@claude-flow/cli 3.11.0 → 3.12.1

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 (192) hide show
  1. package/dist/src/commands/doctor.d.ts.map +1 -1
  2. package/dist/src/commands/doctor.js +175 -1
  3. package/dist/src/commands/doctor.js.map +1 -1
  4. package/dist/src/commands/eject.d.ts +33 -0
  5. package/dist/src/commands/eject.d.ts.map +1 -0
  6. package/dist/src/commands/eject.js +195 -0
  7. package/dist/src/commands/eject.js.map +1 -0
  8. package/dist/src/commands/hooks.d.ts.map +1 -1
  9. package/dist/src/commands/hooks.js +2 -1
  10. package/dist/src/commands/hooks.js.map +1 -1
  11. package/dist/src/commands/index.d.ts.map +1 -1
  12. package/dist/src/commands/index.js +4 -0
  13. package/dist/src/commands/index.js.map +1 -1
  14. package/dist/src/commands/init.d.ts.map +1 -1
  15. package/dist/src/commands/init.js +3 -0
  16. package/dist/src/commands/init.js.map +1 -1
  17. package/dist/src/commands/metaharness.d.ts +39 -0
  18. package/dist/src/commands/metaharness.d.ts.map +1 -0
  19. package/dist/src/commands/metaharness.js +194 -0
  20. package/dist/src/commands/metaharness.js.map +1 -0
  21. package/dist/src/mcp-client.d.ts.map +1 -1
  22. package/dist/src/mcp-client.js +4 -0
  23. package/dist/src/mcp-client.js.map +1 -1
  24. package/dist/src/mcp-tools/index.d.ts +1 -0
  25. package/dist/src/mcp-tools/index.d.ts.map +1 -1
  26. package/dist/src/mcp-tools/index.js +2 -0
  27. package/dist/src/mcp-tools/index.js.map +1 -1
  28. package/dist/src/mcp-tools/metaharness-tools.d.ts +31 -0
  29. package/dist/src/mcp-tools/metaharness-tools.d.ts.map +1 -0
  30. package/dist/src/mcp-tools/metaharness-tools.js +372 -0
  31. package/dist/src/mcp-tools/metaharness-tools.js.map +1 -0
  32. package/dist/src/plugins/store/discovery.d.ts.map +1 -1
  33. package/dist/src/plugins/store/discovery.js +3 -0
  34. package/dist/src/plugins/store/discovery.js.map +1 -1
  35. package/dist/src/plugins/store/types.d.ts +6 -1
  36. package/dist/src/plugins/store/types.d.ts.map +1 -1
  37. package/dist/src/ruvector/model-router.d.ts.map +1 -1
  38. package/dist/src/ruvector/model-router.js +37 -0
  39. package/dist/src/ruvector/model-router.js.map +1 -1
  40. package/dist/src/ruvector/router-parallel-recorder.d.ts +127 -0
  41. package/dist/src/ruvector/router-parallel-recorder.d.ts.map +1 -0
  42. package/dist/src/ruvector/router-parallel-recorder.js +183 -0
  43. package/dist/src/ruvector/router-parallel-recorder.js.map +1 -0
  44. package/dist/tsconfig.tsbuildinfo +1 -1
  45. package/package.json +13 -10
  46. package/plugins/ruflo-metaharness/.claude-flow/data/pending-insights.jsonl +1 -0
  47. package/plugins/ruflo-metaharness/.claude-flow/neural/stats.json +6 -0
  48. package/plugins/ruflo-metaharness/.claude-plugin/plugin.json +32 -0
  49. package/plugins/ruflo-metaharness/README.md +64 -0
  50. package/plugins/ruflo-metaharness/agents/metaharness-architect.md +58 -0
  51. package/plugins/ruflo-metaharness/commands/ruflo-metaharness.md +46 -0
  52. package/plugins/ruflo-metaharness/scripts/_harness.mjs +261 -0
  53. package/plugins/ruflo-metaharness/scripts/_similarity.mjs +161 -0
  54. package/plugins/ruflo-metaharness/scripts/_spike-similarity.mjs +223 -0
  55. package/plugins/ruflo-metaharness/scripts/audit-list.mjs +158 -0
  56. package/plugins/ruflo-metaharness/scripts/audit-trend.mjs +272 -0
  57. package/plugins/ruflo-metaharness/scripts/bench-parse-mcp-scan.mjs +146 -0
  58. package/plugins/ruflo-metaharness/scripts/bench-recordpair-overhead.mjs +186 -0
  59. package/plugins/ruflo-metaharness/scripts/bench-similarity.mjs +177 -0
  60. package/plugins/ruflo-metaharness/scripts/drift-from-history.mjs +363 -0
  61. package/plugins/ruflo-metaharness/scripts/genome.mjs +80 -0
  62. package/plugins/ruflo-metaharness/scripts/mcp-scan.mjs +111 -0
  63. package/plugins/ruflo-metaharness/scripts/mint.mjs +119 -0
  64. package/plugins/ruflo-metaharness/scripts/oia-audit.mjs +228 -0
  65. package/plugins/ruflo-metaharness/scripts/router-parallel-analyze.mjs +250 -0
  66. package/plugins/ruflo-metaharness/scripts/score.mjs +92 -0
  67. package/plugins/ruflo-metaharness/scripts/similarity.mjs +158 -0
  68. package/plugins/ruflo-metaharness/scripts/smoke.sh +2268 -0
  69. package/plugins/ruflo-metaharness/scripts/test-graceful-degradation.mjs +141 -0
  70. package/plugins/ruflo-metaharness/scripts/test-mcp-tools.mjs +437 -0
  71. package/plugins/ruflo-metaharness/scripts/test-parallel-pipeline.mjs +204 -0
  72. package/plugins/ruflo-metaharness/scripts/test-pipeline-roundtrip.mjs +586 -0
  73. package/plugins/ruflo-metaharness/scripts/test-similarity.mjs +334 -0
  74. package/plugins/ruflo-metaharness/scripts/test-with-openrouter.mjs +229 -0
  75. package/plugins/ruflo-metaharness/scripts/threat-model.mjs +59 -0
  76. package/plugins/ruflo-metaharness/skills/harness-drift-from-history/SKILL.md +65 -0
  77. package/plugins/ruflo-metaharness/skills/harness-genome/SKILL.md +54 -0
  78. package/plugins/ruflo-metaharness/skills/harness-mcp-scan/SKILL.md +47 -0
  79. package/plugins/ruflo-metaharness/skills/harness-mint/SKILL.md +72 -0
  80. package/plugins/ruflo-metaharness/skills/harness-oia-audit/SKILL.md +79 -0
  81. package/plugins/ruflo-metaharness/skills/harness-score/SKILL.md +66 -0
  82. package/plugins/ruflo-metaharness/skills/harness-similarity/SKILL.md +67 -0
  83. package/plugins/ruflo-metaharness/skills/harness-threat-model/SKILL.md +39 -0
  84. package/dist/src/benchmarks/capability-tasks.d.ts +0 -126
  85. package/dist/src/benchmarks/capability-tasks.d.ts.map +0 -1
  86. package/dist/src/benchmarks/capability-tasks.js +0 -147
  87. package/dist/src/benchmarks/capability-tasks.js.map +0 -1
  88. package/dist/src/benchmarks/gaia-agent-contrastive.d.ts +0 -53
  89. package/dist/src/benchmarks/gaia-agent-contrastive.d.ts.map +0 -1
  90. package/dist/src/benchmarks/gaia-agent-contrastive.js +0 -177
  91. package/dist/src/benchmarks/gaia-agent-contrastive.js.map +0 -1
  92. package/dist/src/benchmarks/gaia-agent-gemini.d.ts +0 -51
  93. package/dist/src/benchmarks/gaia-agent-gemini.d.ts.map +0 -1
  94. package/dist/src/benchmarks/gaia-agent-gemini.js +0 -352
  95. package/dist/src/benchmarks/gaia-agent-gemini.js.map +0 -1
  96. package/dist/src/benchmarks/gaia-claude-p.d.ts +0 -105
  97. package/dist/src/benchmarks/gaia-claude-p.d.ts.map +0 -1
  98. package/dist/src/benchmarks/gaia-claude-p.js +0 -269
  99. package/dist/src/benchmarks/gaia-claude-p.js.map +0 -1
  100. package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts +0 -20
  101. package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts.map +0 -1
  102. package/dist/src/benchmarks/gaia-claude-p.smoke.js +0 -168
  103. package/dist/src/benchmarks/gaia-claude-p.smoke.js.map +0 -1
  104. package/dist/src/benchmarks/gaia-codeagent-runner.py +0 -556
  105. package/dist/src/benchmarks/gaia-codeagent.d.ts +0 -129
  106. package/dist/src/benchmarks/gaia-codeagent.d.ts.map +0 -1
  107. package/dist/src/benchmarks/gaia-codeagent.js +0 -590
  108. package/dist/src/benchmarks/gaia-codeagent.js.map +0 -1
  109. package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts +0 -33
  110. package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts.map +0 -1
  111. package/dist/src/benchmarks/gaia-codeagent.smoke.js +0 -241
  112. package/dist/src/benchmarks/gaia-codeagent.smoke.js.map +0 -1
  113. package/dist/src/benchmarks/gaia-dag-pilot.d.ts +0 -22
  114. package/dist/src/benchmarks/gaia-dag-pilot.d.ts.map +0 -1
  115. package/dist/src/benchmarks/gaia-dag-pilot.js +0 -158
  116. package/dist/src/benchmarks/gaia-dag-pilot.js.map +0 -1
  117. package/dist/src/benchmarks/gaia-dag.d.ts +0 -110
  118. package/dist/src/benchmarks/gaia-dag.d.ts.map +0 -1
  119. package/dist/src/benchmarks/gaia-dag.js +0 -658
  120. package/dist/src/benchmarks/gaia-dag.js.map +0 -1
  121. package/dist/src/benchmarks/gaia-dag.smoke.d.ts +0 -21
  122. package/dist/src/benchmarks/gaia-dag.smoke.d.ts.map +0 -1
  123. package/dist/src/benchmarks/gaia-dag.smoke.js +0 -290
  124. package/dist/src/benchmarks/gaia-dag.smoke.js.map +0 -1
  125. package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts +0 -17
  126. package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts.map +0 -1
  127. package/dist/src/benchmarks/gaia-ensemble-pilot.js +0 -106
  128. package/dist/src/benchmarks/gaia-ensemble-pilot.js.map +0 -1
  129. package/dist/src/benchmarks/gaia-ensemble.d.ts +0 -103
  130. package/dist/src/benchmarks/gaia-ensemble.d.ts.map +0 -1
  131. package/dist/src/benchmarks/gaia-ensemble.js +0 -462
  132. package/dist/src/benchmarks/gaia-ensemble.js.map +0 -1
  133. package/dist/src/benchmarks/gaia-gemini-pilot.d.ts +0 -17
  134. package/dist/src/benchmarks/gaia-gemini-pilot.d.ts.map +0 -1
  135. package/dist/src/benchmarks/gaia-gemini-pilot.js +0 -168
  136. package/dist/src/benchmarks/gaia-gemini-pilot.js.map +0 -1
  137. package/dist/src/benchmarks/gaia-mode-router.d.ts +0 -67
  138. package/dist/src/benchmarks/gaia-mode-router.d.ts.map +0 -1
  139. package/dist/src/benchmarks/gaia-mode-router.js +0 -126
  140. package/dist/src/benchmarks/gaia-mode-router.js.map +0 -1
  141. package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts +0 -15
  142. package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts.map +0 -1
  143. package/dist/src/benchmarks/gaia-mode-router.smoke.js +0 -131
  144. package/dist/src/benchmarks/gaia-mode-router.smoke.js.map +0 -1
  145. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts +0 -24
  146. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts.map +0 -1
  147. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js +0 -276
  148. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js.map +0 -1
  149. package/dist/src/benchmarks/gaia-tools/image_describe.d.ts +0 -45
  150. package/dist/src/benchmarks/gaia-tools/image_describe.d.ts.map +0 -1
  151. package/dist/src/benchmarks/gaia-tools/image_describe.js +0 -222
  152. package/dist/src/benchmarks/gaia-tools/image_describe.js.map +0 -1
  153. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts +0 -24
  154. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts.map +0 -1
  155. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js +0 -169
  156. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js.map +0 -1
  157. package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts +0 -27
  158. package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts.map +0 -1
  159. package/dist/src/benchmarks/gaia-tools/pdf_read.js +0 -226
  160. package/dist/src/benchmarks/gaia-tools/pdf_read.js.map +0 -1
  161. package/dist/src/benchmarks/gaia-tools/python_exec.d.ts +0 -31
  162. package/dist/src/benchmarks/gaia-tools/python_exec.d.ts.map +0 -1
  163. package/dist/src/benchmarks/gaia-tools/python_exec.js +0 -94
  164. package/dist/src/benchmarks/gaia-tools/python_exec.js.map +0 -1
  165. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts +0 -13
  166. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts.map +0 -1
  167. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js +0 -98
  168. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js.map +0 -1
  169. package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts +0 -52
  170. package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts.map +0 -1
  171. package/dist/src/benchmarks/gaia-tools/visit_webpage.js +0 -304
  172. package/dist/src/benchmarks/gaia-tools/visit_webpage.js.map +0 -1
  173. package/dist/src/benchmarks/gaia-tools/web_browse.d.ts +0 -61
  174. package/dist/src/benchmarks/gaia-tools/web_browse.d.ts.map +0 -1
  175. package/dist/src/benchmarks/gaia-tools/web_browse.js +0 -217
  176. package/dist/src/benchmarks/gaia-tools/web_browse.js.map +0 -1
  177. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts +0 -26
  178. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts.map +0 -1
  179. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js +0 -152
  180. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js.map +0 -1
  181. package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts +0 -24
  182. package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts.map +0 -1
  183. package/dist/src/benchmarks/gaia-tools/web_search.smoke.js +0 -311
  184. package/dist/src/benchmarks/gaia-tools/web_search.smoke.js.map +0 -1
  185. package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts +0 -25
  186. package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts.map +0 -1
  187. package/dist/src/benchmarks/gaia-visit-webpage.smoke.js +0 -304
  188. package/dist/src/benchmarks/gaia-visit-webpage.smoke.js.map +0 -1
  189. package/dist/src/commands/performance-capability.d.ts +0 -30
  190. package/dist/src/commands/performance-capability.d.ts.map +0 -1
  191. package/dist/src/commands/performance-capability.js +0 -349
  192. package/dist/src/commands/performance-capability.js.map +0 -1
@@ -1,98 +0,0 @@
1
- /**
2
- * Smoke tests for python_exec — ADR-133-PR4
3
- *
4
- * Run with:
5
- * cd v3/@claude-flow/cli
6
- * npx tsx src/benchmarks/gaia-tools/python_exec.smoke.ts
7
- *
8
- * Expects python3 to be installed and on PATH.
9
- *
10
- * Refs: ADR-133, #2156
11
- */
12
- import { createPythonExecTool } from './python_exec.js';
13
- async function runTest(name, fn) {
14
- try {
15
- await fn();
16
- return { name, passed: true, message: 'OK' };
17
- }
18
- catch (e) {
19
- return { name, passed: false, message: String(e) };
20
- }
21
- }
22
- function assert(condition, msg) {
23
- if (!condition)
24
- throw new Error(`Assertion failed: ${msg}`);
25
- }
26
- // ---------------------------------------------------------------------------
27
- // Tests
28
- // ---------------------------------------------------------------------------
29
- const tool = createPythonExecTool();
30
- const tests = [
31
- runTest('Test 1: basic arithmetic — print(2+2)', async () => {
32
- const out = await tool.execute({ code: 'print(2+2)' });
33
- assert(out.includes('4'), `Expected stdout to contain "4", got:\n${out}`);
34
- assert(out.includes('exit_code: 0'), `Expected exit_code 0, got:\n${out}`);
35
- }),
36
- runTest('Test 2: large computation — sum(range(1000000))', async () => {
37
- const out = await tool.execute({ code: 'print(sum(range(1000000)))' });
38
- assert(out.includes('499999500000'), `Expected "499999500000", got:\n${out}`);
39
- assert(out.includes('exit_code: 0'), `Expected exit_code 0, got:\n${out}`);
40
- }),
41
- runTest('Test 3: stdlib import — math.sqrt(144)', async () => {
42
- const out = await tool.execute({
43
- code: 'import math; print(math.sqrt(144))',
44
- });
45
- assert(out.includes('12.0'), `Expected "12.0", got:\n${out}`);
46
- assert(out.includes('exit_code: 0'), `Expected exit_code 0, got:\n${out}`);
47
- }),
48
- runTest('Test 4: timeout — infinite loop with 2s limit', async () => {
49
- const start = Date.now();
50
- const out = await tool.execute({
51
- code: 'while True: pass',
52
- timeout_seconds: 2,
53
- });
54
- const elapsed = Date.now() - start;
55
- assert(elapsed < 6_000, `Timeout should have fired within ~2s, took ${elapsed}ms`);
56
- const hasTimedOut = out.includes('timed out') ||
57
- out.includes('exit_code: 124') ||
58
- // SIGKILL on some platforms gives exit code -9 or 137
59
- out.includes('exit_code: -9') ||
60
- out.includes('exit_code: 137');
61
- assert(hasTimedOut, `Expected timeout signal, got:\n${out}`);
62
- }),
63
- runTest('Test 5: syntax error — exit_code != 0, stderr contains SyntaxError', async () => {
64
- const out = await tool.execute({ code: 'x = (' });
65
- assert(out.includes('SyntaxError'), `Expected SyntaxError in output, got:\n${out}`);
66
- const hasNonZeroExit = !out.includes('exit_code: 0') || out.includes('stderr:');
67
- assert(hasNonZeroExit, `Expected non-zero exit code or stderr for syntax error, got:\n${out}`);
68
- }),
69
- ];
70
- // ---------------------------------------------------------------------------
71
- // Runner
72
- // ---------------------------------------------------------------------------
73
- async function main() {
74
- console.log('python_exec smoke tests — ADR-133-PR4\n');
75
- const results = await Promise.all(tests);
76
- let passed = 0;
77
- let failed = 0;
78
- for (const r of results) {
79
- const icon = r.passed ? 'PASS' : 'FAIL';
80
- console.log(`[${icon}] ${r.name}`);
81
- if (!r.passed) {
82
- console.log(` ${r.message}`);
83
- }
84
- if (r.passed)
85
- passed++;
86
- else
87
- failed++;
88
- }
89
- console.log(`\nResults: ${passed}/${results.length} passed`);
90
- if (failed > 0) {
91
- process.exit(1);
92
- }
93
- }
94
- main().catch((e) => {
95
- console.error('Fatal:', e);
96
- process.exit(1);
97
- });
98
- //# sourceMappingURL=python_exec.smoke.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"python_exec.smoke.js","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/python_exec.smoke.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,oBAAoB,EAAE,MAAM,kBAAkB,CAAC;AAYxD,KAAK,UAAU,OAAO,CACpB,IAAY,EACZ,EAAuB;IAEvB,IAAI,CAAC;QACH,MAAM,EAAE,EAAE,CAAC;QACX,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAC/C,CAAC;IAAC,OAAO,CAAU,EAAE,CAAC;QACpB,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;IACrD,CAAC;AACH,CAAC;AAED,SAAS,MAAM,CAAC,SAAkB,EAAE,GAAW;IAC7C,IAAI,CAAC,SAAS;QAAE,MAAM,IAAI,KAAK,CAAC,qBAAqB,GAAG,EAAE,CAAC,CAAC;AAC9D,CAAC;AAED,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,MAAM,IAAI,GAAG,oBAAoB,EAAE,CAAC;AAEpC,MAAM,KAAK,GAAG;IACZ,OAAO,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QAC1D,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;QACvD,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,yCAAyC,GAAG,EAAE,CAAC,CAAC;QAC1E,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,EAAE,+BAA+B,GAAG,EAAE,CAAC,CAAC;IAC7E,CAAC,CAAC;IAEF,OAAO,CAAC,iDAAiD,EAAE,KAAK,IAAI,EAAE;QACpE,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,4BAA4B,EAAE,CAAC,CAAC;QACvE,MAAM,CACJ,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,EAC5B,kCAAkC,GAAG,EAAE,CACxC,CAAC;QACF,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,EAAE,+BAA+B,GAAG,EAAE,CAAC,CAAC;IAC7E,CAAC,CAAC;IAEF,OAAO,CAAC,wCAAwC,EAAE,KAAK,IAAI,EAAE;QAC3D,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC;YAC7B,IAAI,EAAE,oCAAoC;SAC3C,CAAC,CAAC;QACH,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,0BAA0B,GAAG,EAAE,CAAC,CAAC;QAC9D,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,EAAE,+BAA+B,GAAG,EAAE,CAAC,CAAC;IAC7E,CAAC,CAAC;IAEF,OAAO,CAAC,+CAA+C,EAAE,KAAK,IAAI,EAAE;QAClE,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACzB,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC;YAC7B,IAAI,EAAE,kBAAkB;YACxB,eAAe,EAAE,CAAC;SACnB,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;QACnC,MAAM,CACJ,OAAO,GAAG,KAAK,EACf,8CAA8C,OAAO,IAAI,CAC1D,CAAC;QACF,MAAM,WAAW,GACf,GAAG,CAAC,QAAQ,CAAC,WAAW,CAAC;YACzB,GAAG,CAAC,QAAQ,CAAC,gBAAgB,CAAC;YAC9B,sDAAsD;YACtD,GAAG,CAAC,QAAQ,CAAC,eAAe,CAAC;YAC7B,GAAG,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC;QACjC,MAAM,CAAC,WAAW,EAAE,kCAAkC,GAAG,EAAE,CAAC,CAAC;IAC/D,CAAC,CAAC;IAEF,OAAO,CAAC,oEAAoE,EAAE,KAAK,IAAI,EAAE;QACvF,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;QAClD,MAAM,CACJ,GAAG,CAAC,QAAQ,CAAC,aAAa,CAAC,EAC3B,yCAAyC,GAAG,EAAE,CAC/C,CAAC;QACF,MAAM,cAAc,GAClB,CAAC,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAC3D,MAAM,CACJ,cAAc,EACd,iEAAiE,GAAG,EAAE,CACvE,CAAC;IACJ,CAAC,CAAC;CACH,CAAC;AAEF,8EAA8E;AAC9E,SAAS;AACT,8EAA8E;AAE9E,KAAK,UAAU,IAAI;IACjB,OAAO,CAAC,GAAG,CAAC,yCAAyC,CAAC,CAAC;IAEvD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;IACzC,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,MAAM,GAAG,CAAC,CAAC;IAEf,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QACxC,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QACnC,IAAI,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;YACd,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;QACrC,CAAC;QACD,IAAI,CAAC,CAAC,MAAM;YAAE,MAAM,EAAE,CAAC;;YAClB,MAAM,EAAE,CAAC;IAChB,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,cAAc,MAAM,IAAI,OAAO,CAAC,MAAM,SAAS,CAAC,CAAC;IAE7D,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAC3B,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
@@ -1,52 +0,0 @@
1
- /**
2
- * GAIA Tool: visit_webpage — iter-57 (upgraded from iter-54 base)
3
- *
4
- * Fetches the FULL readable text content of a webpage URL — not just snippets.
5
- * This closes the HAL parity gap identified in iter-51 surrender analysis
6
- * (~25-35% of L1 questions need full page reading that snippet-only tools miss).
7
- *
8
- * Upgrade over iter-54 base (iter-57 changes):
9
- * - max_chars input parameter (default 50k, was hardcoded 8k)
10
- * - Test hooks interface (visitWebpageTestHooks) for unit tests without live HTTP
11
- * - URL validation with helpful error messages
12
- * - Link extraction (up to 30 links) — enables follow-up navigation
13
- * - Better HTML extraction: content-block heuristics (main/article selection)
14
- * - AIDefence PII gate (optional dep, graceful skip if not installed)
15
- * - Structured VisitWebpageResult type + diagnostic stderr logging
16
- * - Python bs4 primary path preserved (already in benchmark env)
17
- *
18
- * Extraction pipeline:
19
- * 1. fetch() + Python bs4.get_text() (primary — better quality, handles encoding)
20
- * 2. Regex-based fallback (no external dep) if bs4 unavailable or output too short
21
- * 3. Content passed through AIDefence PII gate before returning to agent
22
- *
23
- * Refs: ADR-138, ADR-133, iter-54, iter-57, #2156
24
- */
25
- import { GaiaTool, ToolDefinition } from './types.js';
26
- export interface VisitWebpageResult {
27
- title: string;
28
- content: string;
29
- links: Array<{
30
- href: string;
31
- text: string;
32
- }>;
33
- sources_consulted: string[];
34
- chars_returned: number;
35
- truncated: boolean;
36
- }
37
- export interface VisitWebpageTestHooks {
38
- /** Override the HTTP fetch so tests never make live calls. */
39
- fetchHtml?: (url: string) => Promise<{
40
- html: string;
41
- finalUrl: string;
42
- }>;
43
- }
44
- export declare const visitWebpageTestHooks: VisitWebpageTestHooks;
45
- export declare function visitWebpage(url: string, maxChars?: number): Promise<VisitWebpageResult>;
46
- export declare class VisitWebpageTool implements GaiaTool {
47
- readonly name = "visit_webpage";
48
- readonly definition: ToolDefinition;
49
- execute(input: Record<string, unknown>): Promise<string>;
50
- }
51
- export declare function createVisitWebpageTool(): VisitWebpageTool;
52
- //# sourceMappingURL=visit_webpage.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"visit_webpage.d.ts","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/visit_webpage.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAGH,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAiBtD,MAAM,WAAW,kBAAkB;IACjC,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC7C,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAC5B,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAMD,MAAM,WAAW,qBAAqB;IACpC,8DAA8D;IAC9D,SAAS,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,OAAO,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAC1E;AAED,eAAO,MAAM,qBAAqB,EAAE,qBAA0B,CAAC;AAkL/D,wBAAsB,YAAY,CAChC,GAAG,EAAE,MAAM,EACX,QAAQ,GAAE,MAA0B,GACnC,OAAO,CAAC,kBAAkB,CAAC,CA+B7B;AA6BD,qBAAa,gBAAiB,YAAW,QAAQ;IAC/C,QAAQ,CAAC,IAAI,mBAAmB;IAEhC,QAAQ,CAAC,UAAU,EAAE,cAAc,CAwBjC;IAEI,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;CAyB/D;AAMD,wBAAgB,sBAAsB,IAAI,gBAAgB,CAEzD"}
@@ -1,304 +0,0 @@
1
- /**
2
- * GAIA Tool: visit_webpage — iter-57 (upgraded from iter-54 base)
3
- *
4
- * Fetches the FULL readable text content of a webpage URL — not just snippets.
5
- * This closes the HAL parity gap identified in iter-51 surrender analysis
6
- * (~25-35% of L1 questions need full page reading that snippet-only tools miss).
7
- *
8
- * Upgrade over iter-54 base (iter-57 changes):
9
- * - max_chars input parameter (default 50k, was hardcoded 8k)
10
- * - Test hooks interface (visitWebpageTestHooks) for unit tests without live HTTP
11
- * - URL validation with helpful error messages
12
- * - Link extraction (up to 30 links) — enables follow-up navigation
13
- * - Better HTML extraction: content-block heuristics (main/article selection)
14
- * - AIDefence PII gate (optional dep, graceful skip if not installed)
15
- * - Structured VisitWebpageResult type + diagnostic stderr logging
16
- * - Python bs4 primary path preserved (already in benchmark env)
17
- *
18
- * Extraction pipeline:
19
- * 1. fetch() + Python bs4.get_text() (primary — better quality, handles encoding)
20
- * 2. Regex-based fallback (no external dep) if bs4 unavailable or output too short
21
- * 3. Content passed through AIDefence PII gate before returning to agent
22
- *
23
- * Refs: ADR-138, ADR-133, iter-54, iter-57, #2156
24
- */
25
- import { execFileSync } from 'node:child_process';
26
- // ---------------------------------------------------------------------------
27
- // Constants
28
- // ---------------------------------------------------------------------------
29
- const DEFAULT_MAX_CHARS = 50_000;
30
- const MIN_CONTENT_CHARS = 50;
31
- const REQUEST_TIMEOUT_MS = 30_000;
32
- const MAX_LINKS = 30;
33
- const UA = 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36';
34
- export const visitWebpageTestHooks = {};
35
- // ---------------------------------------------------------------------------
36
- // URL validation
37
- // ---------------------------------------------------------------------------
38
- function validateUrl(url) {
39
- if (!url || typeof url !== 'string') {
40
- throw new Error('visit_webpage: `url` must be a non-empty string.');
41
- }
42
- if (!url.startsWith('http://') && !url.startsWith('https://')) {
43
- throw new Error(`visit_webpage: URL must start with http:// or https://. Got: "${url}"`);
44
- }
45
- try {
46
- new URL(url);
47
- }
48
- catch {
49
- throw new Error(`visit_webpage: invalid URL: "${url}"`);
50
- }
51
- }
52
- // ---------------------------------------------------------------------------
53
- // Page fetcher (uses Node built-in fetch — Node 18+)
54
- // ---------------------------------------------------------------------------
55
- async function fetchPage(url) {
56
- const controller = new AbortController();
57
- const timer = setTimeout(() => controller.abort(), REQUEST_TIMEOUT_MS);
58
- let res;
59
- try {
60
- res = await fetch(url, {
61
- headers: {
62
- 'User-Agent': UA,
63
- Accept: 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
64
- 'Accept-Language': 'en-US,en;q=0.9',
65
- },
66
- signal: controller.signal,
67
- redirect: 'follow',
68
- });
69
- }
70
- finally {
71
- clearTimeout(timer);
72
- }
73
- if (!res.ok) {
74
- throw new Error(`visit_webpage: HTTP ${res.status} for ${url}`);
75
- }
76
- const ct = res.headers.get('content-type') ?? '';
77
- if (ct.includes('application/pdf')) {
78
- throw new Error(`visit_webpage: URL returns a PDF. Use file_read with the downloaded path instead. URL: ${url}`);
79
- }
80
- const html = await res.text();
81
- const finalUrl = res.url || url;
82
- return { html, finalUrl };
83
- }
84
- // ---------------------------------------------------------------------------
85
- // HTML extraction helpers
86
- // ---------------------------------------------------------------------------
87
- /**
88
- * Extract title from raw HTML.
89
- */
90
- function extractTitle(html) {
91
- const match = /<title[^>]*>([\s\S]*?)<\/title>/i.exec(html);
92
- if (!match)
93
- return '';
94
- return match[1]
95
- .replace(/<[^>]+>/g, '')
96
- .replace(/&amp;/g, '&').replace(/&lt;/g, '<').replace(/&gt;/g, '>')
97
- .replace(/&quot;/g, '"').replace(/&nbsp;/g, ' ')
98
- .trim();
99
- }
100
- /**
101
- * Extract anchor links from HTML (up to MAX_LINKS).
102
- */
103
- function extractLinks(html, baseUrl) {
104
- const links = [];
105
- const anchorRe = /<a[^>]+href="([^"#][^"]*)"[^>]*>([\s\S]*?)<\/a>/gi;
106
- let match;
107
- let base;
108
- try {
109
- base = new URL(baseUrl);
110
- }
111
- catch {
112
- return links;
113
- }
114
- while ((match = anchorRe.exec(html)) !== null && links.length < MAX_LINKS) {
115
- const rawHref = match[1].trim();
116
- const rawText = match[2].replace(/<[^>]+>/g, ' ').replace(/\s+/g, ' ').trim().slice(0, 120);
117
- if (!rawText || rawHref.startsWith('javascript:') || rawHref.startsWith('mailto:'))
118
- continue;
119
- let href;
120
- try {
121
- href = new URL(rawHref, base.origin + base.pathname).href;
122
- }
123
- catch {
124
- continue;
125
- }
126
- links.push({ href, text: rawText });
127
- }
128
- return links;
129
- }
130
- /**
131
- * Primary extraction: Python bs4.get_text().
132
- * bs4 is available in the benchmark env (already installed for gaia-codeagent).
133
- * HTML is passed via stdin to avoid shell-escaping issues.
134
- */
135
- function extractTextViaPython(html) {
136
- const script = [
137
- 'import sys, re',
138
- 'from bs4 import BeautifulSoup',
139
- 'html = sys.stdin.read()',
140
- "soup = BeautifulSoup(html, 'html.parser')",
141
- "for tag in soup(['script', 'style', 'nav', 'footer', 'header', 'aside']):",
142
- ' tag.decompose()',
143
- "text = soup.get_text(separator='\\n', strip=True)",
144
- "text = re.sub(r'\\n{3,}', '\\n\\n', text)",
145
- 'print(text)',
146
- ].join('\n');
147
- try {
148
- return execFileSync('python3', ['-'], {
149
- input: script + '\n' + html,
150
- encoding: 'utf-8',
151
- timeout: 15_000,
152
- env: { ...process.env, PYTHONIOENCODING: 'utf-8' },
153
- maxBuffer: 10 * 1024 * 1024,
154
- }).trim();
155
- }
156
- catch {
157
- return '';
158
- }
159
- }
160
- /**
161
- * Fallback: regex-based HTML stripping (no external dep, less accurate).
162
- */
163
- function extractTextViaRegex(html) {
164
- return html
165
- .replace(/<script[\s\S]*?<\/script>/gi, '')
166
- .replace(/<style[\s\S]*?<\/style>/gi, '')
167
- .replace(/<nav[\s\S]*?<\/nav>/gi, '')
168
- .replace(/<footer[\s\S]*?<\/footer>/gi, '')
169
- .replace(/<header[\s\S]*?<\/header>/gi, '')
170
- .replace(/<\/?(p|div|section|article|main|li|h[1-6]|br)[^>]*>/gi, '\n')
171
- .replace(/<[^>]+>/g, ' ')
172
- .replace(/&amp;/g, '&').replace(/&lt;/g, '<').replace(/&gt;/g, '>')
173
- .replace(/&quot;/g, '"').replace(/&nbsp;/g, ' ').replace(/&#x27;/g, "'")
174
- .replace(/&#(\d+);/g, (_, n) => String.fromCharCode(Number(n)))
175
- .replace(/ +/g, ' ')
176
- .replace(/\n{4,}/g, '\n\n\n')
177
- .replace(/^\s+|\s+$/gm, '')
178
- .trim();
179
- }
180
- // ---------------------------------------------------------------------------
181
- // AIDefence PII gate (optional — graceful skip if not installed)
182
- // ---------------------------------------------------------------------------
183
- async function applyPiiGate(content) {
184
- try {
185
- const aidefence = await import('@claude-flow/aidefence').catch(() => null);
186
- if (!aidefence)
187
- return content;
188
- const { hasPii } = aidefence;
189
- if (typeof hasPii !== 'function')
190
- return content;
191
- const detected = await hasPii(content);
192
- if (detected) {
193
- process.stderr.write('[visit_webpage] AIDefence: PII detected in page content (not blocked)\n');
194
- }
195
- }
196
- catch {
197
- // AIDefence unavailable — continue without gate
198
- }
199
- return content;
200
- }
201
- // ---------------------------------------------------------------------------
202
- // Core visit logic (exported for unit testing via visitWebpageTestHooks)
203
- // ---------------------------------------------------------------------------
204
- export async function visitWebpage(url, maxChars = DEFAULT_MAX_CHARS) {
205
- const sourcesConsulted = [url];
206
- const { html, finalUrl } = visitWebpageTestHooks.fetchHtml
207
- ? await visitWebpageTestHooks.fetchHtml(url)
208
- : await fetchPage(url);
209
- if (finalUrl !== url)
210
- sourcesConsulted.push(finalUrl);
211
- const title = extractTitle(html);
212
- const links = extractLinks(html, finalUrl);
213
- // Primary: Python bs4; fallback: regex
214
- let content = extractTextViaPython(html);
215
- if (!content || content.length < MIN_CONTENT_CHARS) {
216
- content = extractTextViaRegex(html);
217
- }
218
- content = await applyPiiGate(content);
219
- const truncated = content.length > maxChars;
220
- if (truncated)
221
- content = content.slice(0, maxChars);
222
- return {
223
- title,
224
- content,
225
- links: links.slice(0, MAX_LINKS),
226
- sources_consulted: sourcesConsulted,
227
- chars_returned: content.length,
228
- truncated,
229
- };
230
- }
231
- // ---------------------------------------------------------------------------
232
- // Format output for Claude
233
- // ---------------------------------------------------------------------------
234
- function formatResult(result, url) {
235
- const lines = [];
236
- lines.push(`[visit_webpage: ${url}]`);
237
- if (result.title)
238
- lines.push(`Title: ${result.title}`);
239
- if (result.sources_consulted.length > 1) {
240
- lines.push(`Redirected to: ${result.sources_consulted[result.sources_consulted.length - 1]}`);
241
- }
242
- if (result.truncated)
243
- lines.push(`[Content truncated at ${result.chars_returned} chars]`);
244
- lines.push('');
245
- lines.push(result.content || '[No readable content extracted]');
246
- if (result.links.length > 0) {
247
- lines.push('');
248
- lines.push(`Links (${result.links.length}):`);
249
- for (const link of result.links)
250
- lines.push(` ${link.text} → ${link.href}`);
251
- }
252
- return lines.join('\n');
253
- }
254
- // ---------------------------------------------------------------------------
255
- // GaiaTool implementation
256
- // ---------------------------------------------------------------------------
257
- export class VisitWebpageTool {
258
- name = 'visit_webpage';
259
- definition = {
260
- name: 'visit_webpage',
261
- description: 'Fetch the FULL text content of a webpage URL. Unlike web_search (snippets only) ' +
262
- 'or grounded_query (synthesised summary), visit_webpage returns the complete readable ' +
263
- "text — equivalent to HAL's visit_webpage tool. Use this when you need to read a " +
264
- 'specific Wikipedia article, documentation page, government site, or any web page ' +
265
- 'in full to extract detailed facts, tables, or text not available from snippets. ' +
266
- 'Returns: page title, full text content (up to max_chars), and up to 30 extracted links. ' +
267
- `Default max_chars: ${DEFAULT_MAX_CHARS}. Maximum: 100,000.`,
268
- input_schema: {
269
- type: 'object',
270
- properties: {
271
- url: {
272
- type: 'string',
273
- description: 'The full URL (http:// or https://) of the webpage to fetch.',
274
- },
275
- max_chars: {
276
- type: 'number',
277
- description: `Maximum characters of content to return (default: ${DEFAULT_MAX_CHARS}, max: 100000).`,
278
- },
279
- },
280
- required: ['url'],
281
- },
282
- };
283
- async execute(input) {
284
- const url = String(input['url'] ?? '').trim();
285
- validateUrl(url);
286
- const maxChars = Math.min(Math.max(500, Number(input['max_chars'] ?? DEFAULT_MAX_CHARS)), 100_000);
287
- process.stderr.write(`[visit_webpage] url=${url} max_chars=${maxChars}\n`);
288
- const result = await visitWebpage(url, maxChars);
289
- process.stderr.write(`[visit_webpage] title=${JSON.stringify(result.title)} ` +
290
- `chars=${result.chars_returned} truncated=${result.truncated} ` +
291
- `links=${result.links.length}\n`);
292
- if (!result.content || result.content.length < MIN_CONTENT_CHARS) {
293
- return `[visit_webpage: page at ${url} returned no readable text content]`;
294
- }
295
- return formatResult(result, url);
296
- }
297
- }
298
- // ---------------------------------------------------------------------------
299
- // Convenience factory
300
- // ---------------------------------------------------------------------------
301
- export function createVisitWebpageTool() {
302
- return new VisitWebpageTool();
303
- }
304
- //# sourceMappingURL=visit_webpage.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"visit_webpage.js","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/visit_webpage.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAGlD,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,MAAM,iBAAiB,GAAG,MAAM,CAAC;AACjC,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAC7B,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAClC,MAAM,SAAS,GAAG,EAAE,CAAC;AACrB,MAAM,EAAE,GACN,uGAAuG,CAAC;AAwB1G,MAAM,CAAC,MAAM,qBAAqB,GAA0B,EAAE,CAAC;AAE/D,8EAA8E;AAC9E,iBAAiB;AACjB,8EAA8E;AAE9E,SAAS,WAAW,CAAC,GAAW;IAC9B,IAAI,CAAC,GAAG,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;QACpC,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;IACtE,CAAC;IACD,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;QAC9D,MAAM,IAAI,KAAK,CACb,iEAAiE,GAAG,GAAG,CACxE,CAAC;IACJ,CAAC;IACD,IAAI,CAAC;QACH,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC;IACf,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,gCAAgC,GAAG,GAAG,CAAC,CAAC;IAC1D,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,qDAAqD;AACrD,8EAA8E;AAE9E,KAAK,UAAU,SAAS,CAAC,GAAW;IAClC,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,kBAAkB,CAAC,CAAC;IAEvE,IAAI,GAAa,CAAC;IAClB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;YACrB,OAAO,EAAE;gBACP,YAAY,EAAE,EAAE;gBAChB,MAAM,EAAE,iEAAiE;gBACzE,iBAAiB,EAAE,gBAAgB;aACpC;YACD,MAAM,EAAE,UAAU,CAAC,MAAM;YACzB,QAAQ,EAAE,QAAQ;SACnB,CAAC,CAAC;IACL,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAK,CAAC,CAAC;IACtB,CAAC;IAED,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CAAC,uBAAuB,GAAG,CAAC,MAAM,QAAQ,GAAG,EAAE,CAAC,CAAC;IAClE,CAAC;IAED,MAAM,EAAE,GAAG,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC;IACjD,IAAI,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CACb,0FAA0F,GAAG,EAAE,CAChG,CAAC;IACJ,CAAC;IAED,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC;IAC9B,MAAM,QAAQ,GAAG,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC;IAChC,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;AAC5B,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAE9E;;GAEG;AACH,SAAS,YAAY,CAAC,IAAY;IAChC,MAAM,KAAK,GAAG,kCAAkC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5D,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,OAAO,KAAK,CAAC,CAAC,CAAC;SACZ,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,OAAO,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC;SAClE,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC;SAC/C,IAAI,EAAE,CAAC;AACZ,CAAC;AAED;;GAEG;AACH,SAAS,YAAY,CAAC,IAAY,EAAE,OAAe;IACjD,MAAM,KAAK,GAA0C,EAAE,CAAC;IACxD,MAAM,QAAQ,GAAG,mDAAmD,CAAC;IACrE,IAAI,KAA6B,CAAC;IAClC,IAAI,IAAS,CAAC;IACd,IAAI,CAAC;QAAC,IAAI,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC;IAAC,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,KAAK,CAAC;IAAC,CAAC;IAExD,OAAO,CAAC,KAAK,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,IAAI,IAAI,KAAK,CAAC,MAAM,GAAG,SAAS,EAAE,CAAC;QAC1E,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAChC,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;QAE5F,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,UAAU,CAAC,aAAa,CAAC,IAAI,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC;YAAE,SAAS;QAE7F,IAAI,IAAY,CAAC;QACjB,IAAI,CAAC;YAAC,IAAI,GAAG,IAAI,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC;YAAC,SAAS;QAAC,CAAC;QAEtF,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;IACtC,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;;GAIG;AACH,SAAS,oBAAoB,CAAC,IAAY;IACxC,MAAM,MAAM,GAAG;QACb,gBAAgB;QAChB,+BAA+B;QAC/B,yBAAyB;QACzB,2CAA2C;QAC3C,2EAA2E;QAC3E,qBAAqB;QACrB,mDAAmD;QACnD,2CAA2C;QAC3C,aAAa;KACd,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,IAAI,CAAC;QACH,OAAO,YAAY,CAAC,SAAS,EAAE,CAAC,GAAG,CAAC,EAAE;YACpC,KAAK,EAAE,MAAM,GAAG,IAAI,GAAG,IAAI;YAC3B,QAAQ,EAAE,OAAO;YACjB,OAAO,EAAE,MAAM;YACf,GAAG,EAAE,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,gBAAgB,EAAE,OAAO,EAAE;YAClD,SAAS,EAAE,EAAE,GAAG,IAAI,GAAG,IAAI;SAC5B,CAAC,CAAC,IAAI,EAAE,CAAC;IACZ,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED;;GAEG;AACH,SAAS,mBAAmB,CAAC,IAAY;IACvC,OAAO,IAAI;SACR,OAAO,CAAC,6BAA6B,EAAE,EAAE,CAAC;SAC1C,OAAO,CAAC,2BAA2B,EAAE,EAAE,CAAC;SACxC,OAAO,CAAC,uBAAuB,EAAE,EAAE,CAAC;SACpC,OAAO,CAAC,6BAA6B,EAAE,EAAE,CAAC;SAC1C,OAAO,CAAC,6BAA6B,EAAE,EAAE,CAAC;SAC1C,OAAO,CAAC,uDAAuD,EAAE,IAAI,CAAC;SACtE,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC;SACxB,OAAO,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,GAAG,CAAC;SAClE,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC;SACvE,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;SAC9D,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC;SACnB,OAAO,CAAC,SAAS,EAAE,QAAQ,CAAC;SAC5B,OAAO,CAAC,aAAa,EAAE,EAAE,CAAC;SAC1B,IAAI,EAAE,CAAC;AACZ,CAAC;AAED,8EAA8E;AAC9E,iEAAiE;AACjE,8EAA8E;AAE9E,KAAK,UAAU,YAAY,CAAC,OAAe;IACzC,IAAI,CAAC;QACH,MAAM,SAAS,GAAG,MAAM,MAAM,CAAC,wBAAwB,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;QAC3E,IAAI,CAAC,SAAS;YAAE,OAAO,OAAO,CAAC;QAC/B,MAAM,EAAE,MAAM,EAAE,GAAG,SAA4D,CAAC;QAChF,IAAI,OAAO,MAAM,KAAK,UAAU;YAAE,OAAO,OAAO,CAAC;QACjD,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,OAAO,CAAC,CAAC;QACvC,IAAI,QAAQ,EAAE,CAAC;YACb,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,yEAAyE,CAAC,CAAC;QAClG,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,gDAAgD;IAClD,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,8EAA8E;AAC9E,yEAAyE;AACzE,8EAA8E;AAE9E,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,GAAW,EACX,WAAmB,iBAAiB;IAEpC,MAAM,gBAAgB,GAAa,CAAC,GAAG,CAAC,CAAC;IAEzC,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,qBAAqB,CAAC,SAAS;QACxD,CAAC,CAAC,MAAM,qBAAqB,CAAC,SAAS,CAAC,GAAG,CAAC;QAC5C,CAAC,CAAC,MAAM,SAAS,CAAC,GAAG,CAAC,CAAC;IAEzB,IAAI,QAAQ,KAAK,GAAG;QAAE,gBAAgB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAEtD,MAAM,KAAK,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;IACjC,MAAM,KAAK,GAAG,YAAY,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IAE3C,uCAAuC;IACvC,IAAI,OAAO,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAC;IACzC,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,MAAM,GAAG,iBAAiB,EAAE,CAAC;QACnD,OAAO,GAAG,mBAAmB,CAAC,IAAI,CAAC,CAAC;IACtC,CAAC;IAED,OAAO,GAAG,MAAM,YAAY,CAAC,OAAO,CAAC,CAAC;IAEtC,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,GAAG,QAAQ,CAAC;IAC5C,IAAI,SAAS;QAAE,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;IAEpD,OAAO;QACL,KAAK;QACL,OAAO;QACP,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC;QAChC,iBAAiB,EAAE,gBAAgB;QACnC,cAAc,EAAE,OAAO,CAAC,MAAM;QAC9B,SAAS;KACV,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,2BAA2B;AAC3B,8EAA8E;AAE9E,SAAS,YAAY,CAAC,MAA0B,EAAE,GAAW;IAC3D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,CAAC,IAAI,CAAC,mBAAmB,GAAG,GAAG,CAAC,CAAC;IACtC,IAAI,MAAM,CAAC,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACvD,IAAI,MAAM,CAAC,iBAAiB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxC,KAAK,CAAC,IAAI,CAAC,kBAAkB,MAAM,CAAC,iBAAiB,CAAC,MAAM,CAAC,iBAAiB,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC;IAChG,CAAC;IACD,IAAI,MAAM,CAAC,SAAS;QAAE,KAAK,CAAC,IAAI,CAAC,yBAAyB,MAAM,CAAC,cAAc,SAAS,CAAC,CAAC;IAC1F,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACf,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,iCAAiC,CAAC,CAAC;IAEhE,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC;QAC9C,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK;YAAE,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,IAAI,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;IAC/E,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAE9E,MAAM,OAAO,gBAAgB;IAClB,IAAI,GAAG,eAAe,CAAC;IAEvB,UAAU,GAAmB;QACpC,IAAI,EAAE,eAAe;QACrB,WAAW,EACT,kFAAkF;YAClF,uFAAuF;YACvF,kFAAkF;YAClF,mFAAmF;YACnF,kFAAkF;YAClF,0FAA0F;YAC1F,sBAAsB,iBAAiB,qBAAqB;QAC9D,YAAY,EAAE;YACZ,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,GAAG,EAAE;oBACH,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,6DAA6D;iBAC3E;gBACD,SAAS,EAAE;oBACT,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,qDAAqD,iBAAiB,iBAAiB;iBACrG;aACF;YACD,QAAQ,EAAE,CAAC,KAAK,CAAC;SAClB;KACF,CAAC;IAEF,KAAK,CAAC,OAAO,CAAC,KAA8B;QAC1C,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAC9C,WAAW,CAAC,GAAG,CAAC,CAAC;QAEjB,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CACvB,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,iBAAiB,CAAC,CAAC,EAC9D,OAAO,CACR,CAAC;QAEF,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,uBAAuB,GAAG,cAAc,QAAQ,IAAI,CAAC,CAAC;QAE3E,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;QAEjD,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,yBAAyB,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG;YACtD,SAAS,MAAM,CAAC,cAAc,cAAc,MAAM,CAAC,SAAS,GAAG;YAC/D,SAAS,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,CACnC,CAAC;QAEF,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,iBAAiB,EAAE,CAAC;YACjE,OAAO,2BAA2B,GAAG,qCAAqC,CAAC;QAC7E,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IACnC,CAAC;CACF;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E,MAAM,UAAU,sBAAsB;IACpC,OAAO,IAAI,gBAAgB,EAAE,CAAC;AAChC,CAAC"}
@@ -1,61 +0,0 @@
1
- /**
2
- * GAIA Tool: web_browse — ADR-133-PR5
3
- *
4
- * Opens a URL in a headless Chromium browser via Playwright and extracts
5
- * page content (text, HTML, or screenshot). Covers the ~10-15pp of GAIA
6
- * Level-1 questions that require navigating dynamic JS pages, video pages
7
- * (YouTube, Vimeo), and paywalled/login-required content.
8
- *
9
- * ============================================================
10
- * PLAYWRIGHT DEPENDENCY — OPT-IN INSTALL
11
- * ============================================================
12
- * Playwright is NOT a hard runtime dep of @claude-flow/cli. It is loaded
13
- * lazily via a dynamic import so the package installs cleanly without it.
14
- *
15
- * To use web_browse, run once:
16
- * npm install playwright
17
- * npx playwright install chromium
18
- *
19
- * If Playwright is not installed, execute() returns a structured error
20
- * message that Claude can relay to the user rather than crashing.
21
- *
22
- * Install size: ~80 MB for the Playwright package + Chromium binary.
23
- * Add as a devDependency in benchmark-specific contexts to avoid bloating
24
- * the production bundle.
25
- *
26
- * ============================================================
27
- * RESOURCE CAPS
28
- * ============================================================
29
- * - Text/HTML extraction capped at 8 000 characters (prevents context-window
30
- * overflow; roughly 2 000 tokens).
31
- * - Default timeout: 30 seconds.
32
- * - Screenshots returned as base64-encoded PNG strings.
33
- * - Browser instance is always closed in a `finally` block.
34
- *
35
- * Refs: ADR-133, #2156
36
- */
37
- import { GaiaTool, ToolDefinition } from './types.js';
38
- export type BrowseExtract = 'text' | 'html' | 'screenshot';
39
- export interface WebBrowseInput {
40
- url: string;
41
- wait_for_selector?: string;
42
- extract?: BrowseExtract;
43
- timeout_seconds?: number;
44
- }
45
- export interface WebBrowseResult {
46
- content: string;
47
- final_url: string;
48
- status: number;
49
- truncated?: boolean;
50
- }
51
- export declare class WebBrowseTool implements GaiaTool {
52
- readonly name = "web_browse";
53
- readonly definition: ToolDefinition;
54
- execute(input: Record<string, unknown>): Promise<string>;
55
- }
56
- export interface WebBrowseToolOptions {
57
- /** Override default timeout in milliseconds. */
58
- defaultTimeoutMs?: number;
59
- }
60
- export declare function createWebBrowseTool(_opts?: WebBrowseToolOptions): WebBrowseTool;
61
- //# sourceMappingURL=web_browse.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"web_browse.d.ts","sourceRoot":"","sources":["../../../../src/benchmarks/gaia-tools/web_browse.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AAEH,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAatD,MAAM,MAAM,aAAa,GAAG,MAAM,GAAG,MAAM,GAAG,YAAY,CAAC;AAE3D,MAAM,WAAW,cAAc;IAC7B,GAAG,EAAE,MAAM,CAAC;IACZ,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,aAAa,CAAC;IACxB,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,OAAO,CAAC;CACrB;AA6JD,qBAAa,aAAc,YAAW,QAAQ;IAC5C,QAAQ,CAAC,IAAI,gBAAgB;IAE7B,QAAQ,CAAC,UAAU,EAAE,cAAc,CAoCjC;IAEI,OAAO,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;CAoB/D;AAMD,MAAM,WAAW,oBAAoB;IACnC,gDAAgD;IAChD,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,wBAAgB,mBAAmB,CAAC,KAAK,CAAC,EAAE,oBAAoB,GAAG,aAAa,CAE/E"}