agentv 2.5.8 → 2.7.1-next.1

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 (81) hide show
  1. package/README.md +128 -33
  2. package/dist/chunk-3L2L5GIL.js +51 -0
  3. package/dist/{chunk-BKMQNEUD.js.map → chunk-3L2L5GIL.js.map} +1 -1
  4. package/dist/{chunk-BKMQNEUD.js → chunk-5H446C7X.js} +2 -45
  5. package/dist/chunk-6MPKU3K7.js +5476 -0
  6. package/dist/chunk-6MPKU3K7.js.map +1 -0
  7. package/dist/{chunk-LJVS3JAK.js → chunk-BL4PVUAT.js} +6 -4
  8. package/dist/{chunk-LJVS3JAK.js.map → chunk-BL4PVUAT.js.map} +1 -1
  9. package/dist/{chunk-H7Z5TVCB.js → chunk-BWLYFF5N.js} +9824 -11251
  10. package/dist/chunk-BWLYFF5N.js.map +1 -0
  11. package/dist/chunk-C5GOHBQM.js +84 -0
  12. package/dist/chunk-C5GOHBQM.js.map +1 -0
  13. package/dist/chunk-FV32QHPB.js +565 -0
  14. package/dist/chunk-FV32QHPB.js.map +1 -0
  15. package/dist/chunk-H5FFZCKI.js +2957 -0
  16. package/dist/chunk-H5FFZCKI.js.map +1 -0
  17. package/dist/chunk-JK6V4KVD.js +114 -0
  18. package/dist/chunk-JK6V4KVD.js.map +1 -0
  19. package/dist/chunk-LRULMAAA.js +1711 -0
  20. package/dist/chunk-LRULMAAA.js.map +1 -0
  21. package/dist/chunk-SR4I5KET.js +1238 -0
  22. package/dist/chunk-SR4I5KET.js.map +1 -0
  23. package/dist/chunk-VQ2ZO7XJ.js +2098 -0
  24. package/dist/chunk-VQ2ZO7XJ.js.map +1 -0
  25. package/dist/chunk-XALGXSKB.js +21 -0
  26. package/dist/chunk-XALGXSKB.js.map +1 -0
  27. package/dist/cli.js +8 -2
  28. package/dist/cli.js.map +1 -1
  29. package/dist/dist-R3OCWGXH.js +257 -0
  30. package/dist/dist-R3OCWGXH.js.map +1 -0
  31. package/dist/esm-5Q4BZALM-5REQWAUV.js +924 -0
  32. package/dist/esm-5Q4BZALM-5REQWAUV.js.map +1 -0
  33. package/dist/esm-DX3WQKEN.js +32 -0
  34. package/dist/esm-DX3WQKEN.js.map +1 -0
  35. package/dist/esm-QNEMCJPL.js +933 -0
  36. package/dist/esm-QNEMCJPL.js.map +1 -0
  37. package/dist/esm-R77SNOF5.js +65 -0
  38. package/dist/esm-R77SNOF5.js.map +1 -0
  39. package/dist/esm-RVQPUGWH.js +1207 -0
  40. package/dist/esm-RVQPUGWH.js.map +1 -0
  41. package/dist/getMachineId-bsd-HSK5LZMG.js +41 -0
  42. package/dist/getMachineId-bsd-HSK5LZMG.js.map +1 -0
  43. package/dist/getMachineId-darwin-4DP6CCJV.js +41 -0
  44. package/dist/getMachineId-darwin-4DP6CCJV.js.map +1 -0
  45. package/dist/getMachineId-linux-44LJ5UJB.js +33 -0
  46. package/dist/getMachineId-linux-44LJ5UJB.js.map +1 -0
  47. package/dist/getMachineId-unsupported-NVK6IATM.js +24 -0
  48. package/dist/getMachineId-unsupported-NVK6IATM.js.map +1 -0
  49. package/dist/getMachineId-win-YZ36S7VA.js +43 -0
  50. package/dist/getMachineId-win-YZ36S7VA.js.map +1 -0
  51. package/dist/index.js +10 -2
  52. package/dist/interactive-33TCZXLF.js +333 -0
  53. package/dist/interactive-33TCZXLF.js.map +1 -0
  54. package/dist/otlp-json-file-exporter-77FDBRSY-EZAPHWP6.js +9 -0
  55. package/dist/otlp-json-file-exporter-77FDBRSY-EZAPHWP6.js.map +1 -0
  56. package/dist/simple-trace-file-exporter-S76DMABU-5FCJESD2.js +9 -0
  57. package/dist/simple-trace-file-exporter-S76DMABU-5FCJESD2.js.map +1 -0
  58. package/dist/src-2N5EJ2N6.js +1733 -0
  59. package/dist/src-2N5EJ2N6.js.map +1 -0
  60. package/dist/templates/.agents/skills/agentv-chat-to-eval/README.md +84 -0
  61. package/dist/templates/.agents/skills/agentv-chat-to-eval/SKILL.md +144 -0
  62. package/dist/templates/.agents/skills/agentv-chat-to-eval/examples/transcript-json.md +67 -0
  63. package/dist/templates/.agents/skills/agentv-chat-to-eval/examples/transcript-markdown.md +101 -0
  64. package/dist/templates/.agents/skills/agentv-eval-builder/SKILL.md +433 -0
  65. package/dist/templates/.agents/skills/agentv-eval-builder/references/config-schema.json +36 -0
  66. package/dist/templates/.agents/skills/agentv-eval-builder/references/custom-evaluators.md +118 -0
  67. package/dist/templates/.agents/skills/agentv-eval-builder/references/eval-schema.json +251 -0
  68. package/dist/templates/.agents/skills/agentv-eval-builder/references/rubric-evaluator.md +77 -0
  69. package/dist/templates/.agents/skills/agentv-eval-orchestrator/SKILL.md +50 -0
  70. package/dist/templates/.agents/skills/agentv-prompt-optimizer/SKILL.md +78 -0
  71. package/dist/templates/.claude/skills/agentv-eval-builder/SKILL.md +38 -13
  72. package/dist/templates/.claude/skills/agentv-eval-builder/references/custom-evaluators.md +9 -6
  73. package/dist/templates/.claude/skills/agentv-eval-builder/references/eval-schema.json +4 -4
  74. package/dist/templates/.claude/skills/agentv-eval-builder/references/rubric-evaluator.md +7 -9
  75. package/dist/{token-D3IYDJQZ.js → token-POXF46NU.js} +6 -4
  76. package/dist/{token-D3IYDJQZ.js.map → token-POXF46NU.js.map} +1 -1
  77. package/dist/{token-util-FWFPR2BV.js → token-util-6GWYZWGE.js} +4 -3
  78. package/dist/token-util-6GWYZWGE.js.map +1 -0
  79. package/package.json +9 -5
  80. package/dist/chunk-H7Z5TVCB.js.map +0 -1
  81. /package/dist/{token-util-FWFPR2BV.js.map → chunk-5H446C7X.js.map} +0 -0
@@ -5,25 +5,23 @@
5
5
  | Field | Type | Default | Description |
6
6
  |-------|------|---------|-------------|
7
7
  | `id` | string | auto-generated | Unique identifier |
8
- | `expected_outcome` | string | required* | Criterion being evaluated (*optional if `score_ranges` used) |
8
+ | `outcome` | string | required* | Criterion being evaluated (*optional if `score_ranges` used) |
9
9
  | `weight` | number | 1.0 | Relative importance |
10
10
  | `required` | boolean | true | Failing forces verdict to 'fail' (checklist mode) |
11
11
  | `required_min_score` | integer | - | Minimum 0-10 score to pass (score-range mode) |
12
12
  | `score_ranges` | map or array | - | Score range definitions for analytic scoring |
13
13
 
14
- `description` is a backward-compatible alias for `expected_outcome`.
15
-
16
14
  ## Checklist Mode
17
15
 
18
16
  ```yaml
19
17
  rubrics:
20
18
  - Mentions divide-and-conquer approach
21
19
  - id: complexity
22
- expected_outcome: States time complexity correctly
20
+ outcome: States time complexity correctly
23
21
  weight: 2.0
24
22
  required: true
25
23
  - id: examples
26
- expected_outcome: Includes code examples
24
+ outcome: Includes code examples
27
25
  weight: 1.0
28
26
  required: false
29
27
  ```
@@ -51,13 +49,13 @@ Array format is also accepted:
51
49
  ```yaml
52
50
  score_ranges:
53
51
  - score_range: [0, 2]
54
- expected_outcome: Critical bugs
52
+ outcome: Critical bugs
55
53
  - score_range: [3, 5]
56
- expected_outcome: Minor bugs
54
+ outcome: Minor bugs
57
55
  - score_range: [6, 8]
58
- expected_outcome: Correct with minor issues
56
+ outcome: Correct with minor issues
59
57
  - score_range: [9, 10]
60
- expected_outcome: Fully correct
58
+ outcome: Fully correct
61
59
  ```
62
60
 
63
61
  Ranges must be integers 0-10, non-overlapping, covering all values 0-10.
@@ -1,11 +1,13 @@
1
1
  import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
2
2
  import {
3
3
  require_token_util
4
- } from "./chunk-LJVS3JAK.js";
4
+ } from "./chunk-BL4PVUAT.js";
5
5
  import {
6
- __commonJS,
7
6
  require_token_error
8
- } from "./chunk-BKMQNEUD.js";
7
+ } from "./chunk-3L2L5GIL.js";
8
+ import {
9
+ __commonJS
10
+ } from "./chunk-5H446C7X.js";
9
11
 
10
12
  // ../../node_modules/.bun/@vercel+oidc@3.0.5/node_modules/@vercel/oidc/dist/token.js
11
13
  var require_token = __commonJS({
@@ -61,4 +63,4 @@ var require_token = __commonJS({
61
63
  }
62
64
  });
63
65
  export default require_token();
64
- //# sourceMappingURL=token-D3IYDJQZ.js.map
66
+ //# sourceMappingURL=token-POXF46NU.js.map
@@ -1 +1 @@
1
- {"version":3,"sources":["../../../node_modules/.bun/@vercel+oidc@3.0.5/node_modules/@vercel/oidc/dist/token.js"],"sourcesContent":["\"use strict\";\nvar __defProp = Object.defineProperty;\nvar __getOwnPropDesc = Object.getOwnPropertyDescriptor;\nvar __getOwnPropNames = Object.getOwnPropertyNames;\nvar __hasOwnProp = Object.prototype.hasOwnProperty;\nvar __export = (target, all) => {\n for (var name in all)\n __defProp(target, name, { get: all[name], enumerable: true });\n};\nvar __copyProps = (to, from, except, desc) => {\n if (from && typeof from === \"object\" || typeof from === \"function\") {\n for (let key of __getOwnPropNames(from))\n if (!__hasOwnProp.call(to, key) && key !== except)\n __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });\n }\n return to;\n};\nvar __toCommonJS = (mod) => __copyProps(__defProp({}, \"__esModule\", { value: true }), mod);\nvar token_exports = {};\n__export(token_exports, {\n refreshToken: () => refreshToken\n});\nmodule.exports = __toCommonJS(token_exports);\nvar import_token_error = require(\"./token-error\");\nvar import_token_util = require(\"./token-util\");\nasync function refreshToken() {\n const { projectId, teamId } = (0, import_token_util.findProjectInfo)();\n let maybeToken = (0, import_token_util.loadToken)(projectId);\n if (!maybeToken || (0, import_token_util.isExpired)((0, import_token_util.getTokenPayload)(maybeToken.token))) {\n const authToken = (0, import_token_util.getVercelCliToken)();\n if (!authToken) {\n throw new import_token_error.VercelOidcTokenError(\n \"Failed to refresh OIDC token: login to vercel cli\"\n );\n }\n if (!projectId) {\n throw new import_token_error.VercelOidcTokenError(\n \"Failed to refresh OIDC token: project id not found\"\n );\n }\n maybeToken = await (0, import_token_util.getVercelOidcToken)(authToken, projectId, teamId);\n if (!maybeToken) {\n throw new import_token_error.VercelOidcTokenError(\"Failed to refresh OIDC token\");\n }\n (0, import_token_util.saveToken)(maybeToken, projectId);\n }\n process.env.VERCEL_OIDC_TOKEN = maybeToken.token;\n return;\n}\n// Annotate the CommonJS export names for ESM import in node:\n0 && (module.exports = {\n refreshToken\n});\n"],"mappings":";;;;;;;;;;AAAA;AAAA;AACA,QAAI,YAAY,OAAO;AACvB,QAAI,mBAAmB,OAAO;AAC9B,QAAI,oBAAoB,OAAO;AAC/B,QAAI,eAAe,OAAO,UAAU;AACpC,QAAI,WAAW,CAAC,QAAQ,QAAQ;AAC9B,eAAS,QAAQ;AACf,kBAAU,QAAQ,MAAM,EAAE,KAAK,IAAI,IAAI,GAAG,YAAY,KAAK,CAAC;AAAA,IAChE;AACA,QAAI,cAAc,CAAC,IAAI,MAAM,QAAQ,SAAS;AAC5C,UAAI,QAAQ,OAAO,SAAS,YAAY,OAAO,SAAS,YAAY;AAClE,iBAAS,OAAO,kBAAkB,IAAI;AACpC,cAAI,CAAC,aAAa,KAAK,IAAI,GAAG,KAAK,QAAQ;AACzC,sBAAU,IAAI,KAAK,EAAE,KAAK,MAAM,KAAK,GAAG,GAAG,YAAY,EAAE,OAAO,iBAAiB,MAAM,GAAG,MAAM,KAAK,WAAW,CAAC;AAAA,MACvH;AACA,aAAO;AAAA,IACT;AACA,QAAI,eAAe,CAAC,QAAQ,YAAY,UAAU,CAAC,GAAG,cAAc,EAAE,OAAO,KAAK,CAAC,GAAG,GAAG;AACzF,QAAI,gBAAgB,CAAC;AACrB,aAAS,eAAe;AAAA,MACtB,cAAc,MAAM;AAAA,IACtB,CAAC;AACD,WAAO,UAAU,aAAa,aAAa;AAC3C,QAAI,qBAAqB;AACzB,QAAI,oBAAoB;AACxB,mBAAe,eAAe;AAC5B,YAAM,EAAE,WAAW,OAAO,KAAK,GAAG,kBAAkB,iBAAiB;AACrE,UAAI,cAAc,GAAG,kBAAkB,WAAW,SAAS;AAC3D,UAAI,CAAC,eAAe,GAAG,kBAAkB,YAAY,GAAG,kBAAkB,iBAAiB,WAAW,KAAK,CAAC,GAAG;AAC7G,cAAM,aAAa,GAAG,kBAAkB,mBAAmB;AAC3D,YAAI,CAAC,WAAW;AACd,gBAAM,IAAI,mBAAmB;AAAA,YAC3B;AAAA,UACF;AAAA,QACF;AACA,YAAI,CAAC,WAAW;AACd,gBAAM,IAAI,mBAAmB;AAAA,YAC3B;AAAA,UACF;AAAA,QACF;AACA,qBAAa,OAAO,GAAG,kBAAkB,oBAAoB,WAAW,WAAW,MAAM;AACzF,YAAI,CAAC,YAAY;AACf,gBAAM,IAAI,mBAAmB,qBAAqB,8BAA8B;AAAA,QAClF;AACA,SAAC,GAAG,kBAAkB,WAAW,YAAY,SAAS;AAAA,MACxD;AACA,cAAQ,IAAI,oBAAoB,WAAW;AAC3C;AAAA,IACF;AAAA;AAAA;","names":[]}
1
+ {"version":3,"sources":["../../../node_modules/.bun/@vercel+oidc@3.0.5/node_modules/@vercel/oidc/dist/token.js"],"sourcesContent":["\"use strict\";\nvar __defProp = Object.defineProperty;\nvar __getOwnPropDesc = Object.getOwnPropertyDescriptor;\nvar __getOwnPropNames = Object.getOwnPropertyNames;\nvar __hasOwnProp = Object.prototype.hasOwnProperty;\nvar __export = (target, all) => {\n for (var name in all)\n __defProp(target, name, { get: all[name], enumerable: true });\n};\nvar __copyProps = (to, from, except, desc) => {\n if (from && typeof from === \"object\" || typeof from === \"function\") {\n for (let key of __getOwnPropNames(from))\n if (!__hasOwnProp.call(to, key) && key !== except)\n __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });\n }\n return to;\n};\nvar __toCommonJS = (mod) => __copyProps(__defProp({}, \"__esModule\", { value: true }), mod);\nvar token_exports = {};\n__export(token_exports, {\n refreshToken: () => refreshToken\n});\nmodule.exports = __toCommonJS(token_exports);\nvar import_token_error = require(\"./token-error\");\nvar import_token_util = require(\"./token-util\");\nasync function refreshToken() {\n const { projectId, teamId } = (0, import_token_util.findProjectInfo)();\n let maybeToken = (0, import_token_util.loadToken)(projectId);\n if (!maybeToken || (0, import_token_util.isExpired)((0, import_token_util.getTokenPayload)(maybeToken.token))) {\n const authToken = (0, import_token_util.getVercelCliToken)();\n if (!authToken) {\n throw new import_token_error.VercelOidcTokenError(\n \"Failed to refresh OIDC token: login to vercel cli\"\n );\n }\n if (!projectId) {\n throw new import_token_error.VercelOidcTokenError(\n \"Failed to refresh OIDC token: project id not found\"\n );\n }\n maybeToken = await (0, import_token_util.getVercelOidcToken)(authToken, projectId, teamId);\n if (!maybeToken) {\n throw new import_token_error.VercelOidcTokenError(\"Failed to refresh OIDC token\");\n }\n (0, import_token_util.saveToken)(maybeToken, projectId);\n }\n process.env.VERCEL_OIDC_TOKEN = maybeToken.token;\n return;\n}\n// Annotate the CommonJS export names for ESM import in node:\n0 && (module.exports = {\n refreshToken\n});\n"],"mappings":";;;;;;;;;;;;AAAA;AAAA;AACA,QAAI,YAAY,OAAO;AACvB,QAAI,mBAAmB,OAAO;AAC9B,QAAI,oBAAoB,OAAO;AAC/B,QAAI,eAAe,OAAO,UAAU;AACpC,QAAI,WAAW,CAAC,QAAQ,QAAQ;AAC9B,eAAS,QAAQ;AACf,kBAAU,QAAQ,MAAM,EAAE,KAAK,IAAI,IAAI,GAAG,YAAY,KAAK,CAAC;AAAA,IAChE;AACA,QAAI,cAAc,CAAC,IAAI,MAAM,QAAQ,SAAS;AAC5C,UAAI,QAAQ,OAAO,SAAS,YAAY,OAAO,SAAS,YAAY;AAClE,iBAAS,OAAO,kBAAkB,IAAI;AACpC,cAAI,CAAC,aAAa,KAAK,IAAI,GAAG,KAAK,QAAQ;AACzC,sBAAU,IAAI,KAAK,EAAE,KAAK,MAAM,KAAK,GAAG,GAAG,YAAY,EAAE,OAAO,iBAAiB,MAAM,GAAG,MAAM,KAAK,WAAW,CAAC;AAAA,MACvH;AACA,aAAO;AAAA,IACT;AACA,QAAI,eAAe,CAAC,QAAQ,YAAY,UAAU,CAAC,GAAG,cAAc,EAAE,OAAO,KAAK,CAAC,GAAG,GAAG;AACzF,QAAI,gBAAgB,CAAC;AACrB,aAAS,eAAe;AAAA,MACtB,cAAc,MAAM;AAAA,IACtB,CAAC;AACD,WAAO,UAAU,aAAa,aAAa;AAC3C,QAAI,qBAAqB;AACzB,QAAI,oBAAoB;AACxB,mBAAe,eAAe;AAC5B,YAAM,EAAE,WAAW,OAAO,KAAK,GAAG,kBAAkB,iBAAiB;AACrE,UAAI,cAAc,GAAG,kBAAkB,WAAW,SAAS;AAC3D,UAAI,CAAC,eAAe,GAAG,kBAAkB,YAAY,GAAG,kBAAkB,iBAAiB,WAAW,KAAK,CAAC,GAAG;AAC7G,cAAM,aAAa,GAAG,kBAAkB,mBAAmB;AAC3D,YAAI,CAAC,WAAW;AACd,gBAAM,IAAI,mBAAmB;AAAA,YAC3B;AAAA,UACF;AAAA,QACF;AACA,YAAI,CAAC,WAAW;AACd,gBAAM,IAAI,mBAAmB;AAAA,YAC3B;AAAA,UACF;AAAA,QACF;AACA,qBAAa,OAAO,GAAG,kBAAkB,oBAAoB,WAAW,WAAW,MAAM;AACzF,YAAI,CAAC,YAAY;AACf,gBAAM,IAAI,mBAAmB,qBAAqB,8BAA8B;AAAA,QAClF;AACA,SAAC,GAAG,kBAAkB,WAAW,YAAY,SAAS;AAAA,MACxD;AACA,cAAQ,IAAI,oBAAoB,WAAW;AAC3C;AAAA,IACF;AAAA;AAAA;","names":[]}
@@ -1,7 +1,8 @@
1
1
  import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
2
2
  import {
3
3
  require_token_util
4
- } from "./chunk-LJVS3JAK.js";
5
- import "./chunk-BKMQNEUD.js";
4
+ } from "./chunk-BL4PVUAT.js";
5
+ import "./chunk-3L2L5GIL.js";
6
+ import "./chunk-5H446C7X.js";
6
7
  export default require_token_util();
7
- //# sourceMappingURL=token-util-FWFPR2BV.js.map
8
+ //# sourceMappingURL=token-util-6GWYZWGE.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "agentv",
3
- "version": "2.5.8",
3
+ "version": "2.7.1-next.1",
4
4
  "description": "CLI entry point for AgentV",
5
5
  "type": "module",
6
6
  "repository": {
@@ -16,7 +16,7 @@
16
16
  },
17
17
  "files": ["dist", "README.md"],
18
18
  "scripts": {
19
- "dev": "bun --watch src/index.ts",
19
+ "dev": "bun src/cli.ts",
20
20
  "build": "tsup && bun run copy-readme",
21
21
  "copy-readme": "bun -e \"import { cpSync } from 'fs'; cpSync('../../README.md', 'README.md')\"",
22
22
  "prepublishOnly": "bun run copy-readme",
@@ -28,9 +28,13 @@
28
28
  "test:watch": "bun test --watch"
29
29
  },
30
30
  "dependencies": {
31
- "@agentv/core": "2.5.4",
32
- "@mariozechner/pi-agent": "^0.9.0",
33
- "@mariozechner/pi-ai": "^0.37.2",
31
+ "@agentv/core": "2.7.0",
32
+ "@anthropic-ai/claude-agent-sdk": "^0.2.49",
33
+ "@github/copilot-sdk": "^0.1.25",
34
+ "@inquirer/prompts": "^8.2.1",
35
+ "@mariozechner/pi-agent-core": "^0.50.9",
36
+ "@mariozechner/pi-ai": "^0.50.9",
37
+ "@openai/codex-sdk": "^0.104.0",
34
38
  "cmd-ts": "^0.14.3",
35
39
  "dotenv": "^16.4.5",
36
40
  "fast-glob": "^3.3.3",