", "", "```json", JSON.stringify(trace.routing, null, 2), "```", "");
223
- }
224
- const events = [
225
- ...trace.messages.map((message, index) => ({
226
- kind: "message",
227
- seq: message.seq ?? index,
228
- message,
229
- })),
230
- ...trace.toolCalls.map((tool, index) => ({
231
- kind: "tool",
232
- seq: tool.seq ?? 10_000 + index,
233
- tool,
234
- })),
235
- ].sort((a, b) => a.seq - b.seq);
236
- for (const event of events) {
237
- if (event.kind === "message") {
238
- lines.push(`### ${event.message.role}`, "", event.message.content, "");
239
- }
240
- else {
241
- lines.push(`### tool: ${event.tool.name}`, "", "**args**", "", "```", compactText(event.tool.args, TOOL_ARG_COMPACT_CHARS), "```", "");
242
- if (event.tool.result !== undefined) {
243
- lines.push("**result**", "", "```", compactText(event.tool.result, TOOL_RESULT_MAX_CHARS), "```", "");
244
- }
245
- }
246
- }
247
- if (trace.shellCommands.length > 0) {
248
- lines.push("### Shell commands", "");
249
- for (const cmd of trace.shellCommands) {
250
- lines.push(`- \`${cmd}\``);
251
- }
252
- lines.push("");
253
- }
254
- }
255
- if (options.judgeVerdicts?.length) {
256
- lines.push("## Judge", "");
257
- for (const verdict of options.judgeVerdicts) {
258
- const mark = verdict.pass ? "PASS" : "FAIL";
259
- lines.push(`### ${mark} \`${verdict.id}\``, "", verdict.question, "", verdict.rationale, "");
260
- if (verdict.infraError) {
261
- lines.push(`**infraError:** ${verdict.infraError}`, "");
262
- }
263
- if (verdict.rawSdkStatus) {
264
- lines.push(`**rawSdkStatus:** ${verdict.rawSdkStatus}`, "");
265
- }
266
- if (verdict.sdkError) {
267
- lines.push(`**sdkError:** ${JSON.stringify(verdict.sdkError)}`, "");
268
- }
269
- }
270
- }
271
- if (options.failures.length > 0) {
272
- lines.push("## Failures", "");
273
- for (const failure of options.failures) {
274
- lines.push(`- **${failure.category}** \`${failure.matcher}\`: ${failure.message}`);
275
- if (failure.evidence) {
276
- lines.push(` - evidence: ${failure.evidence}`);
277
- }
278
- }
279
- lines.push("");
280
- }
281
- return `${lines.join("\n")}\n`;
282
- }
283
- export function getDebugBundleDir(stagingSessionId, suiteName, scenarioName, getSessionRoot, host) {
284
- const slug = scenarioArtifactSlug(scenarioName);
285
- const leaf = host ? `${slug}.${host}.debug` : `${slug}.debug`;
286
- return join(getSessionRoot(stagingSessionId), suiteName, leaf);
287
- }
288
- /** Persist a scenario debug bundle (trace, failures, transcript, judge, env, rerun). */
289
- export async function writeDebugBundle(options) {
290
- const { dir, result, trace, scenario, environment, rerun } = options;
291
- await mkdir(dir, { recursive: true });
292
- const failuresPath = join(dir, "failures.json");
293
- const transcriptPath = join(dir, "transcript.md");
294
- const environmentPath = join(dir, "environment.json");
295
- const rerunPath = join(dir, "rerun.sh");
296
- const summaryPath = join(dir, "summary.md");
297
- const scenarioPath = join(dir, "scenario.json");
298
- const resultPath = join(dir, "result.json");
299
- const effectiveTrace = trace ?? result.trace;
300
- await writeFile(failuresPath, `${JSON.stringify(result.failures, null, 2)}\n`, "utf8");
301
- await writeFile(environmentPath, `${JSON.stringify(environment, null, 2)}\n`, "utf8");
302
- await writeFile(summaryPath, formatDebugSummaryMarkdown({ scenario, result, trace: effectiveTrace }), "utf8");
303
- await writeFile(scenarioPath, `${JSON.stringify({
304
- name: scenario.name,
305
- prompt: scenario.prompt,
306
- rubric: scenario.rubric,
307
- workspace: scenario.workspace,
308
- allowUserSkills: scenario.allowUserSkills,
309
- seedPatch: scenario.seedPatch,
310
- seedStageOnly: scenario.seedStageOnly,
311
- host: scenario.host,
312
- skip: scenario.skip,
313
- compare: scenario.compare,
314
- }, null, 2)}\n`, "utf8");
315
- await writeFile(resultPath, `${JSON.stringify({
316
- suite: result.suite,
317
- scenario: result.scenario,
318
- passed: result.passed,
319
- skipped: result.skipped,
320
- durationMs: result.durationMs,
321
- failures: result.failures,
322
- usage: result.usage,
323
- agentUsage: result.agentUsage ?? result.usage?.agent,
324
- judgeUsage: result.judgeUsage ?? result.usage?.judge,
325
- skillsInvoked: effectiveTrace?.skillsInvoked ?? [],
326
- routing: effectiveTrace?.routing,
327
- messageCount: effectiveTrace?.messages.length ?? 0,
328
- toolCallCount: effectiveTrace?.toolCalls.length ?? 0,
329
- artifacts: effectiveTrace?.artifacts ?? {},
330
- compare: result.compare,
331
- }, null, 2)}\n`, "utf8");
332
- if (trace) {
333
- await writeFile(transcriptPath, formatTranscriptMarkdown({
334
- scenario,
335
- trace,
336
- failures: result.failures,
337
- judgeVerdicts: result.judgeVerdicts,
338
- result,
339
- }), "utf8");
340
- const payload = {
341
- messages: trace.messages,
342
- toolCalls: trace.toolCalls,
343
- shellCommands: trace.shellCommands,
344
- gitDiff: trace.gitDiff,
345
- prBody: trace.prBody,
346
- artifacts: trace.artifacts,
347
- routing: trace.routing,
348
- skillsInvoked: trace.skillsInvoked,
349
- assistantTextBeforeTools: trace.assistantTextBeforeTools,
350
- usage: trace.usage,
351
- usageBreakdown: result.usage,
352
- judgeVerdicts: trace.judgeVerdicts,
353
- };
354
- await writeFile(join(dir, "trace.json"), `${JSON.stringify(payload, null, 2)}\n`, "utf8");
355
- }
356
- else {
357
- // Parent rewrite after a failed staging load must not clobber a richer child bundle.
358
- let transcriptExists = false;
359
- try {
360
- await access(transcriptPath);
361
- transcriptExists = true;
362
- }
363
- catch {
364
- transcriptExists = false;
365
- }
366
- if (!transcriptExists) {
367
- await writeFile(transcriptPath, formatTranscriptMarkdown({
368
- scenario,
369
- trace,
370
- failures: result.failures,
371
- judgeVerdicts: result.judgeVerdicts,
372
- result,
373
- }), "utf8");
374
- }
375
- }
376
- if (result.judgeVerdicts?.length) {
377
- await writeFile(join(dir, "judge-debug.json"), `${JSON.stringify(result.judgeVerdicts, null, 2)}\n`, "utf8");
378
- }
379
- const rerunBody = [
380
- "#!/usr/bin/env bash",
381
- "set -euo pipefail",
382
- `# Re-run failed scenario ${shellCommentText(scenario.name)}`,
383
- `# Requires host auth. Default is subscription after agent-test login. Pass --auth-mode api-key plus the host API key to bill the API.`,
384
- `cd ${shellQuote(rerun.cwd)}`,
385
- buildRerunCommand(rerun),
386
- "",
387
- ].join("\n");
388
- await writeFile(rerunPath, rerunBody, "utf8");
389
- await chmod(rerunPath, 0o755);
390
- return dir;
391
- }
392
- //# sourceMappingURL=debug-bundle.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"debug-bundle.js","sourceRoot":"","sources":["../src/debug-bundle.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACnE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAIjC,OAAO,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAyDzD,MAAM,mBAAmB,GAAG;IAC3B,oBAAoB;IACpB,0BAA0B;IAC1B,sBAAsB;IACtB,8BAA8B;IAC9B,kBAAkB;IAClB,uBAAuB;IACvB,2BAA2B;IAC3B,wBAAwB;IACxB,uBAAuB;IACvB,+BAA+B;IAC/B,6BAA6B;IAC7B,kBAAkB;IAClB,kBAAkB;CACT,CAAC;AAEX,MAAM,sBAAsB,GAAG,GAAG,CAAC;AACnC,MAAM,qBAAqB,GAAG,KAAK,CAAC;AAEpC,kDAAkD;AAClD,MAAM,UAAU,UAAU,CAAC,KAAa;IACvC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,IAAI,CAAC;IACb,CAAC;IACD,IAAI,0BAA0B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5C,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC;AAC5C,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,gBAAgB,CAAC,KAAa;IAC7C,OAAO,KAAK,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;AAC9C,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,OAA0B;IAC3D,MAAM,IAAI,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IAC/B,IAAI,CAAC,IAAI,CACR,cAAc,EACd,OAAO,CAAC,SAAS,EACjB,SAAS,EACT,OAAO,CAAC,KAAK,EACb,YAAY,EACZ,OAAO,CAAC,QAAQ,CAChB,CAAC;IACF,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;QACxB,IAAI,CAAC,IAAI,CAAC,eAAe,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;IAChD,CAAC;IACD,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;QAClB,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,IAAI,CAAC,CAAC;IACnC,CAAC;IACD,KAAK,MAAM,UAAU,IAAI,OAAO,CAAC,cAAc,IAAI,EAAE,EAAE,CAAC;QACvD,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC;IACpC,CAAC;IACD,IAAI,OAAO,CAAC,KAAK,KAAK,KAAK,EAAE,CAAC;QAC7B,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IACzB,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,KAAK,KAAK,EAAE,CAAC;QAChC,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAC5B,CAAC;IACD,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QACvB,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;IAC3B,CAAC;SAAM,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;QAC5C,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;IACtD,CAAC;IACD,IAAI,OAAO,CAAC,cAAc,EAAE,CAAC;QAC5B,IAAI,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;IACjC,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;QACtB,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC5C,CAAC;IACD,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACrB,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;QACtB,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IAC5C,CAAC;IACD,IAAI,OAAO,CAAC,cAAc,KAAK,KAAK,EAAE,CAAC;QACtC,IAAI,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAChC,CAAC;IACD,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAC1E,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,OAQvC;IACA,MAAM,YAAY,GAAuC,EAAE,CAAC;IAC5D,KAAK,MAAM,GAAG,IAAI,mBAAmB,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC/B,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;YACzB,YAAY,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC;QAC3B,CAAC;IACF,CAAC;IACD,OAAO;QACN,WAAW,EAAE,OAAO,CAAC,QAAQ,CAAC,IAAI;QAClC,UAAU,EAAE,OAAO,CAAC,QAAQ,CAAC,GAAG;QAChC,cAAc,EAAE,OAAO,CAAC,cAAc;QACtC,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,eAAe,EAAE,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,EAAE,IAAI,EAAE,CAAC;QAC5D,gBAAgB,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;QAChD,gBAAgB,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;QAChD,sBAAsB,EAAE,OAAO,CAAC,GAAG,CAAC,wBAAwB;QAC5D,kBAAkB,EAAE,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,IAAI,EAAE,CAAC;QAClE,gBAAgB,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;QAChD,aAAa,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe;QAC1C,SAAS,EAAE,OAAO,CAAC,SAAS;QAC5B,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,YAAY;KACZ,CAAC;AACH,CAAC;AAED,SAAS,WAAW,CAAC,KAAc,EAAE,QAAgB;IACpD,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;QACzB,OAAO,EAAE,CAAC;IACX,CAAC;IACD,MAAM,GAAG,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACtE,IAAI,GAAG,CAAC,MAAM,IAAI,QAAQ,EAAE,CAAC;QAC5B,OAAO,GAAG,CAAC;IACZ,CAAC;IACD,OAAO,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,GAAG,CAAC;AACrC,CAAC;AAED,SAAS,OAAO,CAAC,QAAyB;IACzC,QAAQ,QAAQ,EAAE,CAAC;QAClB,KAAK,eAAe;YACnB,OAAO,iHAAiH,CAAC;QAC1H,KAAK,eAAe;YACnB,OAAO,yHAAyH,CAAC;QAClI,KAAK,aAAa;YACjB,OAAO,6FAA6F,CAAC;QACtG,KAAK,aAAa;YACjB,OAAO,uEAAuE,CAAC;QAChF,KAAK,iBAAiB;YACrB,OAAO,4FAA4F,CAAC;QACrG;YACC,OAAO,8FAA8F,CAAC;IACxG,CAAC;AACF,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,OAGrC;IACA,MAAM,KAAK,GAAa,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IACvC,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;QACrC,OAAO,KAAK,CAAC;IACd,CAAC;IACD,IAAI,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1C,KAAK,CAAC,IAAI,CAAC,qCAAqC,EAAE,EAAE,CAAC,CAAC;QACtD,OAAO,KAAK,CAAC;IACd,CAAC;IACD,KAAK,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,EAAE,EAAE,CAAC;QAClE,KAAK,CAAC,IAAI,CAAC,OAAO,KAAK,GAAG,CAAC,OAAO,OAAO,CAAC,QAAQ,QAAQ,OAAO,CAAC,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;QACjF,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QAChC,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;YACtB,KAAK,CAAC,IAAI,CAAC,KAAK,EAAE,OAAO,CAAC,QAAQ,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;QAChD,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,IAAI,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IAClD,CAAC;IACD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC;IACpD,IAAI,KAAK,EAAE,CAAC;QACX,MAAM,MAAM,GAAG,KAAK,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;QACvF,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC;QACvC,MAAM,IAAI,GAAG,SAAS,EAAE,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC;QAC7C,MAAM,UAAU,GAAa,EAAE,CAAC;QAChC,IAAI,SAAS,EAAE,KAAK,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;YACjD,MAAM,CAAC,GAAG,SAAS,CAAC,KAAK,CAAC;YAC1B,UAAU,CAAC,IAAI,CACd,yBAAyB,CAAC,CAAC,WAAW,EAAE,cAAc,EAAE,IAAI,GAAG,KAAK,CAAC,CAAC,WAAW,EAAE,cAAc,EAAE,IAAI,GAAG,SAAS,CAAC,CAAC,YAAY,EAAE,cAAc,EAAE,IAAI,GAAG,OAAO,CACjK,CAAC;QACH,CAAC;aAAM,IAAI,IAAI,EAAE,CAAC;YACjB,UAAU,CAAC,IAAI,CACd,IAAI,CAAC,WAAW,KAAK,SAAS;gBAC7B,CAAC,CAAC,kBAAkB,IAAI,CAAC,WAAW,EAAE;gBACtC,CAAC,CAAC,eAAe,IAAI,CAAC,WAAW,IAAI,GAAG,QAAQ,IAAI,CAAC,YAAY,IAAI,GAAG,EAAE,CAC3E,CAAC;QACH,CAAC;aAAM,CAAC;YACP,UAAU,CAAC,IAAI,CAAC,uBAAuB,CAAC,CAAC;QAC1C,CAAC;QACD,IAAI,SAAS,EAAE,KAAK,EAAE,CAAC;YACtB,UAAU,CAAC,IAAI,CACd,yBAAyB,SAAS,CAAC,KAAK,CAAC,WAAW,EAAE,cAAc,EAAE,IAAI,GAAG,QAAQ,CACrF,CAAC;QACH,CAAC;QACD,IAAI,SAAS,EAAE,KAAK,EAAE,CAAC;YACtB,UAAU,CAAC,IAAI,CACd,yBAAyB,SAAS,CAAC,KAAK,CAAC,WAAW,EAAE,cAAc,EAAE,IAAI,GAAG,QAAQ,CACrF,CAAC;QACH,CAAC;QACD,KAAK,CAAC,IAAI,CACT,kBAAkB,EAClB,EAAE,EACF,eAAe,KAAK,CAAC,QAAQ,CAAC,MAAM,EAAE,EACtC,gBAAgB,KAAK,CAAC,SAAS,CAAC,MAAM,EAAE,EACxC,oBAAoB,MAAM,EAAE,EAC5B,mBAAmB,KAAK,CAAC,OAAO,EAAE,IAAI,IAAI,QAAQ,EAAE,EACpD,GAAG,UAAU,EACb,EAAE,CACF,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACd,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,OAI1C;IACA,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;IACjG,MAAM,KAAK,GAAa;QACvB,KAAK,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE;QAC5B,EAAE;QACF,kBAAkB,OAAO,EAAE;QAC3B,qBAAqB,OAAO,CAAC,MAAM,CAAC,UAAU,EAAE;QAChD,mBAAmB,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,EAAE;QACnD,EAAE;QACF,GAAG,qBAAqB,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;KAC1E,CAAC;IACF,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,OAMxC;IACA,MAAM,MAAM,GAAmB,OAAO,CAAC,MAAM,IAAI;QAChD,KAAK,EAAE,EAAE;QACT,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC,IAAI;QAC/B,MAAM,EAAE,OAAO,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC;QACrC,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,UAAU,EAAE,CAAC;QACb,aAAa,EAAE,OAAO,CAAC,aAAa;QACpC,KAAK,EAAE,OAAO,CAAC,KAAK;KACpB,CAAC;IAEF,MAAM,KAAK,GAAa;QACvB,KAAK,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE;QAC5B,EAAE;QACF,GAAG,qBAAqB,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;QAC1D,WAAW;QACX,EAAE;QACF,KAAK;QACL,OAAO,CAAC,QAAQ,CAAC,MAAM;QACvB,KAAK;QACL,EAAE;QACF,WAAW;QACX,EAAE;QACF,SAAS;QACT,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;QAChD,KAAK;QACL,EAAE;QACF,eAAe;QACf,EAAE;KACF,CAAC;IAEF,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,IAAI,CAAC,KAAK,EAAE,CAAC;QACZ,KAAK,CAAC,IAAI,CAAC,uBAAuB,EAAE,EAAE,CAAC,CAAC;IACzC,CAAC;SAAM,CAAC;QACP,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YACnB,KAAK,CAAC,IAAI,CAAC,aAAa,EAAE,EAAE,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;QAC7F,CAAC;QAMD,MAAM,MAAM,GAAY;YACvB,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;gBAC1C,IAAI,EAAE,SAAkB;gBACxB,GAAG,EAAE,OAAO,CAAC,GAAG,IAAI,KAAK;gBACzB,OAAO;aACP,CAAC,CAAC;YACH,GAAG,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;gBACxC,IAAI,EAAE,MAAe;gBACrB,GAAG,EAAE,IAAI,CAAC,GAAG,IAAI,MAAM,GAAG,KAAK;gBAC/B,IAAI;aACJ,CAAC,CAAC;SACH,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QAEhC,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC5B,IAAI,KAAK,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;gBAC9B,KAAK,CAAC,IAAI,CAAC,OAAO,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,KAAK,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YACxE,CAAC;iBAAM,CAAC;gBACP,KAAK,CAAC,IAAI,CACT,aAAa,KAAK,CAAC,IAAI,CAAC,IAAI,EAAE,EAC9B,EAAE,EACF,UAAU,EACV,EAAE,EACF,KAAK,EACL,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,EAAE,sBAAsB,CAAC,EACpD,KAAK,EACL,EAAE,CACF,CAAC;gBACF,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;oBACrC,KAAK,CAAC,IAAI,CACT,YAAY,EACZ,EAAE,EACF,KAAK,EACL,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,qBAAqB,CAAC,EACrD,KAAK,EACL,EAAE,CACF,CAAC;gBACH,CAAC;YACF,CAAC;QACF,CAAC;QAED,IAAI,KAAK,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;YACrC,KAAK,MAAM,GAAG,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;gBACvC,KAAK,CAAC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,CAAC;YAC5B,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAChB,CAAC;IACF,CAAC;IAED,IAAI,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,CAAC;QACnC,KAAK,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;QAC3B,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,aAAa,EAAE,CAAC;YAC7C,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;YAC5C,KAAK,CAAC,IAAI,CAAC,OAAO,IAAI,MAAM,OAAO,CAAC,EAAE,IAAI,EAAE,EAAE,EAAE,OAAO,CAAC,QAAQ,EAAE,EAAE,EAAE,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;YAC7F,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;gBACxB,KAAK,CAAC,IAAI,CAAC,mBAAmB,OAAO,CAAC,UAAU,EAAE,EAAE,EAAE,CAAC,CAAC;YACzD,CAAC;YACD,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;gBAC1B,KAAK,CAAC,IAAI,CAAC,qBAAqB,OAAO,CAAC,YAAY,EAAE,EAAE,EAAE,CAAC,CAAC;YAC7D,CAAC;YACD,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACtB,KAAK,CAAC,IAAI,CAAC,iBAAiB,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;YACrE,CAAC;QACF,CAAC;IACF,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjC,KAAK,CAAC,IAAI,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC;QAC9B,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC,OAAO,OAAO,CAAC,QAAQ,QAAQ,OAAO,CAAC,OAAO,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;YACnF,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACtB,KAAK,CAAC,IAAI,CAAC,iBAAiB,OAAO,CAAC,QAAQ,EAAE,CAAC,CAAC;YACjD,CAAC;QACF,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAChB,CAAC;IAED,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,iBAAiB,CAChC,gBAAwB,EACxB,SAAiB,EACjB,YAAoB,EACpB,cAA6C,EAC7C,IAAa;IAEb,MAAM,IAAI,GAAG,oBAAoB,CAAC,YAAY,CAAC,CAAC;IAChD,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC,GAAG,IAAI,QAAQ,CAAC;IAC9D,OAAO,IAAI,CAAC,cAAc,CAAC,gBAAgB,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;AAChE,CAAC;AAED,wFAAwF;AACxF,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,OAAgC;IACtE,MAAM,EAAE,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,GAAG,OAAO,CAAC;IACrE,MAAM,KAAK,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAEtC,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,EAAE,eAAe,CAAC,CAAC;IAChD,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,EAAE,eAAe,CAAC,CAAC;IAClD,MAAM,eAAe,GAAG,IAAI,CAAC,GAAG,EAAE,kBAAkB,CAAC,CAAC;IACtD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;IACxC,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;IAC5C,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,EAAE,eAAe,CAAC,CAAC;IAChD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;IAE5C,MAAM,cAAc,GAAG,KAAK,IAAI,MAAM,CAAC,KAAK,CAAC;IAE7C,MAAM,SAAS,CAAC,YAAY,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACvF,MAAM,SAAS,CAAC,eAAe,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACtF,MAAM,SAAS,CACd,WAAW,EACX,0BAA0B,CAAC,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,cAAc,EAAE,CAAC,EACvE,MAAM,CACN,CAAC;IACF,MAAM,SAAS,CACd,YAAY,EACZ,GAAG,IAAI,CAAC,SAAS,CAChB;QACC,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,SAAS,EAAE,QAAQ,CAAC,SAAS;QAC7B,eAAe,EAAE,QAAQ,CAAC,eAAe;QACzC,SAAS,EAAE,QAAQ,CAAC,SAAS;QAC7B,aAAa,EAAE,QAAQ,CAAC,aAAa;QACrC,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,OAAO,EAAE,QAAQ,CAAC,OAAO;KACzB,EACD,IAAI,EACJ,CAAC,CACD,IAAI,EACL,MAAM,CACN,CAAC;IACF,MAAM,SAAS,CACd,UAAU,EACV,GAAG,IAAI,CAAC,SAAS,CAChB;QACC,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,KAAK,EAAE,KAAK;QACpD,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,KAAK,EAAE,KAAK;QACpD,aAAa,EAAE,cAAc,EAAE,aAAa,IAAI,EAAE;QAClD,OAAO,EAAE,cAAc,EAAE,OAAO;QAChC,YAAY,EAAE,cAAc,EAAE,QAAQ,CAAC,MAAM,IAAI,CAAC;QAClD,aAAa,EAAE,cAAc,EAAE,SAAS,CAAC,MAAM,IAAI,CAAC;QACpD,SAAS,EAAE,cAAc,EAAE,SAAS,IAAI,EAAE;QAC1C,OAAO,EAAE,MAAM,CAAC,OAAO;KACvB,EACD,IAAI,EACJ,CAAC,CACD,IAAI,EACL,MAAM,CACN,CAAC;IAEF,IAAI,KAAK,EAAE,CAAC;QACX,MAAM,SAAS,CACd,cAAc,EACd,wBAAwB,CAAC;YACxB,QAAQ;YACR,KAAK;YACL,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,aAAa,EAAE,MAAM,CAAC,aAAa;YACnC,MAAM;SACN,CAAC,EACF,MAAM,CACN,CAAC;QACF,MAAM,OAAO,GAAG;YACf,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,aAAa,EAAE,KAAK,CAAC,aAAa;YAClC,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,aAAa,EAAE,KAAK,CAAC,aAAa;YAClC,wBAAwB,EAAE,KAAK,CAAC,wBAAwB;YACxD,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,cAAc,EAAE,MAAM,CAAC,KAAK;YAC5B,aAAa,EAAE,KAAK,CAAC,aAAa;SAClC,CAAC;QACF,MAAM,SAAS,CAAC,IAAI,CAAC,GAAG,EAAE,YAAY,CAAC,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC3F,CAAC;SAAM,CAAC;QACP,qFAAqF;QACrF,IAAI,gBAAgB,GAAG,KAAK,CAAC;QAC7B,IAAI,CAAC;YACJ,MAAM,MAAM,CAAC,cAAc,CAAC,CAAC;YAC7B,gBAAgB,GAAG,IAAI,CAAC;QACzB,CAAC;QAAC,MAAM,CAAC;YACR,gBAAgB,GAAG,KAAK,CAAC;QAC1B,CAAC;QACD,IAAI,CAAC,gBAAgB,EAAE,CAAC;YACvB,MAAM,SAAS,CACd,cAAc,EACd,wBAAwB,CAAC;gBACxB,QAAQ;gBACR,KAAK;gBACL,QAAQ,EAAE,MAAM,CAAC,QAAQ;gBACzB,aAAa,EAAE,MAAM,CAAC,aAAa;gBACnC,MAAM;aACN,CAAC,EACF,MAAM,CACN,CAAC;QACH,CAAC;IACF,CAAC;IAED,IAAI,MAAM,CAAC,aAAa,EAAE,MAAM,EAAE,CAAC;QAClC,MAAM,SAAS,CACd,IAAI,CAAC,GAAG,EAAE,kBAAkB,CAAC,EAC7B,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EACpD,MAAM,CACN,CAAC;IACH,CAAC;IAED,MAAM,SAAS,GAAG;QACjB,qBAAqB;QACrB,mBAAmB;QACnB,4BAA4B,gBAAgB,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE;QAC7D,uIAAuI;QACvI,MAAM,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE;QAC7B,iBAAiB,CAAC,KAAK,CAAC;QACxB,EAAE;KACF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACb,MAAM,SAAS,CAAC,SAAS,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IAC9C,MAAM,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IAE9B,OAAO,GAAG,CAAC;AACZ,CAAC"}
@@ -1,2 +0,0 @@
1
- export declare function discoverSuites(suitesDir: string): Promise<string[]>;
2
- //# sourceMappingURL=discover-suites.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"discover-suites.d.ts","sourceRoot":"","sources":["../src/discover-suites.ts"],"names":[],"mappings":"AAGA,wBAAsB,cAAc,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAczE"}
@@ -1,19 +0,0 @@
1
- import { access, readdir } from "node:fs/promises";
2
- import { join } from "node:path";
3
- export async function discoverSuites(suitesDir) {
4
- const entries = await readdir(suitesDir, { withFileTypes: true });
5
- const paths = [];
6
- for (const entry of entries) {
7
- if (!entry.isDirectory()) {
8
- continue;
9
- }
10
- const scenarioPath = join(suitesDir, entry.name, "scenarios.json");
11
- try {
12
- await access(scenarioPath);
13
- paths.push(scenarioPath);
14
- }
15
- catch { }
16
- }
17
- return paths;
18
- }
19
- //# sourceMappingURL=discover-suites.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"discover-suites.js","sourceRoot":"","sources":["../src/discover-suites.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,SAAiB;IACrD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,SAAS,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAClE,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;YAC1B,SAAS;QACV,CAAC;QACD,MAAM,YAAY,GAAG,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,EAAE,gBAAgB,CAAC,CAAC;QACnE,IAAI,CAAC;YACJ,MAAM,MAAM,CAAC,YAAY,CAAC,CAAC;YAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC1B,CAAC;QAAC,MAAM,CAAC,CAAA,CAAC;IACX,CAAC;IACD,OAAO,KAAK,CAAC;AACd,CAAC"}
package/dist/doctor.d.ts DELETED
@@ -1,24 +0,0 @@
1
- import { type AgentHost, HEALTH_CHECK_PATH } from "@post-print/agent-harness";
2
- export interface DoctorReport {
3
- ok: boolean;
4
- nodeMajor: number;
5
- nodeOk: boolean;
6
- cliPresent: boolean;
7
- cursorApiKeySet: boolean;
8
- cursorSdkAuthFilePresent: boolean;
9
- cursorSdkPresent: boolean;
10
- anthropicApiKeySet: boolean;
11
- claudeBinPresent: boolean;
12
- openaiApiKeySet: boolean;
13
- openaiBinPresent: boolean;
14
- liveReady: boolean;
15
- health: typeof HEALTH_CHECK_PATH;
16
- messages: string[];
17
- }
18
- /** Local diagnostics for agent-test install and live-run readiness. */
19
- export declare function runDoctor(options?: {
20
- cliPath?: string;
21
- }): DoctorReport;
22
- /** Missing credential or binary for a live host adapter. Undefined when that host can run. */
23
- export declare function missingAgentAuth(host: AgentHost): string | undefined;
24
- //# sourceMappingURL=doctor.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"doctor.d.ts","sourceRoot":"","sources":["../src/doctor.ts"],"names":[],"mappings":"AAKA,OAAO,EACN,KAAK,SAAS,EAQd,iBAAiB,EAOjB,MAAM,2BAA2B,CAAC;AAEnC,MAAM,WAAW,YAAY;IAC5B,EAAE,EAAE,OAAO,CAAC;IACZ,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,OAAO,CAAC;IAChB,UAAU,EAAE,OAAO,CAAC;IACpB,eAAe,EAAE,OAAO,CAAC;IACzB,wBAAwB,EAAE,OAAO,CAAC;IAClC,gBAAgB,EAAE,OAAO,CAAC;IAC1B,kBAAkB,EAAE,OAAO,CAAC;IAC5B,gBAAgB,EAAE,OAAO,CAAC;IAC1B,eAAe,EAAE,OAAO,CAAC;IACzB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,SAAS,EAAE,OAAO,CAAC;IACnB,MAAM,EAAE,OAAO,iBAAiB,CAAC;IACjC,QAAQ,EAAE,MAAM,EAAE,CAAC;CACnB;AA+CD,uEAAuE;AACvE,wBAAgB,SAAS,CAAC,OAAO,CAAC,EAAE;IAAE,OAAO,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,YAAY,CAuLtE;AAED,8FAA8F;AAC9F,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,SAAS,GAAG,MAAM,GAAG,SAAS,CAqDpE"}