harness-everything 0.28.0 → 0.30.0

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 (109) hide show
  1. package/.claude-plugin/marketplace.json +1 -1
  2. package/.claude-plugin/plugin.json +1 -1
  3. package/AGENTS.md +1 -0
  4. package/CHANGELOG.md +47 -0
  5. package/PRIVACY.md +9 -0
  6. package/benchmarks/fixtures/system-one-holdout-reviews.json +18 -18
  7. package/benchmarks/fixtures/system-one-holdout.json +9 -9
  8. package/benchmarks/fixtures/system-one-intent-holdout-draft.json +2228 -0
  9. package/benchmarks/fixtures/system-one-intent-holdout.json +16509 -0
  10. package/benchmarks/fixtures/system-one-intent-reviews.json +2723 -0
  11. package/benchmarks/results/system-one/harness-routing-v1-cpu-2026-09-24/README.md +75 -0
  12. package/benchmarks/results/system-one/harness-routing-v1-cpu-2026-09-24/report.json +10832 -0
  13. package/benchmarks/results/system-one/harness-routing-v1-ngram-2026-09-24/README.md +72 -0
  14. package/benchmarks/results/system-one/harness-routing-v1-ngram-2026-09-24/report.json +10817 -0
  15. package/benchmarks/results/system-one/harness-routing-v3-ngram-2026-09-24/README.md +63 -0
  16. package/benchmarks/results/system-one/harness-routing-v3-ngram-2026-09-24/report.json +10817 -0
  17. package/benchmarks/results/system-one/intent-ngram-v1-2026-09-24/README.md +64 -0
  18. package/benchmarks/results/system-one/intent-ngram-v1-2026-09-24/report.json +15177 -0
  19. package/benchmarks/results/system-one/labeler-holdout-check-2026-09-24/report.json +1104 -0
  20. package/benchmarks/results/system-one/labeler-holdout-check-codex-2026-09-24/report.json +1104 -0
  21. package/benchmarks/results/system-one/labeler-holdout-check-codex-tier3-2026-09-24/report.json +1104 -0
  22. package/benchmarks/results/system-one/labeler-holdout-check-intent-2026-09-24/report.json +2976 -0
  23. package/benchmarks/results/system-one/labeler-holdout-check-intent-scores-haiku-2026-09-24/report-005-steps.json +6125 -0
  24. package/benchmarks/results/system-one/labeler-holdout-check-intent-scores-haiku-2026-09-24/report-haiku-sonnet-escalation.json +6344 -0
  25. package/benchmarks/results/system-one/labeler-holdout-check-intent-scores-haiku-2026-09-24/report-scaled-margin-escalation.json +6206 -0
  26. package/benchmarks/results/system-one/labeler-holdout-check-intent-scores-haiku-2026-09-24/report.json +6103 -0
  27. package/benchmarks/results/system-one/laya-ft-2026-09-25/README.md +83 -0
  28. package/benchmarks/results/system-one/laya-ft-2026-09-25/report.json +71 -0
  29. package/benchmarks/results/system-one/laya-ft-2026-09-25/shadow-report.json +44 -0
  30. package/benchmarks/results/system-one/laya-zs-2026-09-25/README.md +68 -0
  31. package/benchmarks/results/system-one/laya-zs-2026-09-25/report.json +173 -0
  32. package/ci/mechanism-34-observations.test.js +550 -0
  33. package/ci/mechanism-system-one-collect.test.js +16 -0
  34. package/ci/mechanism-system-one-evaluation.test.js +17 -0
  35. package/ci/mechanism-system-one-intent.test.js +406 -0
  36. package/ci/mechanism-system-one-label.test.js +27 -0
  37. package/ci/mechanism-system-one-laya.test.js +45 -0
  38. package/ci/mechanism-system-one-ngram.test.js +161 -0
  39. package/ci/mechanism-system-one-relevance.test.js +112 -0
  40. package/ci/mechanism-system-one-resident.test.js +16 -0
  41. package/ci/mechanism-system-one-train.test.js +12 -0
  42. package/ci/system-one-laya-calibrate-test.py +65 -0
  43. package/ci/system-one-laya-export-test.py +155 -0
  44. package/ci/system-one-laya-label-test.py +201 -0
  45. package/ci/system-one-laya-perintent-test.py +60 -0
  46. package/ci/system-one-laya-provider-test.py +115 -0
  47. package/ci/system-one-laya-train-test.py +93 -0
  48. package/ci/system-one-ngram-parity.py +19 -0
  49. package/ci/system-one-train-test.py +6 -0
  50. package/docs/routing.md +25 -0
  51. package/docs/system-one-corpus.md +31 -3
  52. package/docs/system-one-intent-gates.md +102 -0
  53. package/docs/system-one-intent.md +173 -0
  54. package/docs/system-one-laya-training.md +38 -0
  55. package/docs/system-one-observations.md +206 -0
  56. package/docs/system-one-routing.md +102 -5
  57. package/docs/system-one-suggestion-gates.md +292 -0
  58. package/docs/system-one-training.md +69 -7
  59. package/harness-everything/SKILL.md +1 -1
  60. package/harness-everything/scripts/kernel-router.js +10 -0
  61. package/harness-everything/scripts/system-one/catalogs.js +35 -0
  62. package/harness-everything/scripts/system-one/contract.js +45 -1
  63. package/harness-everything/scripts/system-one/evaluate.js +118 -21
  64. package/harness-everything/scripts/system-one/intent.js +22 -0
  65. package/harness-everything/scripts/system-one/laya_adapter.py +141 -0
  66. package/harness-everything/scripts/system-one/ngram.js +84 -0
  67. package/harness-everything/scripts/system-one/provider.js +24 -7
  68. package/harness-everything/scripts/system-one/resident.js +8 -3
  69. package/harness-everything/scripts/system-one/router.js +1 -1
  70. package/hooks/hooks.json +40 -0
  71. package/hooks/scripts/codex-user-prompt.js +39 -0
  72. package/hooks/scripts/lib/observations.js +581 -0
  73. package/hooks/scripts/observation-hook.js +25 -0
  74. package/opencode-plugin/plugin.json +1 -1
  75. package/package.json +1 -1
  76. package/plugins/harness-everything/.codex-plugin/plugin.json +1 -1
  77. package/plugins/harness-everything/hooks/hooks.json +35 -2
  78. package/plugins/harness-everything/hooks/scripts/codex-user-prompt.js +39 -0
  79. package/plugins/harness-everything/hooks/scripts/lib/observations.js +581 -0
  80. package/plugins/harness-everything/hooks/scripts/observation-hook.js +25 -0
  81. package/plugins/harness-everything/plugin.json +1 -1
  82. package/plugins/harness-everything/skills/harness-everything/SKILL.md +1 -1
  83. package/plugins/harness-everything/skills/harness-everything/scripts/kernel-router.js +10 -0
  84. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/catalogs.js +35 -0
  85. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/contract.js +45 -1
  86. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/evaluate.js +118 -21
  87. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/intent.js +22 -0
  88. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/laya_adapter.py +141 -0
  89. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/ngram.js +84 -0
  90. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/provider.js +24 -7
  91. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/resident.js +8 -3
  92. package/plugins/harness-everything/skills/harness-everything/scripts/system-one/router.js +1 -1
  93. package/sbom.cdx.json +7 -7
  94. package/scripts/evaluate-system-one.js +32 -13
  95. package/scripts/system-one-calibrate-laya.py +165 -0
  96. package/scripts/system-one-calibrate.js +42 -13
  97. package/scripts/system-one-collect.js +7 -3
  98. package/scripts/system-one-compare-laya.py +156 -0
  99. package/scripts/system-one-corpus.js +22 -15
  100. package/scripts/system-one-eval-laya.js +112 -0
  101. package/scripts/system-one-export-laya.py +197 -0
  102. package/scripts/system-one-label-laya.py +243 -0
  103. package/scripts/system-one-label.js +192 -37
  104. package/scripts/system-one-observations-export.js +106 -0
  105. package/scripts/system-one-observations-review.js +86 -0
  106. package/scripts/system-one-perintent-laya.py +134 -0
  107. package/scripts/system-one-train-laya.py +344 -0
  108. package/scripts/system-one-train-ngram.py +181 -0
  109. package/scripts/system-one-train.py +14 -1
@@ -21,13 +21,24 @@
21
21
  "hooks": [
22
22
  {
23
23
  "type": "command",
24
- "command": "node \"${PLUGIN_ROOT}/skills/harness-everything/scripts/kernel-router.js\"",
25
- "commandWindows": "node \"${PLUGIN_ROOT}\\skills\\harness-everything\\scripts\\kernel-router.js\"",
24
+ "command": "node \"${PLUGIN_ROOT}/hooks/scripts/codex-user-prompt.js\"",
25
+ "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\codex-user-prompt.js\"",
26
26
  "timeout": 30,
27
27
  "statusMessage": "Routing with Harness workflow contract",
28
28
  "additionalContextLimit": 2500
29
29
  }
30
30
  ]
31
+ },
32
+ {
33
+ "hooks": [
34
+ {
35
+ "type": "command",
36
+ "command": "node \"${PLUGIN_ROOT}/hooks/scripts/observation-hook.js\"",
37
+ "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\observation-hook.js\"",
38
+ "timeout": 10,
39
+ "statusMessage": "Recording a local System One turn observation"
40
+ }
41
+ ]
31
42
  }
32
43
  ],
33
44
  "PreToolUse": [
@@ -205,6 +216,17 @@
205
216
  "statusMessage": "Recording local Harness skill telemetry"
206
217
  }
207
218
  ]
219
+ },
220
+ {
221
+ "hooks": [
222
+ {
223
+ "type": "command",
224
+ "command": "node \"${PLUGIN_ROOT}/hooks/scripts/observation-hook.js\"",
225
+ "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\observation-hook.js\"",
226
+ "timeout": 10,
227
+ "statusMessage": "Recording a local System One turn observation"
228
+ }
229
+ ]
208
230
  }
209
231
  ],
210
232
  "SubagentStart": [
@@ -269,6 +291,17 @@
269
291
  "statusMessage": "Recording local Harness skill telemetry"
270
292
  }
271
293
  ]
294
+ },
295
+ {
296
+ "hooks": [
297
+ {
298
+ "type": "command",
299
+ "command": "node \"${PLUGIN_ROOT}/hooks/scripts/observation-hook.js\"",
300
+ "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\observation-hook.js\"",
301
+ "timeout": 10,
302
+ "statusMessage": "Recording a local System One turn observation"
303
+ }
304
+ ]
272
305
  }
273
306
  ]
274
307
  }
@@ -0,0 +1,39 @@
1
+ #!/usr/bin/env node
2
+ 'use strict';
3
+ // Codex UserPromptSubmit adapter (docs/system-one-observations.md#hosts).
4
+ // Codex adds prompt-time context to the model only from JSON
5
+ // hookSpecificOutput.additionalContext; the router's plain stdout never
6
+ // reaches the model. This runs the kernel router and wraps its output.
7
+
8
+ const fs = require('fs');
9
+ const path = require('path');
10
+ const { spawnSync } = require('child_process');
11
+
12
+ // Packaged plugin layout first, then the canonical repository layout.
13
+ const ROUTER = [
14
+ path.join(__dirname, '..', '..', 'skills', 'harness-everything', 'scripts', 'kernel-router.js'),
15
+ path.join(__dirname, '..', '..', 'harness-everything', 'scripts', 'kernel-router.js'),
16
+ ].find(file => fs.existsSync(file));
17
+
18
+ let input = '';
19
+ let finished = false;
20
+
21
+ function finish() {
22
+ if (finished) return;
23
+ finished = true;
24
+ clearTimeout(timer);
25
+ if (!ROUTER) process.exit(0);
26
+ const result = spawnSync(process.execPath, [ROUTER], { input, encoding: 'utf8', env: process.env, timeout: 25000 });
27
+ if (result.stderr) process.stderr.write(result.stderr);
28
+ const context = String(result.stdout || '').trim();
29
+ if (context) {
30
+ process.stdout.write(JSON.stringify({ hookSpecificOutput: { hookEventName: 'UserPromptSubmit', additionalContext: context } }));
31
+ }
32
+ process.exit(0);
33
+ }
34
+
35
+ const timer = setTimeout(finish, 1000);
36
+ process.stdin.setEncoding('utf8');
37
+ process.stdin.on('data', chunk => { input += chunk; });
38
+ process.stdin.on('end', finish);
39
+ process.stdin.on('error', finish);