agentv 4.21.0 → 4.22.0

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.
@@ -15,7 +15,7 @@ import {
15
15
  resolveWorkspaceOrFilePath,
16
16
  toSnakeCaseDeep,
17
17
  writeArtifactsFromResults
18
- } from "./chunk-FZUNMXBT.js";
18
+ } from "./chunk-CM2AVQAM.js";
19
19
  import {
20
20
  ResponseCache,
21
21
  RunBudgetTracker,
@@ -31,7 +31,7 @@ import {
31
31
  shouldSkipCacheForTemperature,
32
32
  stageResultsArtifacts,
33
33
  syncResultsRepo
34
- } from "./chunk-27VT2KU2.js";
34
+ } from "./chunk-762UZSGR.js";
35
35
  import {
36
36
  CLI_PLACEHOLDERS,
37
37
  COMMON_TARGET_SETTINGS,
@@ -62,68 +62,7 @@ import {
62
62
  subscribeToCopilotSdkLogEntries,
63
63
  subscribeToPiLogEntries,
64
64
  toCamelCaseDeep
65
- } from "./chunk-ERSBQAGK.js";
66
-
67
- // package.json
68
- var package_default = {
69
- name: "agentv",
70
- version: "4.21.0",
71
- description: "CLI entry point for AgentV",
72
- type: "module",
73
- repository: {
74
- type: "git",
75
- url: "https://github.com/EntityProcess/agentv.git"
76
- },
77
- homepage: "https://agentv.dev",
78
- bugs: {
79
- url: "https://github.com/EntityProcess/agentv/issues"
80
- },
81
- bin: {
82
- agentv: "./dist/cli.js"
83
- },
84
- files: ["dist", "README.md"],
85
- scripts: {
86
- dev: "bun src/cli.ts",
87
- build: "tsup && bun run copy-readme",
88
- "copy-readme": `bun -e "import { cpSync } from 'fs'; cpSync('../../README.md', 'README.md')"`,
89
- prepublishOnly: `node -e "if(process.env.ALLOW_PUBLISH!=='1'){console.error('ERROR: Use bun run publish:next, then bun run promote:latest');process.exit(1)}"`,
90
- typecheck: "tsc --noEmit",
91
- lint: "biome check .",
92
- format: "biome format --write .",
93
- fix: "biome check --write .",
94
- test: "bun test",
95
- "test:watch": "bun test --watch"
96
- },
97
- dependencies: {
98
- "@ai-sdk/openai": "^3.0.0",
99
- "@anthropic-ai/claude-agent-sdk": "^0.2.49",
100
- "@github/copilot-sdk": "^0.1.25",
101
- "@hono/node-server": "^1.19.11",
102
- "@inquirer/prompts": "^8.2.1",
103
- "@openai/codex-sdk": "^0.104.0",
104
- "cmd-ts": "^0.14.3",
105
- dotenv: "^16.4.5",
106
- "fast-glob": "^3.3.3",
107
- hono: "^4.12.9",
108
- json5: "^2.2.3",
109
- micromatch: "^4.0.8",
110
- semver: "^7.7.4",
111
- yaml: "^2.8.3"
112
- },
113
- peerDependencies: {
114
- "@mariozechner/pi-coding-agent": "^0.62.0"
115
- },
116
- peerDependenciesMeta: {
117
- "@mariozechner/pi-coding-agent": {
118
- optional: true
119
- }
120
- },
121
- devDependencies: {
122
- "@agentv/core": "workspace:*",
123
- "@types/semver": "^7.7.1",
124
- execa: "^9.3.0"
125
- }
126
- };
65
+ } from "./chunk-4YZDPAGZ.js";
127
66
 
128
67
  // src/commands/eval/shared.ts
129
68
  import { constants } from "node:fs";
@@ -236,6 +175,67 @@ async function findRepoRoot(start) {
236
175
  return fallback;
237
176
  }
238
177
 
178
+ // package.json
179
+ var package_default = {
180
+ name: "agentv",
181
+ version: "4.22.0",
182
+ description: "CLI entry point for AgentV",
183
+ type: "module",
184
+ repository: {
185
+ type: "git",
186
+ url: "https://github.com/EntityProcess/agentv.git"
187
+ },
188
+ homepage: "https://agentv.dev",
189
+ bugs: {
190
+ url: "https://github.com/EntityProcess/agentv/issues"
191
+ },
192
+ bin: {
193
+ agentv: "./dist/cli.js"
194
+ },
195
+ files: ["dist", "README.md"],
196
+ scripts: {
197
+ dev: "bun src/cli.ts",
198
+ build: "tsup && bun run copy-readme",
199
+ "copy-readme": `bun -e "import { cpSync } from 'fs'; cpSync('../../README.md', 'README.md')"`,
200
+ prepublishOnly: `node -e "if(process.env.ALLOW_PUBLISH!=='1'){console.error('ERROR: Use bun run publish:next, then bun run promote:latest');process.exit(1)}"`,
201
+ typecheck: "tsc --noEmit",
202
+ lint: "biome check .",
203
+ format: "biome format --write .",
204
+ fix: "biome check --write .",
205
+ test: "bun test",
206
+ "test:watch": "bun test --watch"
207
+ },
208
+ dependencies: {
209
+ "@ai-sdk/openai": "^3.0.0",
210
+ "@anthropic-ai/claude-agent-sdk": "^0.2.49",
211
+ "@github/copilot-sdk": "^0.1.25",
212
+ "@hono/node-server": "^1.19.11",
213
+ "@inquirer/prompts": "^8.2.1",
214
+ "@openai/codex-sdk": "^0.104.0",
215
+ "cmd-ts": "^0.14.3",
216
+ dotenv: "^16.4.5",
217
+ "fast-glob": "^3.3.3",
218
+ hono: "^4.12.9",
219
+ json5: "^2.2.3",
220
+ micromatch: "^4.0.8",
221
+ semver: "^7.7.4",
222
+ yaml: "^2.8.3"
223
+ },
224
+ peerDependencies: {
225
+ "@mariozechner/pi-coding-agent": "^0.62.0"
226
+ },
227
+ peerDependenciesMeta: {
228
+ "@mariozechner/pi-coding-agent": {
229
+ optional: true
230
+ }
231
+ },
232
+ devDependencies: {
233
+ "@agentv/core": "workspace:*",
234
+ "@types/semver": "^7.7.1",
235
+ execa: "^9.3.0"
236
+ }
237
+ };
238
+
239
239
  // src/utils/targets.ts
240
240
  import { constants as constants2 } from "node:fs";
241
241
  import { access as access2 } from "node:fs/promises";
@@ -4483,6 +4483,7 @@ async function fileExists3(filePath) {
4483
4483
  var ANSI_YELLOW3 = "\x1B[33m";
4484
4484
  var ANSI_RED3 = "\x1B[31m";
4485
4485
  var ANSI_RESET3 = "\x1B[0m";
4486
+ var DRY_RUN_MOCK_RESPONSE = '{"score":1,"assertions":[],"checks":[],"overall_reasoning":"dry-run mock"}';
4486
4487
  function isTTY() {
4487
4488
  return process.stdout.isTTY ?? false;
4488
4489
  }
@@ -4584,7 +4585,7 @@ Errors in ${targetsFilePath}:`);
4584
4585
  name: `${targetDefinition.name}-dry-run`,
4585
4586
  graderTarget: void 0,
4586
4587
  config: {
4587
- response: '{"answer":"Mock dry-run response"}',
4588
+ response: DRY_RUN_MOCK_RESPONSE,
4588
4589
  delayMs: dryRunDelay,
4589
4590
  delayMinMs: dryRunDelayMin,
4590
4591
  delayMaxMs: dryRunDelayMax
@@ -4687,7 +4688,7 @@ Errors in ${targetsFilePath}:`);
4687
4688
  name: `${targetDefinition.name}-dry-run`,
4688
4689
  graderTarget: void 0,
4689
4690
  config: {
4690
- response: '{"answer":"Mock dry-run response"}',
4691
+ response: DRY_RUN_MOCK_RESPONSE,
4691
4692
  delayMs: dryRunDelay,
4692
4693
  delayMinMs: dryRunDelayMin,
4693
4694
  delayMaxMs: dryRunDelayMax
@@ -5041,7 +5042,9 @@ async function prepareFileMetadata(params) {
5041
5042
  name: `${targetDefinition.name}-dry-run`,
5042
5043
  graderTarget: void 0,
5043
5044
  config: {
5044
- response: '{"answer":"Mock dry-run response"}',
5045
+ // Schema-valid grader response so --dry-run works end-to-end with LLM graders.
5046
+ // Satisfies freeform (score), rubric (checks, overall_reasoning), and score-range (checks) without real LLM calls.
5047
+ response: '{"score":1,"assertions":[],"checks":[],"overall_reasoning":"dry-run mock"}',
5045
5048
  delayMs: options.dryRunDelay,
5046
5049
  delayMinMs: options.dryRunDelayMin,
5047
5050
  delayMaxMs: options.dryRunDelayMax
@@ -5432,7 +5435,7 @@ async function runEvalCommand(input) {
5432
5435
  const useFileExport = !!options.otelFile;
5433
5436
  if (options.exportOtel || useFileExport) {
5434
5437
  try {
5435
- const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-RSAA3T6F.js");
5438
+ const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-6LNJDZAI.js");
5436
5439
  let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
5437
5440
  let headers = {};
5438
5441
  if (options.otelBackend) {
@@ -5631,7 +5634,7 @@ async function runEvalCommand(input) {
5631
5634
  const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
5632
5635
  let transcriptProviderFactory;
5633
5636
  if (options.transcript) {
5634
- const { TranscriptProvider } = await import("./dist-RSAA3T6F.js");
5637
+ const { TranscriptProvider } = await import("./dist-6LNJDZAI.js");
5635
5638
  const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
5636
5639
  const totalTests = [...fileMetadata.values()].reduce(
5637
5640
  (sum, meta) => sum + meta.testCases.length,
@@ -5800,7 +5803,7 @@ async function runEvalCommand(input) {
5800
5803
  if (usesDefaultArtifactWorkspace && allResults.length > 0) {
5801
5804
  const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
5802
5805
  if (isResumeAppend) {
5803
- const { writePerTestArtifacts } = await import("./artifact-writer-E775664W.js");
5806
+ const { writePerTestArtifacts } = await import("./artifact-writer-S7S6K3XA.js");
5804
5807
  await writePerTestArtifacts(allResults, runDir, {
5805
5808
  experiment: normalizeExperimentName(options.experiment)
5806
5809
  });
@@ -5975,6 +5978,8 @@ function filterByCategory(files, category) {
5975
5978
  }
5976
5979
 
5977
5980
  export {
5981
+ resolveEvalPaths,
5982
+ findRepoRoot,
5978
5983
  package_default,
5979
5984
  parseResultManifest,
5980
5985
  resolveResultSourcePath,
@@ -5986,8 +5991,6 @@ export {
5986
5991
  fetchLatestVersion,
5987
5992
  performSelfUpdate,
5988
5993
  enforceRequiredVersion,
5989
- resolveEvalPaths,
5990
- findRepoRoot,
5991
5994
  c,
5992
5995
  padRight,
5993
5996
  padLeft,
@@ -6023,4 +6026,4 @@ export {
6023
6026
  getCategories,
6024
6027
  filterByCategory
6025
6028
  };
6026
- //# sourceMappingURL=chunk-GPGX24OI.js.map
6029
+ //# sourceMappingURL=chunk-5QF7PKZ4.js.map