@hassangameryt/freeflow 0.1.0 → 0.3.0

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 (630) hide show
  1. package/.agents/plugins/marketplace.json +1 -1
  2. package/.claude-plugin/marketplace.json +2 -2
  3. package/{plugins/freeflow/.claude-plugin → .claude-plugin}/plugin.json +1 -1
  4. package/{plugins/freeflow/.codex-plugin → .codex-plugin}/plugin.json +1 -1
  5. package/README.md +25 -16
  6. package/assets/freeflow-plugin-icon.png +0 -0
  7. package/{plugins/freeflow/command-surface.json → command-surface.json} +10 -10
  8. package/evals/README.md +32 -0
  9. package/evals/fixtures/copy-conventions.md +31 -0
  10. package/evals/fixtures/output-router/README.md +12 -0
  11. package/evals/fixtures/output-router/failed-command-output.txt +9 -0
  12. package/evals/fixtures/output-router/large-router-manual.md +63 -0
  13. package/evals/fixtures/output-router/native-large-output.txt +40 -0
  14. package/evals/fixtures/output-router/noisy-test-output.txt +30 -0
  15. package/evals/fixtures/output-router/verification-output.txt +8 -0
  16. package/evals/fixtures/tiny-artifact-review-app/README.md +5 -0
  17. package/evals/fixtures/tiny-artifact-review-app/docs/billing-policy.md +10 -0
  18. package/evals/fixtures/tiny-artifact-review-app/docs/specs/billing-webhook-api-draft.md +21 -0
  19. package/evals/fixtures/tiny-artifact-review-app/docs/specs/failed-payment-grace-period.md +24 -0
  20. package/evals/fixtures/tiny-artifact-review-app/docs/specs/immediate-downgrade.md +17 -0
  21. package/evals/fixtures/tiny-artifact-review-app/tests/billing.test.md +8 -0
  22. package/evals/fixtures/tiny-commit-ambiguous-app/.freeflow-eval-setup.sh +15 -0
  23. package/evals/fixtures/tiny-commit-ambiguous-app/README.md +3 -0
  24. package/evals/fixtures/tiny-commit-ambiguous-app/src/billing.ts +5 -0
  25. package/evals/fixtures/tiny-commit-ambiguous-app/src/settings-page.ts +1 -0
  26. package/evals/fixtures/tiny-commit-ambiguous-app/tests/settings.test.md +5 -0
  27. package/evals/fixtures/tiny-commit-app/.freeflow-eval-setup.sh +14 -0
  28. package/evals/fixtures/tiny-commit-app/README.md +3 -0
  29. package/evals/fixtures/tiny-commit-app/docs/specs/settings-title.md +5 -0
  30. package/evals/fixtures/tiny-commit-app/src/billing.ts +5 -0
  31. package/evals/fixtures/tiny-commit-app/src/settings-page.ts +1 -0
  32. package/evals/fixtures/tiny-commit-app/tests/settings.test.md +5 -0
  33. package/evals/fixtures/tiny-commit-everything-app/.freeflow-eval-setup.sh +15 -0
  34. package/evals/fixtures/tiny-commit-everything-app/README.md +3 -0
  35. package/evals/fixtures/tiny-commit-everything-app/docs/specs/settings-title.md +5 -0
  36. package/evals/fixtures/tiny-commit-everything-app/src/billing.ts +5 -0
  37. package/evals/fixtures/tiny-commit-everything-app/src/settings-page.ts +1 -0
  38. package/evals/fixtures/tiny-commit-everything-app/tests/settings.test.md +5 -0
  39. package/evals/fixtures/tiny-commit-staged-app/.freeflow-eval-setup.sh +15 -0
  40. package/evals/fixtures/tiny-commit-staged-app/README.md +3 -0
  41. package/evals/fixtures/tiny-commit-staged-app/docs/specs/settings-title.md +5 -0
  42. package/evals/fixtures/tiny-commit-staged-app/src/billing.ts +5 -0
  43. package/evals/fixtures/tiny-commit-staged-app/src/settings-page.ts +1 -0
  44. package/evals/fixtures/tiny-commit-staged-app/tests/settings.test.md +5 -0
  45. package/evals/fixtures/tiny-commit-staged-mixed-app/.freeflow-eval-setup.sh +15 -0
  46. package/evals/fixtures/tiny-commit-staged-mixed-app/README.md +3 -0
  47. package/evals/fixtures/tiny-commit-staged-mixed-app/docs/billing-policy.md +3 -0
  48. package/evals/fixtures/tiny-commit-staged-mixed-app/docs/release-notes.md +3 -0
  49. package/evals/fixtures/tiny-commit-staged-mixed-app/src/billing.ts +5 -0
  50. package/evals/fixtures/tiny-dashboard-stale-app/README.md +11 -0
  51. package/evals/fixtures/tiny-dashboard-stale-app/src/api-client.ts +13 -0
  52. package/evals/fixtures/tiny-dashboard-stale-app/src/dashboard-data.ts +19 -0
  53. package/evals/fixtures/tiny-dashboard-stale-app/tests/dashboard-data.test.md +9 -0
  54. package/evals/fixtures/tiny-decision-destination-app/README.md +6 -0
  55. package/evals/fixtures/tiny-decision-destination-app/src/risk.ts +8 -0
  56. package/evals/fixtures/tiny-decision-destination-app/tests/risk.test.md +4 -0
  57. package/evals/fixtures/tiny-design-depth-app/README.md +15 -0
  58. package/evals/fixtures/tiny-design-depth-app/src/routes.ts +25 -0
  59. package/evals/fixtures/tiny-design-depth-app/src/sendEmail.ts +3 -0
  60. package/evals/fixtures/tiny-design-depth-app/src/sendSms.ts +3 -0
  61. package/evals/fixtures/tiny-design-depth-app/src/telemetry.ts +3 -0
  62. package/evals/fixtures/tiny-design-depth-app/tests/routes.test.md +12 -0
  63. package/evals/fixtures/tiny-eval-method-app/README.md +8 -0
  64. package/evals/fixtures/tiny-eval-method-app/evals/reports/review-pr-eval-result.md +5 -0
  65. package/evals/fixtures/tiny-eval-method-app/evals/reports/triage-question-failure.md +17 -0
  66. package/evals/fixtures/tiny-eval-method-app/evals/runs/review-pr-with-skill-output.diff +11 -0
  67. package/evals/fixtures/tiny-eval-method-app/evals/runs/review-pr-with-skill-output.md +3 -0
  68. package/evals/fixtures/tiny-eval-method-app/skills/review-pr/SKILL.md +10 -0
  69. package/evals/fixtures/tiny-eval-method-app/skills/triage-question/SKILL.md +10 -0
  70. package/evals/fixtures/tiny-execute-plan-app/README.md +5 -0
  71. package/evals/fixtures/tiny-execute-plan-app/checks/settings-and-billing-check.sh +5 -0
  72. package/evals/fixtures/tiny-execute-plan-app/docs/billing-policy.md +10 -0
  73. package/evals/fixtures/tiny-execute-plan-app/docs/plans/immediate-downgrade-plan.md +11 -0
  74. package/evals/fixtures/tiny-execute-plan-app/docs/plans/settings-title-plan.md +12 -0
  75. package/evals/fixtures/tiny-execute-plan-app/docs/plans/settings-title-with-bad-check.md +15 -0
  76. package/evals/fixtures/tiny-execute-plan-app/docs/plans/unverified-billing-plan.md +9 -0
  77. package/evals/fixtures/tiny-execute-plan-app/docs/specs/failed-payment-grace-period.md +9 -0
  78. package/evals/fixtures/tiny-execute-plan-app/docs/specs/settings-title.md +9 -0
  79. package/evals/fixtures/tiny-execute-plan-app/src/billing.ts +7 -0
  80. package/evals/fixtures/tiny-execute-plan-app/src/settings-page.ts +1 -0
  81. package/evals/fixtures/tiny-execute-plan-app/tests/billing.test.md +7 -0
  82. package/evals/fixtures/tiny-execute-plan-app/tests/settings.test.md +5 -0
  83. package/evals/fixtures/tiny-execute-review-loop-app/README.md +5 -0
  84. package/evals/fixtures/tiny-execute-review-loop-app/docs/billing-policy.md +9 -0
  85. package/evals/fixtures/tiny-execute-review-loop-app/docs/plans/settings-title-plan.md +12 -0
  86. package/evals/fixtures/tiny-execute-review-loop-app/docs/review/previous-agent-summary.md +15 -0
  87. package/evals/fixtures/tiny-execute-review-loop-app/docs/review/review-pass-1-non-pass.md +11 -0
  88. package/evals/fixtures/tiny-execute-review-loop-app/docs/specs/settings-title.md +9 -0
  89. package/evals/fixtures/tiny-execute-review-loop-app/src/billing.ts +7 -0
  90. package/evals/fixtures/tiny-execute-review-loop-app/src/settings-page.ts +1 -0
  91. package/evals/fixtures/tiny-execute-review-loop-app/tests/billing.test.md +7 -0
  92. package/evals/fixtures/tiny-execute-review-loop-app/tests/settings.test.md +5 -0
  93. package/evals/fixtures/tiny-flaky-session-app/README.md +10 -0
  94. package/evals/fixtures/tiny-flaky-session-app/src/session-timeout.ts +13 -0
  95. package/evals/fixtures/tiny-flaky-session-app/tests/session-timeout.test.md +8 -0
  96. package/evals/fixtures/tiny-handoff-app/README.md +5 -0
  97. package/evals/fixtures/tiny-handoff-app/docs/billing-policy.md +10 -0
  98. package/evals/fixtures/tiny-handoff-app/docs/handoffs/old-agent-handoff.md +8 -0
  99. package/evals/fixtures/tiny-handoff-app/docs/plans/billing-retry-plan.md +11 -0
  100. package/evals/fixtures/tiny-handoff-app/src/billing.ts +7 -0
  101. package/evals/fixtures/tiny-handoff-report-app/README.md +3 -0
  102. package/evals/fixtures/tiny-handoff-report-app/evals/runs/handoff-3/hof-003/with-skill-output.diff +1 -0
  103. package/evals/fixtures/tiny-handoff-report-app/evals/runs/handoff-3/hof-003/with-skill-output.md +1 -0
  104. package/evals/fixtures/tiny-handoff-report-app/evals/scores/hof-003-score.txt +4 -0
  105. package/evals/fixtures/tiny-plan-app/README.md +5 -0
  106. package/evals/fixtures/tiny-plan-app/docs/billing-policy.md +10 -0
  107. package/evals/fixtures/tiny-plan-app/docs/specs/failed-payment-grace-period.md +25 -0
  108. package/evals/fixtures/tiny-plan-app/src/billing.ts +17 -0
  109. package/evals/fixtures/tiny-plan-app/src/settings-page.ts +1 -0
  110. package/evals/fixtures/tiny-plan-app/tests/billing.test.md +8 -0
  111. package/evals/fixtures/tiny-plan-app/tests/settings.test.md +5 -0
  112. package/evals/fixtures/tiny-plan-strict-api-app/README.md +3 -0
  113. package/evals/fixtures/tiny-plan-strict-api-app/docs/billing-policy.md +5 -0
  114. package/evals/fixtures/tiny-plan-strict-api-app/docs/specs/billing-webhook-api.md +25 -0
  115. package/evals/fixtures/tiny-plan-strict-api-app/src/billing.ts +5 -0
  116. package/evals/fixtures/tiny-plan-strict-api-app/src/webhooks.ts +3 -0
  117. package/evals/fixtures/tiny-plan-strict-api-app/tests/billing-webhook.test.md +8 -0
  118. package/evals/fixtures/tiny-post-setup-invalid-config-app/.freeflow/config.json +3 -0
  119. package/evals/fixtures/tiny-post-setup-invalid-config-app/AGENTS.md +15 -0
  120. package/evals/fixtures/tiny-post-setup-invalid-config-app/README.md +16 -0
  121. package/evals/fixtures/tiny-post-setup-invalid-config-app/docs/auth-review.md +15 -0
  122. package/evals/fixtures/tiny-post-setup-invalid-config-app/docs/billing-policy.md +10 -0
  123. package/evals/fixtures/tiny-post-setup-invalid-config-app/docs/copy-conventions.md +15 -0
  124. package/evals/fixtures/tiny-post-setup-invalid-config-app/src/auth-middleware.ts +12 -0
  125. package/evals/fixtures/tiny-post-setup-invalid-config-app/src/billing.ts +20 -0
  126. package/evals/fixtures/tiny-post-setup-invalid-config-app/src/login-copy.ts +1 -0
  127. package/evals/fixtures/tiny-post-setup-invalid-config-app/src/settings-page.ts +1 -0
  128. package/evals/fixtures/tiny-post-setup-invalid-config-app/tests/auth-middleware.test.md +8 -0
  129. package/evals/fixtures/tiny-post-setup-invalid-config-app/tests/billing.test.md +8 -0
  130. package/evals/fixtures/tiny-post-setup-invalid-config-app/tests/copy.test.md +7 -0
  131. package/evals/fixtures/tiny-post-setup-missing-config-app/AGENTS.md +15 -0
  132. package/evals/fixtures/tiny-post-setup-missing-config-app/README.md +16 -0
  133. package/evals/fixtures/tiny-post-setup-missing-config-app/docs/auth-review.md +15 -0
  134. package/evals/fixtures/tiny-post-setup-missing-config-app/docs/billing-policy.md +10 -0
  135. package/evals/fixtures/tiny-post-setup-missing-config-app/docs/copy-conventions.md +15 -0
  136. package/evals/fixtures/tiny-post-setup-missing-config-app/src/auth-middleware.ts +12 -0
  137. package/evals/fixtures/tiny-post-setup-missing-config-app/src/billing.ts +20 -0
  138. package/evals/fixtures/tiny-post-setup-missing-config-app/src/login-copy.ts +1 -0
  139. package/evals/fixtures/tiny-post-setup-missing-config-app/src/settings-page.ts +1 -0
  140. package/evals/fixtures/tiny-post-setup-missing-config-app/tests/auth-middleware.test.md +8 -0
  141. package/evals/fixtures/tiny-post-setup-missing-config-app/tests/billing.test.md +8 -0
  142. package/evals/fixtures/tiny-post-setup-missing-config-app/tests/copy.test.md +7 -0
  143. package/evals/fixtures/tiny-post-setup-source-conflict-app/.freeflow/config.json +3 -0
  144. package/evals/fixtures/tiny-post-setup-source-conflict-app/AGENTS.md +15 -0
  145. package/evals/fixtures/tiny-post-setup-source-conflict-app/README.md +16 -0
  146. package/evals/fixtures/tiny-post-setup-source-conflict-app/docs/auth-review.md +15 -0
  147. package/evals/fixtures/tiny-post-setup-source-conflict-app/docs/billing-policy.md +10 -0
  148. package/evals/fixtures/tiny-post-setup-source-conflict-app/docs/copy-conventions.md +15 -0
  149. package/evals/fixtures/tiny-post-setup-source-conflict-app/src/auth-middleware.ts +12 -0
  150. package/evals/fixtures/tiny-post-setup-source-conflict-app/src/billing.ts +20 -0
  151. package/evals/fixtures/tiny-post-setup-source-conflict-app/src/login-copy.ts +1 -0
  152. package/evals/fixtures/tiny-post-setup-source-conflict-app/src/settings-page.ts +1 -0
  153. package/evals/fixtures/tiny-post-setup-source-conflict-app/tests/auth-middleware.test.md +8 -0
  154. package/evals/fixtures/tiny-post-setup-source-conflict-app/tests/billing.test.md +8 -0
  155. package/evals/fixtures/tiny-post-setup-source-conflict-app/tests/copy.test.md +7 -0
  156. package/evals/fixtures/tiny-report-performance-app/README.md +11 -0
  157. package/evals/fixtures/tiny-report-performance-app/docs/performance.md +10 -0
  158. package/evals/fixtures/tiny-report-performance-app/src/report-summary.ts +29 -0
  159. package/evals/fixtures/tiny-report-performance-app/tests/report-performance.test.md +8 -0
  160. package/evals/fixtures/tiny-review-request-app/README.md +3 -0
  161. package/evals/fixtures/tiny-review-request-app/docs/billing-policy.md +5 -0
  162. package/evals/fixtures/tiny-review-request-app/docs/plans/billing-webhook-api.md +19 -0
  163. package/evals/fixtures/tiny-review-request-app/docs/review/previous-agent-summary.md +20 -0
  164. package/evals/fixtures/tiny-review-request-app/docs/specs/billing-webhook-api.md +25 -0
  165. package/evals/fixtures/tiny-review-request-app/src/billing.ts +11 -0
  166. package/evals/fixtures/tiny-review-request-app/src/webhooks.ts +19 -0
  167. package/evals/fixtures/tiny-review-request-app/tests/billing-webhook.test.md +8 -0
  168. package/evals/fixtures/tiny-review-work-app/README.md +5 -0
  169. package/evals/fixtures/tiny-review-work-app/docs/billing-policy.md +9 -0
  170. package/evals/fixtures/tiny-review-work-app/docs/plans/settings-title-plan.md +12 -0
  171. package/evals/fixtures/tiny-review-work-app/docs/review/previous-agent-summary.md +15 -0
  172. package/evals/fixtures/tiny-review-work-app/docs/specs/settings-title.md +9 -0
  173. package/evals/fixtures/tiny-review-work-app/src/billing.ts +7 -0
  174. package/evals/fixtures/tiny-review-work-app/src/settings-page.ts +1 -0
  175. package/evals/fixtures/tiny-review-work-app/tests/billing.test.md +7 -0
  176. package/evals/fixtures/tiny-review-work-app/tests/settings.test.md +5 -0
  177. package/evals/fixtures/tiny-saas-app/README.md +16 -0
  178. package/evals/fixtures/tiny-saas-app/docs/auth-review.md +15 -0
  179. package/evals/fixtures/tiny-saas-app/docs/billing-policy.md +10 -0
  180. package/evals/fixtures/tiny-saas-app/docs/copy-conventions.md +15 -0
  181. package/evals/fixtures/tiny-saas-app/docs/specs/settings-title.md +13 -0
  182. package/evals/fixtures/tiny-saas-app/src/auth-middleware.ts +12 -0
  183. package/evals/fixtures/tiny-saas-app/src/billing.ts +20 -0
  184. package/evals/fixtures/tiny-saas-app/src/login-copy.ts +1 -0
  185. package/evals/fixtures/tiny-saas-app/src/settings-page.ts +1 -0
  186. package/evals/fixtures/tiny-saas-app/tests/auth-middleware.test.md +8 -0
  187. package/evals/fixtures/tiny-saas-app/tests/billing.test.md +8 -0
  188. package/evals/fixtures/tiny-saas-app/tests/copy.test.md +7 -0
  189. package/evals/fixtures/tiny-setup-agents-app/AGENTS.md +6 -0
  190. package/evals/fixtures/tiny-setup-both-app/AGENTS.md +5 -0
  191. package/evals/fixtures/tiny-setup-both-app/CLAUDE.md +5 -0
  192. package/evals/fixtures/tiny-setup-claude-app/CLAUDE.md +6 -0
  193. package/evals/fixtures/tiny-setup-codex-rules-app/.codex/rules/shell.rules +5 -0
  194. package/evals/fixtures/tiny-setup-codex-rules-app/AGENTS.md +6 -0
  195. package/evals/fixtures/tiny-setup-conflict-app/AGENTS.md +6 -0
  196. package/evals/fixtures/tiny-setup-existing-block-app/AGENTS.md +14 -0
  197. package/evals/fixtures/tiny-setup-four-rules-app/CLAUDE.md +6 -0
  198. package/evals/fixtures/tiny-skill-dev-app/README.md +10 -0
  199. package/evals/fixtures/tiny-skill-dev-app/evals/prompts/review-pr-001.txt +8 -0
  200. package/evals/fixtures/tiny-skill-dev-app/evals/reports/review-pr-failure.md +17 -0
  201. package/evals/fixtures/tiny-skill-dev-app/skills/review-pr/SKILL.md +13 -0
  202. package/evals/fixtures/tiny-team-app/README.md +5 -0
  203. package/evals/fixtures/tiny-team-app/docs/product.md +6 -0
  204. package/evals/fixtures/tiny-team-app/src/teams.ts +9 -0
  205. package/evals/fixtures/tiny-team-app/src/users.ts +5 -0
  206. package/evals/fixtures/tiny-team-app/tests/teams.test.md +7 -0
  207. package/evals/fixtures/tiny-verify-app/README.md +5 -0
  208. package/evals/fixtures/tiny-verify-app/src/settings-nav.ts +1 -0
  209. package/evals/fixtures/tiny-verify-app/src/settings-page.ts +1 -0
  210. package/evals/fixtures/tiny-verify-app/tests/settings.test.md +7 -0
  211. package/evals/prompts/aon-001.txt +2 -0
  212. package/evals/prompts/byp-001.txt +2 -0
  213. package/evals/prompts/byp-002.txt +2 -0
  214. package/evals/prompts/cap-001.txt +3 -0
  215. package/evals/prompts/cap-002.txt +3 -0
  216. package/evals/prompts/cmd-001.txt +3 -0
  217. package/evals/prompts/cmd-002.txt +3 -0
  218. package/evals/prompts/cmd-003.txt +6 -0
  219. package/evals/prompts/cmd-004.txt +5 -0
  220. package/evals/prompts/cmd-005.txt +5 -0
  221. package/evals/prompts/cmd-006.txt +5 -0
  222. package/evals/prompts/cmd-007.txt +5 -0
  223. package/evals/prompts/cmd-008.txt +5 -0
  224. package/evals/prompts/cmd-009.txt +8 -0
  225. package/evals/prompts/cmd-010.txt +5 -0
  226. package/evals/prompts/cmd-011.txt +5 -0
  227. package/evals/prompts/cmd-012.txt +5 -0
  228. package/evals/prompts/cmd-013.txt +5 -0
  229. package/evals/prompts/cmd-014.txt +7 -0
  230. package/evals/prompts/cmd-015.txt +7 -0
  231. package/evals/prompts/cmd-016.txt +5 -0
  232. package/evals/prompts/cmt-001.txt +1 -0
  233. package/evals/prompts/cmt-002.txt +1 -0
  234. package/evals/prompts/cmt-003.txt +1 -0
  235. package/evals/prompts/cmt-004.txt +1 -0
  236. package/evals/prompts/cmt-005.txt +1 -0
  237. package/evals/prompts/dfd-001.txt +3 -0
  238. package/evals/prompts/dia-001.txt +3 -0
  239. package/evals/prompts/dia-002.txt +3 -0
  240. package/evals/prompts/dia-003.txt +3 -0
  241. package/evals/prompts/dis-001.txt +5 -0
  242. package/evals/prompts/dis-002.txt +5 -0
  243. package/evals/prompts/esk-001.txt +5 -0
  244. package/evals/prompts/esk-002.txt +5 -0
  245. package/evals/prompts/esk-003.txt +5 -0
  246. package/evals/prompts/grc-001.txt +3 -0
  247. package/evals/prompts/hof-001.txt +3 -0
  248. package/evals/prompts/hof-002.txt +1 -0
  249. package/evals/prompts/hof-003.txt +1 -0
  250. package/evals/prompts/hof-004.txt +1 -0
  251. package/evals/prompts/hof-005.txt +1 -0
  252. package/evals/prompts/ivg-001.txt +1 -0
  253. package/evals/prompts/ivg-002.txt +11 -0
  254. package/evals/prompts/ivg-003.txt +1 -0
  255. package/evals/prompts/ivg-004.txt +1 -0
  256. package/evals/prompts/mode-001.txt +3 -0
  257. package/evals/prompts/mode-002.txt +1 -0
  258. package/evals/prompts/mode-003.txt +3 -0
  259. package/evals/prompts/mode-004.txt +3 -0
  260. package/evals/prompts/mode-005.txt +3 -0
  261. package/evals/prompts/mode-006.txt +3 -0
  262. package/evals/prompts/mode-007.txt +3 -0
  263. package/evals/prompts/otr-001.txt +3 -0
  264. package/evals/prompts/otr-002.txt +3 -0
  265. package/evals/prompts/rar-001.txt +5 -0
  266. package/evals/prompts/rar-002.txt +3 -0
  267. package/evals/prompts/rar-003.txt +6 -0
  268. package/evals/prompts/rar-004.txt +15 -0
  269. package/evals/prompts/rar-005.txt +10 -0
  270. package/evals/prompts/rbr-001.txt +3 -0
  271. package/evals/prompts/rbr-002.txt +3 -0
  272. package/evals/prompts/rbr-003.txt +7 -0
  273. package/evals/prompts/rev-001.txt +3 -0
  274. package/evals/prompts/rev-002.txt +5 -0
  275. package/evals/prompts/rev-003.txt +6 -0
  276. package/evals/prompts/rev-004.txt +1 -0
  277. package/evals/prompts/rev-005.txt +15 -0
  278. package/evals/prompts/rev-006.txt +5 -0
  279. package/evals/prompts/stp-001.txt +1 -0
  280. package/evals/prompts/stp-002.txt +1 -0
  281. package/evals/prompts/stp-003.txt +1 -0
  282. package/evals/prompts/stp-004.txt +1 -0
  283. package/evals/prompts/stp-005.txt +1 -0
  284. package/evals/prompts/stp-006.txt +1 -0
  285. package/evals/prompts/stp-007.txt +1 -0
  286. package/evals/prompts/stp-008.txt +1 -0
  287. package/evals/prompts/stp-009.txt +1 -0
  288. package/evals/prompts/stp-010.txt +1 -0
  289. package/evals/prompts/stp-011.txt +1 -0
  290. package/evals/prompts/vfy-003.txt +1 -0
  291. package/evals/prompts/wpl-001.txt +3 -0
  292. package/evals/prompts/wpl-002.txt +5 -0
  293. package/evals/prompts/wpl-003.txt +3 -0
  294. package/evals/prompts/wpl-004.txt +5 -0
  295. package/evals/prompts/wpl-005.txt +1 -0
  296. package/evals/prompts/wrc-001.txt +10 -0
  297. package/evals/prompts/wrc-002.txt +3 -0
  298. package/evals/prompts/wsk-001.txt +6 -0
  299. package/evals/prompts/wsp-001.txt +5 -0
  300. package/evals/prompts/wsp-002.txt +15 -0
  301. package/evals/prompts/wsp-003.txt +3 -0
  302. package/evals/prompts/wsp-004.txt +15 -0
  303. package/evals/prompts/wsp-005.txt +5 -0
  304. package/evals/prompts/wsp-006.txt +12 -0
  305. package/evals/prompts/wsp-007.txt +9 -0
  306. package/evals/prompts/wsp-008.txt +13 -0
  307. package/evals/prompts/xpl-001.txt +1 -0
  308. package/evals/prompts/xpl-002.txt +3 -0
  309. package/evals/prompts/xpl-003.txt +3 -0
  310. package/evals/prompts/xpl-004.txt +3 -0
  311. package/evals/prompts/xpl-005.txt +5 -0
  312. package/evals/prompts/xpl-006.txt +3 -0
  313. package/evals/registries/adversarial-evals.json +39 -0
  314. package/evals/registries/fixture-evals.json +2513 -0
  315. package/evals/registries/handoff-evals.json +74 -0
  316. package/evals/registries/legacy-evals.json +147 -0
  317. package/evals/registries/skill-evidence.json +193 -0
  318. package/evals/registries/verify-work-evals.json +46 -0
  319. package/evals/reports/acceptance/v0.1-acceptance-report.md +130 -0
  320. package/evals/reports/by-command-surface/command-surface-1-report.md +45 -0
  321. package/evals/reports/by-command-surface/command-surface-10-report.md +76 -0
  322. package/evals/reports/by-command-surface/command-surface-11-report.md +77 -0
  323. package/evals/reports/by-command-surface/command-surface-12-report.md +91 -0
  324. package/evals/reports/by-command-surface/command-surface-13-report.md +64 -0
  325. package/evals/reports/by-command-surface/command-surface-14-report.md +66 -0
  326. package/evals/reports/by-command-surface/command-surface-15-report.md +124 -0
  327. package/evals/reports/by-command-surface/command-surface-16-report.md +39 -0
  328. package/evals/reports/by-command-surface/command-surface-2-report.md +80 -0
  329. package/evals/reports/by-command-surface/command-surface-3-report.md +86 -0
  330. package/evals/reports/by-command-surface/command-surface-4-report.md +76 -0
  331. package/evals/reports/by-command-surface/command-surface-5-report.md +80 -0
  332. package/evals/reports/by-command-surface/command-surface-6-report.md +74 -0
  333. package/evals/reports/by-command-surface/command-surface-7-report.md +84 -0
  334. package/evals/reports/by-command-surface/command-surface-8-report.md +79 -0
  335. package/evals/reports/by-command-surface/command-surface-9-report.md +79 -0
  336. package/evals/reports/by-command-surface/command-surface-matrix.md +91 -0
  337. package/evals/reports/by-skill/bypass-1-report.md +124 -0
  338. package/evals/reports/by-skill/capture-decisions-1-report.md +99 -0
  339. package/evals/reports/by-skill/capture-decisions-2-report.md +102 -0
  340. package/evals/reports/by-skill/commit-work-1-report.md +165 -0
  341. package/evals/reports/by-skill/commit-work-2-report.md +109 -0
  342. package/evals/reports/by-skill/design-for-depth-1-report.md +95 -0
  343. package/evals/reports/by-skill/diagnose-failure-1-report.md +91 -0
  344. package/evals/reports/by-skill/diagnose-failure-2-report.md +94 -0
  345. package/evals/reports/by-skill/discover-1-report.md +118 -0
  346. package/evals/reports/by-skill/evaluate-skill-2-report.md +71 -0
  347. package/evals/reports/by-skill/execute-plan-1-report.md +103 -0
  348. package/evals/reports/by-skill/execute-plan-2-report.md +112 -0
  349. package/evals/reports/by-skill/execute-plan-3-report.md +87 -0
  350. package/evals/reports/by-skill/grill-context-1-report.md +103 -0
  351. package/evals/reports/by-skill/handoff-1-report.md +123 -0
  352. package/evals/reports/by-skill/handoff-2-report.md +105 -0
  353. package/evals/reports/by-skill/handoff-3-report.md +100 -0
  354. package/evals/reports/by-skill/handoff-7-report.md +77 -0
  355. package/evals/reports/by-skill/interview-gate-1-report.md +63 -0
  356. package/evals/reports/by-skill/interview-gate-2-report.md +105 -0
  357. package/evals/reports/by-skill/mode-contract-1-report.md +74 -0
  358. package/evals/reports/by-skill/mode-contract-2-report.md +87 -0
  359. package/evals/reports/by-skill/mode-contract-3-report.md +65 -0
  360. package/evals/reports/by-skill/mode-contract-4-report.md +71 -0
  361. package/evals/reports/by-skill/mode-contract-5-report.md +70 -0
  362. package/evals/reports/by-skill/output-router-1-report.md +76 -0
  363. package/evals/reports/by-skill/output-router-2-report.md +65 -0
  364. package/evals/reports/by-skill/research-1-report.md +144 -0
  365. package/evals/reports/by-skill/research-brief-1-report.md +94 -0
  366. package/evals/reports/by-skill/research-brief-2-report.md +75 -0
  367. package/evals/reports/by-skill/review-artifact-1-report.md +91 -0
  368. package/evals/reports/by-skill/review-artifact-2-report.md +79 -0
  369. package/evals/reports/by-skill/review-artifact-3-report.md +120 -0
  370. package/evals/reports/by-skill/review-artifact-4-report.md +89 -0
  371. package/evals/reports/by-skill/review-artifact-5-report.md +66 -0
  372. package/evals/reports/by-skill/review-work-1-report.md +85 -0
  373. package/evals/reports/by-skill/review-work-2-report.md +87 -0
  374. package/evals/reports/by-skill/review-work-3-report.md +104 -0
  375. package/evals/reports/by-skill/review-work-4-report.md +108 -0
  376. package/evals/reports/by-skill/review-work-5-report.md +73 -0
  377. package/evals/reports/by-skill/review-work-6-report.md +66 -0
  378. package/evals/reports/by-skill/setup-freeflow-1-report.md +199 -0
  379. package/evals/reports/by-skill/setup-freeflow-2-report.md +105 -0
  380. package/evals/reports/by-skill/setup-freeflow-3-report.md +105 -0
  381. package/evals/reports/by-skill/setup-freeflow-4-report.md +88 -0
  382. package/evals/reports/by-skill/setup-freeflow-5-report.md +115 -0
  383. package/evals/reports/by-skill/skill-authoring-1-report.md +81 -0
  384. package/evals/reports/by-skill/verify-work-1-report.md +111 -0
  385. package/evals/reports/by-skill/verify-work-2-report.md +120 -0
  386. package/evals/reports/by-skill/workflow-1-report.md +65 -0
  387. package/evals/reports/by-skill/write-plan-1-report.md +123 -0
  388. package/evals/reports/by-skill/write-plan-2-report.md +98 -0
  389. package/evals/reports/by-skill/write-plan-3-report.md +110 -0
  390. package/evals/reports/by-skill/write-spec-1-report.md +88 -0
  391. package/evals/reports/by-skill/write-spec-2-report.md +89 -0
  392. package/evals/reports/by-skill/write-spec-3-report.md +90 -0
  393. package/evals/reports/by-skill/write-spec-4-report.md +167 -0
  394. package/evals/reports/harness/claude-harness-1-report.md +103 -0
  395. package/evals/reports/harness/eval-harness-1-report.md +91 -0
  396. package/evals/reports/iterations/iteration-1-report.md +71 -0
  397. package/evals/reports/iterations/iteration-2-report.md +78 -0
  398. package/evals/reports/iterations/iteration-3-report.md +118 -0
  399. package/evals/reports/iterations/iteration-4-report.md +131 -0
  400. package/evals/reports/iterations/iteration-5-report.md +149 -0
  401. package/evals/reports/iterations/iteration-6-report.md +136 -0
  402. package/evals/reports/runtime/always-on-runtime-1-report.md +107 -0
  403. package/evals/reports/runtime/context-mode-normalized-benchmark-1-report.md +52 -0
  404. package/evals/reports/runtime/context-mode-real-deep-baseline-1-report.md +410 -0
  405. package/evals/reports/runtime/context-mode-real-deep-final-slice-11-report.md +203 -0
  406. package/evals/reports/runtime/context-mode-real-deep-final-slice-11-review.md +51 -0
  407. package/evals/reports/runtime/context-saving-native-tools-baseline-1-report.md +99 -0
  408. package/evals/reports/runtime/eryx-python-compatibility-reprobe-1-report.md +166 -0
  409. package/evals/reports/runtime/eryx-python-proof-spike-1-report.md +96 -0
  410. package/evals/reports/runtime/eryx-python-proof-spike-2-report.md +151 -0
  411. package/evals/reports/runtime/eryx-python-script-derive-execution-1-report.md +233 -0
  412. package/evals/reports/runtime/jq-script-derive-execution-1-report.md +215 -0
  413. package/evals/reports/runtime/jq-wasm-proof-spike-1-report.md +67 -0
  414. package/evals/reports/runtime/output-router-benchmark-1-report.md +81 -0
  415. package/evals/reports/runtime/output-router-capture-derive-provider-eval-1-report.md +35 -0
  416. package/evals/reports/runtime/output-router-codex-large-index-sqlite-adhoc-report.md +150 -0
  417. package/evals/reports/runtime/output-router-codex-qa-benchmark-1-report.md +46 -0
  418. package/evals/reports/runtime/output-router-command-benchmark-1-report.md +62 -0
  419. package/evals/reports/runtime/output-router-index-benchmark-1-report.md +64 -0
  420. package/evals/reports/runtime/output-router-regression-1-report.md +81 -0
  421. package/evals/reports/runtime/output-router-transform-eval-1-report.md +28 -0
  422. package/evals/reports/runtime/pi-observed-routing-eval-1-report.md +44 -0
  423. package/evals/reports/runtime/quickjs-script-derive-execution-1-report.md +159 -0
  424. package/evals/reports/runtime/quickjs-wasi-proof-spike-1-report.md +55 -0
  425. package/evals/reports/runtime/render-format-row-protocol-1-report.md +58 -0
  426. package/evals/reports/runtime/script-sandbox-adapter-selection-review-1-report.md +84 -0
  427. package/evals/reports/runtime/script-sandbox-probe-resource-hardening-1-report.md +51 -0
  428. package/evals/reports/runtime/storage-policy-benchmark-1-report.md +176 -0
  429. package/evals/reports/runtime/vault-index-storage-spike-1-report.md +39 -0
  430. package/evals/reports/runtime/workflow-context-hook-1-report.md +82 -0
  431. package/evals/runbooks/adversarial-eval-runbook.md +28 -0
  432. package/evals/runbooks/fixture-eval-runbook.md +137 -0
  433. package/evals/runbooks/handoff-eval-runbook.md +22 -0
  434. package/evals/runbooks/verify-work-eval-runbook.md +27 -0
  435. package/evals/scripts/audit-command-surface.sh +160 -0
  436. package/evals/scripts/check-activation-contract.sh +121 -0
  437. package/evals/scripts/check-runtime-context-hook.sh +183 -0
  438. package/evals/scripts/grade-fixture-eval.sh +446 -0
  439. package/evals/scripts/run-claude-fixture-eval.sh +26 -0
  440. package/evals/scripts/run-codex-fixture-eval.sh +26 -0
  441. package/evals/scripts/run-eryx-python-proof-spike.js +559 -0
  442. package/evals/scripts/run-fixture-eval-by-id.sh +23 -0
  443. package/evals/scripts/run-fixture-eval.sh +464 -0
  444. package/evals/scripts/run-jq-wasm-proof-spike.js +299 -0
  445. package/evals/scripts/run-output-router-transform-eval.js +115 -0
  446. package/evals/scripts/run-pi-observed-routing-eval.js +478 -0
  447. package/evals/scripts/run-quickjs-wasi-proof-spike.js +339 -0
  448. package/evals/scripts/skill-evidence.sh +316 -0
  449. package/evals/scripts/validate-release-metadata.sh +486 -0
  450. package/evals/suites/v0.1-acceptance-suite.md +131 -0
  451. package/{plugins/freeflow/hooks → hooks}/freeflow-runtime-context.mjs +65 -14
  452. package/package.json +46 -13
  453. package/pi-extension/dist/host-producer-identification.js +186 -0
  454. package/pi-extension/dist/index.js +85 -0
  455. package/pi-extension/dist/mcp-config.js +30 -0
  456. package/pi-extension/dist/native-safety-net.js +209 -0
  457. package/pi-extension/dist/observed-tool-routing.js +114 -0
  458. package/pi-extension/dist/renderers.js +835 -0
  459. package/pi-extension/dist/router-tools.js +438 -0
  460. package/pi-extension/dist/runtime-context.js +288 -0
  461. package/pi-extension/dist/schemas.js +254 -0
  462. package/pi-extension/dist/status.js +449 -0
  463. package/pi-extension/dist/utils.js +467 -0
  464. package/pi-extension/src/host-producer-identification.ts +226 -0
  465. package/pi-extension/src/index.ts +121 -0
  466. package/pi-extension/src/mcp-config.ts +32 -0
  467. package/pi-extension/src/native-safety-net.ts +252 -0
  468. package/pi-extension/src/observed-tool-routing.ts +126 -0
  469. package/pi-extension/src/renderers.ts +917 -0
  470. package/pi-extension/src/router-tools.ts +478 -0
  471. package/pi-extension/src/runtime-context.ts +347 -0
  472. package/pi-extension/src/schemas.ts +259 -0
  473. package/pi-extension/src/status.ts +505 -0
  474. package/pi-extension/src/utils.ts +492 -0
  475. package/pi-extension/tsconfig.json +16 -0
  476. package/plugin-docs/README.md +40 -0
  477. package/plugin-docs/adr/0002-evals-before-hooks.md +17 -0
  478. package/{plugins/freeflow/docs → plugin-docs}/adr/0003-release-boundary.md +2 -2
  479. package/plugin-docs/adr/0004-discover-replaces-shallow-discovery-skills.md +26 -0
  480. package/plugin-docs/adr/README.md +10 -0
  481. package/plugin-docs/architecture.md +89 -0
  482. package/plugin-docs/output-router.md +396 -0
  483. package/plugin-docs/release-evidence.md +77 -0
  484. package/{plugins/freeflow/docs → plugin-docs}/skills.md +7 -7
  485. package/{plugins/freeflow/docs → plugin-docs}/workflow.md +4 -4
  486. package/router/dist/benchmarks/benchmark-harness.d.ts +32 -0
  487. package/router/dist/benchmarks/benchmark-harness.js +92 -0
  488. package/router/dist/benchmarks/benchmarks.d.ts +102 -0
  489. package/router/dist/benchmarks/benchmarks.js +880 -0
  490. package/router/dist/benchmarks/codex-qa-benchmarks.d.ts +91 -0
  491. package/router/dist/benchmarks/codex-qa-benchmarks.js +461 -0
  492. package/router/dist/benchmarks/command-benchmarks.d.ts +128 -0
  493. package/router/dist/benchmarks/command-benchmarks.js +653 -0
  494. package/router/dist/benchmarks/context-mode-normalized-benchmarks.d.ts +74 -0
  495. package/router/dist/benchmarks/context-mode-normalized-benchmarks.js +707 -0
  496. package/router/dist/benchmarks/context-mode-real-deep-benchmark.d.ts +124 -0
  497. package/router/dist/benchmarks/context-mode-real-deep-benchmark.js +1295 -0
  498. package/router/dist/benchmarks/context-saving-baseline.d.ts +54 -0
  499. package/router/dist/benchmarks/context-saving-baseline.js +439 -0
  500. package/router/dist/benchmarks/index-benchmarks.d.ts +98 -0
  501. package/router/dist/benchmarks/index-benchmarks.js +762 -0
  502. package/router/dist/benchmarks/storage-policy-benchmarks.d.ts +90 -0
  503. package/router/dist/benchmarks/storage-policy-benchmarks.js +538 -0
  504. package/router/dist/benchmarks/vault-index-benchmarks.d.ts +47 -0
  505. package/router/dist/benchmarks/vault-index-benchmarks.js +257 -0
  506. package/router/dist/config/config.d.ts +96 -0
  507. package/router/dist/config/config.js +453 -0
  508. package/router/dist/config/router-contract.d.ts +10 -0
  509. package/router/dist/config/router-contract.js +54 -0
  510. package/router/dist/config/schema.d.ts +18 -0
  511. package/router/dist/config/schema.js +600 -0
  512. package/router/dist/config/types.d.ts +521 -0
  513. package/router/dist/config/types.js +30 -0
  514. package/router/dist/evidence/bounded-evidence.d.ts +40 -0
  515. package/router/dist/evidence/bounded-evidence.js +310 -0
  516. package/router/dist/evidence/evidence-range-selector.d.ts +22 -0
  517. package/router/dist/evidence/evidence-range-selector.js +231 -0
  518. package/router/dist/evidence/evidence-search.d.ts +25 -0
  519. package/router/dist/evidence/evidence-search.js +466 -0
  520. package/router/dist/evidence/evidence.d.ts +38 -0
  521. package/router/dist/evidence/evidence.js +196 -0
  522. package/router/dist/evidence/failure-contracts.d.ts +27 -0
  523. package/router/dist/evidence/failure-contracts.js +160 -0
  524. package/router/dist/evidence/line-ranges.d.ts +22 -0
  525. package/router/dist/evidence/line-ranges.js +22 -0
  526. package/router/dist/experiments/local-index.d.ts +58 -0
  527. package/router/dist/experiments/local-index.js +562 -0
  528. package/router/dist/index.d.ts +30 -0
  529. package/router/dist/index.js +32 -0
  530. package/router/dist/processing/engine.d.ts +108 -0
  531. package/router/dist/processing/engine.js +435 -0
  532. package/router/dist/processing/reducers.d.ts +266 -0
  533. package/router/dist/processing/reducers.js +1495 -0
  534. package/router/dist/processing/renderers.d.ts +25 -0
  535. package/router/dist/processing/renderers.js +186 -0
  536. package/router/dist/processing/scripts.d.ts +78 -0
  537. package/router/dist/processing/scripts.js +333 -0
  538. package/router/dist/repo/repo-traversal.d.ts +18 -0
  539. package/router/dist/repo/repo-traversal.js +242 -0
  540. package/router/dist/routing/observed-reducers.d.ts +35 -0
  541. package/router/dist/routing/observed-reducers.js +647 -0
  542. package/router/dist/routing/observed-routing.d.ts +17 -0
  543. package/router/dist/routing/observed-routing.js +377 -0
  544. package/router/dist/routing/parsers.d.ts +16 -0
  545. package/router/dist/routing/parsers.js +473 -0
  546. package/router/dist/routing/run-filters.d.ts +56 -0
  547. package/router/dist/routing/run-filters.js +230 -0
  548. package/router/dist/routing/run-reducers.d.ts +30 -0
  549. package/router/dist/routing/run-reducers.js +110 -0
  550. package/router/dist/sandbox/adapter-roots.d.ts +38 -0
  551. package/router/dist/sandbox/adapter-roots.js +64 -0
  552. package/router/dist/sandbox/eryx-python-adapter.d.ts +9 -0
  553. package/router/dist/sandbox/eryx-python-adapter.js +797 -0
  554. package/router/dist/sandbox/jq-wasm-adapter.d.ts +9 -0
  555. package/router/dist/sandbox/jq-wasm-adapter.js +421 -0
  556. package/router/dist/sandbox/quickjs-wasi-adapter.d.ts +9 -0
  557. package/router/dist/sandbox/quickjs-wasi-adapter.js +381 -0
  558. package/router/dist/sandbox/script-sandbox.d.ts +113 -0
  559. package/router/dist/sandbox/script-sandbox.js +303 -0
  560. package/router/dist/setup/script-transform-adapters.d.ts +28 -0
  561. package/router/dist/setup/script-transform-adapters.js +259 -0
  562. package/router/dist/tools/batch.d.ts +22 -0
  563. package/router/dist/tools/batch.js +509 -0
  564. package/router/dist/tools/run.d.ts +50 -0
  565. package/router/dist/tools/run.js +1219 -0
  566. package/router/dist/tools/search.d.ts +44 -0
  567. package/router/dist/tools/search.js +1205 -0
  568. package/router/dist/transform/engine.d.ts +143 -0
  569. package/router/dist/transform/engine.js +1955 -0
  570. package/router/dist/vault/vault-index.d.ts +106 -0
  571. package/router/dist/vault/vault-index.js +411 -0
  572. package/router/dist/vault/vault.d.ts +103 -0
  573. package/router/dist/vault/vault.js +635 -0
  574. package/{plugins/freeflow/skills → skills}/commit-work/SKILL.md +17 -16
  575. package/{plugins/freeflow/skills → skills}/commit-work/references/staging-decisions.md +3 -3
  576. package/skills/design-for-depth/SKILL.md +90 -0
  577. package/skills/design-for-depth/references/design-pressure-signals.md +265 -0
  578. package/skills/design-for-depth/references/software-design-philosophy.md +219 -0
  579. package/{plugins/freeflow/skills → skills}/diagnose-failure/SKILL.md +6 -1
  580. package/skills/discover/ARTIFACT-DESTINATIONS.md +58 -0
  581. package/skills/discover/CHECKPOINTS.md +74 -0
  582. package/skills/discover/SKILL.md +173 -0
  583. package/skills/execute-plan/SKILL.md +133 -0
  584. package/skills/execute-plan/references/execution-map.md +122 -0
  585. package/{plugins/freeflow/skills → skills}/mode-contract/SKILL.md +10 -5
  586. package/skills/output-router/SKILL.md +202 -0
  587. package/skills/output-router/references/safety-policy.md +40 -0
  588. package/{plugins/freeflow/skills → skills}/review-artifact/SKILL.md +37 -1
  589. package/{plugins/freeflow/skills → skills}/review-artifact/references/reviewer-prompt.md +14 -2
  590. package/skills/review-work/SKILL.md +101 -0
  591. package/{plugins/freeflow/skills → skills}/review-work/references/reviewer-prompt.md +8 -0
  592. package/skills/setup-freeflow/SKILL.md +106 -0
  593. package/{plugins/freeflow/skills → skills}/setup-freeflow/references/activation-contract.md +14 -8
  594. package/{plugins/freeflow/skills → skills}/setup-freeflow/references/host-setup.md +5 -3
  595. package/skills/setup-freeflow/references/output-router-setup.md +102 -0
  596. package/{plugins/freeflow/skills → skills}/verify-work/SKILL.md +3 -1
  597. package/{plugins/freeflow/skills → skills}/workflow/SKILL.md +23 -13
  598. package/{plugins/freeflow/skills → skills}/workflow/references/workflow-map.md +21 -20
  599. package/{plugins/freeflow/skills → skills}/write-plan/SKILL.md +9 -1
  600. package/{plugins/freeflow/skills → skills}/write-skill/SKILL.md +1 -1
  601. package/{plugins/freeflow/skills → skills}/write-spec/SKILL.md +9 -6
  602. package/plugins/freeflow/docs/README.md +0 -39
  603. package/plugins/freeflow/docs/adr/0002-evals-before-hooks.md +0 -17
  604. package/plugins/freeflow/docs/adr/README.md +0 -7
  605. package/plugins/freeflow/docs/architecture.md +0 -86
  606. package/plugins/freeflow/docs/release-evidence.md +0 -49
  607. package/plugins/freeflow/pi-extension/index.js +0 -250
  608. package/plugins/freeflow/skills/capture-decisions/SKILL.md +0 -95
  609. package/plugins/freeflow/skills/capture-decisions/references/destination-guide.md +0 -65
  610. package/plugins/freeflow/skills/execute-plan/SKILL.md +0 -67
  611. package/plugins/freeflow/skills/grill-context/SKILL.md +0 -79
  612. package/plugins/freeflow/skills/grill-context/references/approach-framing.md +0 -52
  613. package/plugins/freeflow/skills/research-brief/SKILL.md +0 -69
  614. package/plugins/freeflow/skills/review-work/SKILL.md +0 -73
  615. package/plugins/freeflow/skills/setup-freeflow/SKILL.md +0 -84
  616. /package/{plugins/freeflow/assets → assets}/freeflow-plugin-icon-transparent.png +0 -0
  617. /package/{plugins/freeflow/hooks → hooks}/hooks.json +0 -0
  618. /package/{plugins/freeflow/docs → plugin-docs}/adr/0001-three-modes.md +0 -0
  619. /package/{plugins/freeflow/skills → skills}/bypass/SKILL.md +0 -0
  620. /package/{plugins/freeflow/skills → skills}/diagnose-failure/references/feedback-loop-catalog.md +0 -0
  621. /package/{plugins/freeflow/skills → skills}/diagnose-failure/references/flaky-and-performance.md +0 -0
  622. /package/{plugins/freeflow/skills → skills}/evaluate-skill/SKILL.md +0 -0
  623. /package/{plugins/freeflow/skills → skills}/evaluate-skill/references/eval-patterns.md +0 -0
  624. /package/{plugins/freeflow/skills → skills}/evaluate-skill/references/grading-priority.md +0 -0
  625. /package/{plugins/freeflow/skills → skills}/handoff/SKILL.md +0 -0
  626. /package/{plugins/freeflow/skills → skills}/handoff/references/templates.md +0 -0
  627. /package/{plugins/freeflow/skills → skills}/interview-gate/SKILL.md +0 -0
  628. /package/{plugins/freeflow/skills → skills}/write-plan/references/plan-shapes.md +0 -0
  629. /package/{plugins/freeflow/skills → skills}/write-spec/references/artifact-standards.md +0 -0
  630. /package/{plugins/freeflow/skills → skills}/write-spec/references/spec-shapes.md +0 -0
@@ -0,0 +1,797 @@
1
+ import { spawn } from "node:child_process";
2
+ import { createHash } from "node:crypto";
3
+ import { access, cp, lstat, mkdir, mkdtemp, readdir, readFile, rm, writeFile } from "node:fs/promises";
4
+ import { tmpdir } from "node:os";
5
+ import { dirname, join, resolve } from "node:path";
6
+ import { fileURLToPath, pathToFileURL } from "node:url";
7
+ import { Worker } from "node:worker_threads";
8
+ import { defaultScriptTransformNodeBinary, resolveScriptTransformAdapterRoot } from "./adapter-roots.js";
9
+ import { SCRIPT_SANDBOX_REQUIRED_PROOFS, scriptSandboxProofFixturesForLanguage } from "./script-sandbox.js";
10
+ const DEFAULT_PROBE_TIMEOUT_MS = 250;
11
+ const DEFAULT_PROBE_OUTPUT_BYTES = 4096;
12
+ const DEFAULT_WORKER_OLD_GENERATION_MB = 256;
13
+ const DEFAULT_WORKER_YOUNG_GENERATION_MB = 32;
14
+ const SECRET_SENTINEL = "FREEFLOW_SANDBOX_SECRET_SENTINEL_VALUE";
15
+ const HOST_HOME = process.env.HOME ?? "__no_home__";
16
+ const ERYX_CHILD_MODE_ARG = "__freeflow_eryx_child__";
17
+ const ERYX_CHILD_TIMEOUT_FLOOR_MS = 30_000;
18
+ export const ERYX_ROOT_ENV = "FREEFLOW_ERYX_ROOT";
19
+ const PREVIEW2_IMPORTS = new Map([
20
+ ["@bytecodealliance/preview2-shim/cli", "../../@bytecodealliance/preview2-shim/lib/browser/cli.js"],
21
+ ["@bytecodealliance/preview2-shim/clocks", "../../@bytecodealliance/preview2-shim/lib/browser/clocks.js"],
22
+ ["@bytecodealliance/preview2-shim/filesystem", "../../@bytecodealliance/preview2-shim/lib/browser/filesystem.js"],
23
+ ["@bytecodealliance/preview2-shim/io", "../../@bytecodealliance/preview2-shim/lib/browser/io.js"],
24
+ ["@bytecodealliance/preview2-shim/random", "../../@bytecodealliance/preview2-shim/lib/browser/random.js"],
25
+ ]);
26
+ const eryxProbeCache = new Map();
27
+ export async function discoverEryxPythonSandboxAdaptersFromEnv(env = process.env) {
28
+ const candidate = await resolveScriptTransformAdapterRoot("python", env);
29
+ if (!candidate) {
30
+ return [];
31
+ }
32
+ try {
33
+ return [await createEryxPythonSandboxAdapter({ packageRoot: candidate.packageRoot })];
34
+ }
35
+ catch (error) {
36
+ return [unavailableEryxAdapter(`eryx-python adapter could not load from ${candidate.source === "env" ? candidate.envVar : "global adapter cache"}: ${errorMessage(error)}`)];
37
+ }
38
+ }
39
+ export async function createEryxPythonSandboxAdapter(options) {
40
+ const runtime = await loadEryxRuntime(options.packageRoot);
41
+ return {
42
+ id: options.id ?? "eryx-python",
43
+ version: options.version ?? runtime.packageVersion,
44
+ languages: ["python"],
45
+ async probe(language, config) {
46
+ if (language !== "python") {
47
+ return {
48
+ status: "unavailable",
49
+ reason: `eryx-python supports python, not ${language}.`,
50
+ passedProofs: [],
51
+ failedProofs: [...SCRIPT_SANDBOX_REQUIRED_PROOFS],
52
+ runtime: runtimeInfo(runtime),
53
+ };
54
+ }
55
+ return probeEryxRuntime(runtime, config);
56
+ },
57
+ async execute(request) {
58
+ if (request.language !== "python") {
59
+ return {
60
+ status: "policy_violation",
61
+ stdout: "",
62
+ stderr: "",
63
+ outputFiles: [],
64
+ exitCode: null,
65
+ reason: `eryx-python supports python, not ${request.language}.`,
66
+ };
67
+ }
68
+ if (request.network !== "off") {
69
+ return {
70
+ status: "policy_violation",
71
+ stdout: "",
72
+ stderr: "",
73
+ outputFiles: [],
74
+ exitCode: null,
75
+ reason: "eryx-python adapter only supports network=off.",
76
+ };
77
+ }
78
+ if (!runtime.jspiAvailable) {
79
+ return executeEryxInJspiChild(runtime, request);
80
+ }
81
+ const inputs = {};
82
+ for (const source of request.sources) {
83
+ inputs[source.alias] = await readFile(source.path, "utf8");
84
+ }
85
+ const run = await runEryx(runtime, {
86
+ code: pythonPrelude(inputs) + request.code,
87
+ timeoutMs: request.limits.timeoutMs,
88
+ outputBytes: request.limits.maxOutputBytes,
89
+ inputs,
90
+ });
91
+ return eryxRunToExecutionResult(run);
92
+ },
93
+ };
94
+ }
95
+ function unavailableEryxAdapter(reason) {
96
+ return {
97
+ id: "eryx-python",
98
+ version: "unavailable",
99
+ languages: ["python"],
100
+ async probe() {
101
+ return {
102
+ status: "unavailable",
103
+ reason,
104
+ passedProofs: [],
105
+ failedProofs: [...SCRIPT_SANDBOX_REQUIRED_PROOFS],
106
+ };
107
+ },
108
+ async execute() {
109
+ return {
110
+ status: "policy_violation",
111
+ stdout: "",
112
+ stderr: "",
113
+ outputFiles: [],
114
+ exitCode: null,
115
+ reason,
116
+ };
117
+ },
118
+ };
119
+ }
120
+ async function loadEryxRuntime(packageRoot) {
121
+ const resolvedRoot = resolve(packageRoot);
122
+ const packageJson = JSON.parse(await readFile(join(resolvedRoot, "package.json"), "utf8"));
123
+ const preview2ShimRoot = resolve(resolvedRoot, "..", "..", "@bytecodealliance", "preview2-shim");
124
+ const preview2ShimJson = JSON.parse(await readFile(join(preview2ShimRoot, "package.json"), "utf8"));
125
+ const runtimeHash = await hashEryxRuntime(resolvedRoot, preview2ShimRoot);
126
+ return {
127
+ packageName: String(packageJson.name ?? "@bsull/eryx"),
128
+ packageVersion: String(packageJson.version ?? "unknown"),
129
+ packageRoot: resolvedRoot,
130
+ preview2ShimRoot,
131
+ preview2ShimVersion: String(preview2ShimJson.version ?? "unknown"),
132
+ runtimeHash,
133
+ jspiAvailable: nodeJspiAvailable(),
134
+ };
135
+ }
136
+ async function hashEryxRuntime(packageRoot, preview2ShimRoot) {
137
+ const hash = createHash("sha256");
138
+ hash.update("eryx-runtime-tree-v1\0");
139
+ await hashRuntimeTree(hash, packageRoot, "eryx");
140
+ await hashRuntimeTree(hash, preview2ShimRoot, "preview2-shim");
141
+ return `sha256_${hash.digest("hex")}`;
142
+ }
143
+ async function hashRuntimeTree(hash, root, label, relativeDir = ".") {
144
+ const dir = join(root, relativeDir);
145
+ const entries = await readdir(dir, { withFileTypes: true });
146
+ entries.sort((left, right) => left.name.localeCompare(right.name));
147
+ for (const entry of entries) {
148
+ const relativePath = relativeDir === "." ? entry.name : join(relativeDir, entry.name);
149
+ const fullPath = join(root, relativePath);
150
+ const stats = await lstat(fullPath);
151
+ if (stats.isSymbolicLink()) {
152
+ throw new Error(`${label} runtime tree contains unsupported symlink: ${relativePath}`);
153
+ }
154
+ if (stats.isDirectory()) {
155
+ hash.update(`${label}:dir:${relativePath}\0`);
156
+ await hashRuntimeTree(hash, root, label, relativePath);
157
+ continue;
158
+ }
159
+ if (!stats.isFile()) {
160
+ throw new Error(`${label} runtime tree contains unsupported non-file entry: ${relativePath}`);
161
+ }
162
+ hash.update(`${label}:file:${relativePath}:${stats.size}\0`);
163
+ hash.update(await readFile(fullPath));
164
+ }
165
+ }
166
+ function nodeJspiAvailable() {
167
+ const nodeOptions = process.env.NODE_OPTIONS ?? "";
168
+ const flagPresent = process.execArgv.includes("--experimental-wasm-jspi") || nodeOptions.split(/\s+/).includes("--experimental-wasm-jspi");
169
+ return flagPresent && typeof WebAssembly.Suspending === "function";
170
+ }
171
+ function runtimeInfo(runtime) {
172
+ return {
173
+ name: runtime.packageName,
174
+ version: `${runtime.packageVersion}+preview2-shim-${runtime.preview2ShimVersion}`,
175
+ };
176
+ }
177
+ async function probeEryxRuntime(runtime, config) {
178
+ const timeoutMs = Math.min(config.limits.timeoutMs, DEFAULT_PROBE_TIMEOUT_MS);
179
+ const outputBytes = Math.min(config.limits.maxOutputBytes, DEFAULT_PROBE_OUTPUT_BYTES);
180
+ const currentRuntimeHash = await hashEryxRuntime(runtime.packageRoot, runtime.preview2ShimRoot);
181
+ if (currentRuntimeHash !== runtime.runtimeHash) {
182
+ return {
183
+ status: "unavailable",
184
+ reason: "eryx-python explicit package root changed after adapter creation; refusing to reuse stale proof cache.",
185
+ passedProofs: [],
186
+ failedProofs: [...SCRIPT_SANDBOX_REQUIRED_PROOFS],
187
+ runtime: runtimeInfo(runtime),
188
+ };
189
+ }
190
+ const cacheKey = [runtime.packageVersion, runtime.preview2ShimVersion, currentRuntimeHash, runtime.jspiAvailable, timeoutMs, outputBytes, config.network].join(":");
191
+ const cached = eryxProbeCache.get(cacheKey);
192
+ if (cached) {
193
+ return cloneProbeResult(await cached);
194
+ }
195
+ const probePromise = runEryxProbe(runtime, config, timeoutMs, outputBytes);
196
+ eryxProbeCache.set(cacheKey, probePromise);
197
+ try {
198
+ return cloneProbeResult(await probePromise);
199
+ }
200
+ catch (error) {
201
+ eryxProbeCache.delete(cacheKey);
202
+ throw error;
203
+ }
204
+ }
205
+ async function runEryxProbe(runtime, config, timeoutMs, outputBytes) {
206
+ if (config.network !== "off") {
207
+ return {
208
+ status: "unavailable",
209
+ reason: "eryx-python adapter only supports network=off.",
210
+ passedProofs: [],
211
+ failedProofs: [...SCRIPT_SANDBOX_REQUIRED_PROOFS],
212
+ runtime: runtimeInfo(runtime),
213
+ };
214
+ }
215
+ if (!runtime.jspiAvailable) {
216
+ return probeEryxInJspiChild(runtime, config);
217
+ }
218
+ const previousSecret = process.env.FREEFLOW_SANDBOX_SECRET_SENTINEL;
219
+ process.env.FREEFLOW_SANDBOX_SECRET_SENTINEL = SECRET_SENTINEL;
220
+ try {
221
+ const passedProofs = [];
222
+ const failedProofs = [];
223
+ const fixtures = scriptSandboxProofFixturesForLanguage("python");
224
+ for (const fixture of fixtures) {
225
+ const run = await runEryx(runtime, {
226
+ code: fixture.program,
227
+ timeoutMs,
228
+ outputBytes,
229
+ });
230
+ const assessment = assessEryxProof(fixture.proof, run, timeoutMs, outputBytes);
231
+ if (assessment) {
232
+ passedProofs.push(fixture.proof);
233
+ }
234
+ else {
235
+ failedProofs.push(fixture.proof);
236
+ }
237
+ }
238
+ const allPassed = SCRIPT_SANDBOX_REQUIRED_PROOFS.every((proof) => passedProofs.includes(proof));
239
+ return {
240
+ status: allPassed ? "available" : "unavailable",
241
+ reason: allPassed ? "eryx-python passed every required Python sandbox proof." : "eryx-python did not pass every required Python sandbox proof.",
242
+ passedProofs,
243
+ failedProofs,
244
+ runtime: runtimeInfo(runtime),
245
+ };
246
+ }
247
+ finally {
248
+ if (previousSecret === undefined) {
249
+ delete process.env.FREEFLOW_SANDBOX_SECRET_SENTINEL;
250
+ }
251
+ else {
252
+ process.env.FREEFLOW_SANDBOX_SECRET_SENTINEL = previousSecret;
253
+ }
254
+ }
255
+ }
256
+ async function probeEryxInJspiChild(runtime, config) {
257
+ try {
258
+ const result = await runEryxChildRequest({
259
+ kind: "probe",
260
+ packageRoot: runtime.packageRoot,
261
+ config,
262
+ }, childTimeoutMsForProbe(config));
263
+ if (isProbeResult(result)) {
264
+ return {
265
+ ...result,
266
+ reason: result.status === "available"
267
+ ? `${result.reason} via child Node --experimental-wasm-jspi.`
268
+ : result.reason,
269
+ };
270
+ }
271
+ return unavailableProbeResult("eryx-python child probe returned an execution result instead of a probe result.", runtime);
272
+ }
273
+ catch (error) {
274
+ return unavailableProbeResult(`eryx-python child JSPI probe failed: ${errorMessage(error)}`, runtime);
275
+ }
276
+ }
277
+ async function executeEryxInJspiChild(runtime, request) {
278
+ try {
279
+ const result = await runEryxChildRequest({
280
+ kind: "execute",
281
+ packageRoot: runtime.packageRoot,
282
+ request,
283
+ }, childTimeoutMsForExecution(request));
284
+ if (isExecutionResult(result)) {
285
+ return result;
286
+ }
287
+ return {
288
+ status: "failed",
289
+ stdout: "",
290
+ stderr: "eryx-python child execution returned a probe result instead of an execution result.",
291
+ outputFiles: [],
292
+ exitCode: null,
293
+ reason: "Invalid child execution response.",
294
+ };
295
+ }
296
+ catch (error) {
297
+ const message = errorMessage(error);
298
+ return {
299
+ status: "failed",
300
+ stdout: "",
301
+ stderr: message,
302
+ outputFiles: [],
303
+ exitCode: null,
304
+ reason: `eryx-python child JSPI execution failed: ${message}`,
305
+ };
306
+ }
307
+ }
308
+ async function runEryxChildRequest(request, timeoutMs) {
309
+ const root = await mkdtemp(join(tmpdir(), "freeflow-eryx-child-"));
310
+ const requestPath = join(root, "request.json");
311
+ const responsePath = join(root, "response.json");
312
+ try {
313
+ await writeFile(requestPath, JSON.stringify(request), "utf8");
314
+ await runEryxChildProcess(requestPath, responsePath, timeoutMs);
315
+ const response = JSON.parse(await readFile(responsePath, "utf8"));
316
+ if (!response.ok) {
317
+ throw new Error(response.message);
318
+ }
319
+ return response.result;
320
+ }
321
+ finally {
322
+ await rm(root, { recursive: true, force: true });
323
+ }
324
+ }
325
+ async function runEryxChildProcess(requestPath, responsePath, timeoutMs) {
326
+ const nodeBinary = await resolveEryxChildNodeBinary();
327
+ const child = spawn(nodeBinary, ["--experimental-wasm-jspi", fileURLToPath(import.meta.url), ERYX_CHILD_MODE_ARG, requestPath, responsePath], {
328
+ stdio: ["ignore", "ignore", "pipe"],
329
+ env: process.env,
330
+ });
331
+ let stderr = "";
332
+ child.stderr.setEncoding("utf8");
333
+ child.stderr.on("data", (chunk) => {
334
+ stderr += String(chunk);
335
+ });
336
+ await new Promise((resolvePromise, reject) => {
337
+ const timer = setTimeout(() => {
338
+ child.kill("SIGKILL");
339
+ reject(new Error(`eryx-python child process timed out after ${timeoutMs}ms.`));
340
+ }, timeoutMs);
341
+ child.on("error", (error) => {
342
+ clearTimeout(timer);
343
+ reject(error);
344
+ });
345
+ child.on("close", (code) => {
346
+ clearTimeout(timer);
347
+ if (code === 0) {
348
+ resolvePromise();
349
+ }
350
+ else {
351
+ reject(new Error(`eryx-python child process ${nodeBinary} exited with code ${code}. ${stderr.trim()}`));
352
+ }
353
+ });
354
+ });
355
+ }
356
+ async function resolveEryxChildNodeBinary() {
357
+ const configured = defaultScriptTransformNodeBinary();
358
+ if (await fileExists(configured)) {
359
+ return configured;
360
+ }
361
+ return process.execPath;
362
+ }
363
+ async function fileExists(path) {
364
+ try {
365
+ await access(path);
366
+ return true;
367
+ }
368
+ catch {
369
+ return false;
370
+ }
371
+ }
372
+ function childTimeoutMsForProbe(config) {
373
+ return Math.max(ERYX_CHILD_TIMEOUT_FLOOR_MS, Math.min(config.limits.timeoutMs, DEFAULT_PROBE_TIMEOUT_MS) * SCRIPT_SANDBOX_REQUIRED_PROOFS.length * 8);
374
+ }
375
+ function childTimeoutMsForExecution(request) {
376
+ return Math.max(ERYX_CHILD_TIMEOUT_FLOOR_MS, request.limits.timeoutMs + 5_000);
377
+ }
378
+ function unavailableProbeResult(reason, runtime) {
379
+ return {
380
+ status: "unavailable",
381
+ reason,
382
+ passedProofs: [],
383
+ failedProofs: [...SCRIPT_SANDBOX_REQUIRED_PROOFS],
384
+ runtime: runtimeInfo(runtime),
385
+ };
386
+ }
387
+ function isProbeResult(value) {
388
+ return Array.isArray(value.passedProofs) && Array.isArray(value.failedProofs);
389
+ }
390
+ function isExecutionResult(value) {
391
+ return Array.isArray(value.outputFiles);
392
+ }
393
+ async function runEryx(runtime, options) {
394
+ const start = Date.now();
395
+ const patched = await createPatchedPackageTree(runtime);
396
+ try {
397
+ const eryxEntryUrl = pathToFileURL(join(patched.eryxRoot, "index.js")).href;
398
+ return await runEryxWorker({ eryxEntryUrl, code: options.code, timeoutMs: options.timeoutMs, outputBytes: options.outputBytes, startedAt: start });
399
+ }
400
+ finally {
401
+ await rm(patched.root, { recursive: true, force: true });
402
+ }
403
+ }
404
+ async function createPatchedPackageTree(runtime) {
405
+ const root = await mkdtemp(join(tmpdir(), "freeflow-eryx-adapter-"));
406
+ const eryxRoot = join(root, "node_modules", "@bsull", "eryx");
407
+ const preview2ShimRoot = join(root, "node_modules", "@bytecodealliance", "preview2-shim");
408
+ try {
409
+ await mkdir(dirname(eryxRoot), { recursive: true });
410
+ await mkdir(dirname(preview2ShimRoot), { recursive: true });
411
+ await cp(runtime.packageRoot, eryxRoot, { recursive: true, dereference: false });
412
+ await cp(runtime.preview2ShimRoot, preview2ShimRoot, { recursive: true, dereference: false });
413
+ const copiedRuntimeHash = await hashEryxRuntime(eryxRoot, preview2ShimRoot);
414
+ if (copiedRuntimeHash !== runtime.runtimeHash) {
415
+ throw new Error("eryx-python explicit package root changed while preparing sandbox copy; refusing to execute unproven runtime files.");
416
+ }
417
+ for (const file of [join(eryxRoot, "index.js"), join(eryxRoot, "eryx-sandbox.js")]) {
418
+ let text = await readFile(file, "utf8");
419
+ for (const [from, to] of PREVIEW2_IMPORTS) {
420
+ text = text.split(from).join(to);
421
+ }
422
+ await writeFile(file, text, "utf8");
423
+ }
424
+ await writeInstrumentedNetworkDenyShim(eryxRoot);
425
+ return { root, eryxRoot };
426
+ }
427
+ catch (error) {
428
+ await rm(root, { recursive: true, force: true });
429
+ throw error;
430
+ }
431
+ }
432
+ async function writeInstrumentedNetworkDenyShim(eryxRoot) {
433
+ await writeFile(join(eryxRoot, "shims", "net.js"), `
434
+ const recordNetworkDeny = (kind, operation) => {
435
+ globalThis.__freeflowEryxNetworkEvents ??= [];
436
+ globalThis.__freeflowEryxNetworkEvents.push({ kind, operation, decision: "not-permitted" });
437
+ };
438
+
439
+ const notPermitted = { tag: "not-permitted", val: "networking is not available in the Freeflow Eryx adapter" };
440
+
441
+ export const tcp = {
442
+ connect(_host, _port) {
443
+ recordNetworkDeny("tcp", "connect");
444
+ return { tag: "err", val: notPermitted };
445
+ },
446
+ read(_handle, _len) {
447
+ recordNetworkDeny("tcp", "read");
448
+ return { tag: "err", val: { tag: "invalid-handle" } };
449
+ },
450
+ write(_handle, _data) {
451
+ recordNetworkDeny("tcp", "write");
452
+ return { tag: "err", val: { tag: "invalid-handle" } };
453
+ },
454
+ close(_handle) {
455
+ recordNetworkDeny("tcp", "close");
456
+ },
457
+ };
458
+
459
+ export const tls = {
460
+ upgrade(_tcp, _hostname) {
461
+ recordNetworkDeny("tls", "upgrade");
462
+ return { tag: "err", val: { tag: "tcp", val: notPermitted } };
463
+ },
464
+ read(_handle, _len) {
465
+ recordNetworkDeny("tls", "read");
466
+ return { tag: "err", val: { tag: "invalid-handle" } };
467
+ },
468
+ write(_handle, _data) {
469
+ recordNetworkDeny("tls", "write");
470
+ return { tag: "err", val: { tag: "invalid-handle" } };
471
+ },
472
+ close(_handle) {
473
+ recordNetworkDeny("tls", "close");
474
+ },
475
+ };
476
+ `, "utf8");
477
+ }
478
+ async function runEryxWorker(options) {
479
+ const workerSource = `
480
+ import { parentPort, workerData } from 'node:worker_threads';
481
+
482
+ let stdout = '';
483
+ let stderr = '';
484
+ let rawStdoutBytes = 0;
485
+ let rawStderrBytes = 0;
486
+ let truncated = false;
487
+
488
+ const byteLength = (value) => Buffer.byteLength(String(value ?? ''), 'utf8');
489
+ const stdoutCap = Math.max(1, Math.floor(workerData.outputBytes / 2));
490
+ const stderrCap = Math.max(1, workerData.outputBytes - stdoutCap);
491
+ const append = (stream, value) => {
492
+ const text = String(value ?? '');
493
+ const bytes = byteLength(text);
494
+ const cap = stream === 'stderr' ? stderrCap : stdoutCap;
495
+ const currentBytes = stream === 'stderr' ? byteLength(stderr) : byteLength(stdout);
496
+ if (stream === 'stderr') rawStderrBytes += bytes;
497
+ else rawStdoutBytes += bytes;
498
+ const remaining = cap - currentBytes;
499
+ if (remaining <= 0) {
500
+ truncated = truncated || bytes > 0;
501
+ return;
502
+ }
503
+ const buffer = Buffer.from(text, 'utf8');
504
+ const nextText = buffer.byteLength > remaining ? buffer.subarray(0, remaining).toString('utf8') : text;
505
+ if (buffer.byteLength > remaining) truncated = true;
506
+ if (stream === 'stderr') stderr += nextText;
507
+ else stdout += nextText;
508
+ };
509
+ const formatArgs = (args) => args.map((arg) => typeof arg === 'string' ? arg : JSON.stringify(arg)).join(' ');
510
+ globalThis.console = {
511
+ log: (...args) => append('stdout', formatArgs(args) + '\\n'),
512
+ info: (...args) => append('stdout', formatArgs(args) + '\\n'),
513
+ warn: (...args) => append('stderr', formatArgs(args) + '\\n'),
514
+ error: (...args) => append('stderr', formatArgs(args) + '\\n'),
515
+ debug: (...args) => append('stderr', formatArgs(args) + '\\n'),
516
+ };
517
+
518
+ const finish = (result) => {
519
+ parentPort.postMessage({
520
+ ...result,
521
+ stdout,
522
+ stderr,
523
+ outputFiles: [],
524
+ networkEvents: Array.isArray(globalThis.__freeflowEryxNetworkEvents) ? globalThis.__freeflowEryxNetworkEvents.slice(0, 20) : [],
525
+ rawStdoutBytes,
526
+ rawStderrBytes,
527
+ stdoutBytes: byteLength(stdout),
528
+ stderrBytes: byteLength(stderr),
529
+ outputBytes: byteLength(stdout) + byteLength(stderr),
530
+ truncated,
531
+ });
532
+ };
533
+
534
+ try {
535
+ const mod = await import(workerData.eryxEntryUrl);
536
+ const sandbox = new mod.Sandbox();
537
+ const result = await sandbox.execute(workerData.code);
538
+ append('stdout', result?.stdout ?? '');
539
+ append('stderr', result?.stderr ?? '');
540
+ finish({ status: 'success', exitCode: 0, result: result?.result ?? null, resultError: result?.resultError ?? null });
541
+ } catch (error) {
542
+ const message = String(error?.message ?? error);
543
+ append('stderr', message);
544
+ finish({ status: 'error', errorName: error?.name ?? 'Error', errorMessage: message, exitCode: 1 });
545
+ }
546
+ `;
547
+ const worker = new Worker(workerSource, {
548
+ eval: true,
549
+ type: "module",
550
+ workerData: { eryxEntryUrl: options.eryxEntryUrl, code: options.code, outputBytes: options.outputBytes },
551
+ stdout: true,
552
+ stderr: true,
553
+ resourceLimits: {
554
+ maxOldGenerationSizeMb: DEFAULT_WORKER_OLD_GENERATION_MB,
555
+ maxYoungGenerationSizeMb: DEFAULT_WORKER_YOUNG_GENERATION_MB,
556
+ },
557
+ });
558
+ let externalStdout = "";
559
+ let externalStderr = "";
560
+ let externalRawStdoutBytes = 0;
561
+ let externalRawStderrBytes = 0;
562
+ let externalTruncated = false;
563
+ const appendExternal = (stream, chunk) => {
564
+ const text = String(chunk ?? "");
565
+ const bytes = Buffer.byteLength(text, "utf8");
566
+ if (stream === "stderr")
567
+ externalRawStderrBytes += bytes;
568
+ else
569
+ externalRawStdoutBytes += bytes;
570
+ const currentBytes = Buffer.byteLength(externalStdout, "utf8") + Buffer.byteLength(externalStderr, "utf8");
571
+ const remaining = options.outputBytes - currentBytes;
572
+ if (remaining <= 0) {
573
+ externalTruncated = externalTruncated || bytes > 0;
574
+ return;
575
+ }
576
+ const buffer = Buffer.from(text, "utf8");
577
+ const nextText = buffer.byteLength > remaining ? buffer.subarray(0, remaining).toString("utf8") : text;
578
+ if (buffer.byteLength > remaining)
579
+ externalTruncated = true;
580
+ if (stream === "stderr")
581
+ externalStderr += nextText;
582
+ else
583
+ externalStdout += nextText;
584
+ };
585
+ worker.stdout?.on("data", (chunk) => appendExternal("stdout", chunk));
586
+ worker.stderr?.on("data", (chunk) => appendExternal("stderr", chunk));
587
+ return new Promise((resolve) => {
588
+ let settled = false;
589
+ const finish = (result) => {
590
+ if (settled)
591
+ return;
592
+ settled = true;
593
+ clearTimeout(timer);
594
+ const merged = mergeWorkerResult(result, {
595
+ externalStdout,
596
+ externalStderr,
597
+ externalRawStdoutBytes,
598
+ externalRawStderrBytes,
599
+ externalTruncated,
600
+ outputBytes: options.outputBytes,
601
+ });
602
+ resolve({ ...merged, durationMs: Date.now() - options.startedAt });
603
+ };
604
+ const timer = setTimeout(() => {
605
+ if (settled)
606
+ return;
607
+ finish({ status: "timed_out", stdout: "", stderr: "", exitCode: null, rawStdoutBytes: 0, rawStderrBytes: 0, stdoutBytes: 0, stderrBytes: 0, outputBytes: 0, truncated: false, outputFiles: [], networkEvents: [] });
608
+ void worker.terminate();
609
+ }, options.timeoutMs);
610
+ worker.on("message", (message) => finish(message));
611
+ worker.on("error", (error) => {
612
+ const message = errorMessage(error);
613
+ finish({ status: "error", stdout: "", stderr: message, exitCode: null, rawStdoutBytes: 0, rawStderrBytes: Buffer.byteLength(message, "utf8"), truncated: false, outputFiles: [], networkEvents: [], errorName: error.name, errorMessage: message });
614
+ });
615
+ worker.on("exit", (code) => {
616
+ if (!settled && code !== 0) {
617
+ finish({ status: "error", stdout: "", stderr: `worker exited with code ${code}`, exitCode: code, rawStdoutBytes: 0, rawStderrBytes: 0, truncated: false, outputFiles: [], networkEvents: [] });
618
+ }
619
+ });
620
+ });
621
+ }
622
+ function mergeWorkerResult(result, external) {
623
+ let stdout = "";
624
+ let stderr = "";
625
+ let truncated = Boolean(result.truncated || external.externalTruncated);
626
+ const append = (stream, value) => {
627
+ const currentBytes = Buffer.byteLength(stdout, "utf8") + Buffer.byteLength(stderr, "utf8");
628
+ const remaining = external.outputBytes - currentBytes;
629
+ if (remaining <= 0) {
630
+ truncated = truncated || Buffer.byteLength(value, "utf8") > 0;
631
+ return;
632
+ }
633
+ const buffer = Buffer.from(value, "utf8");
634
+ const nextText = buffer.byteLength > remaining ? buffer.subarray(0, remaining).toString("utf8") : value;
635
+ if (buffer.byteLength > remaining)
636
+ truncated = true;
637
+ if (stream === "stderr")
638
+ stderr += nextText;
639
+ else
640
+ stdout += nextText;
641
+ };
642
+ append("stdout", external.externalStdout);
643
+ append("stderr", external.externalStderr);
644
+ append("stdout", result.stdout ?? "");
645
+ append("stderr", result.stderr ?? "");
646
+ const rawStdoutBytes = (result.rawStdoutBytes ?? 0) + external.externalRawStdoutBytes;
647
+ const rawStderrBytes = (result.rawStderrBytes ?? 0) + external.externalRawStderrBytes;
648
+ const merged = {
649
+ status: result.status ?? "error",
650
+ stdout,
651
+ stderr,
652
+ rawStdoutBytes,
653
+ rawStderrBytes,
654
+ stdoutBytes: Buffer.byteLength(stdout, "utf8"),
655
+ stderrBytes: Buffer.byteLength(stderr, "utf8"),
656
+ outputBytes: Buffer.byteLength(stdout, "utf8") + Buffer.byteLength(stderr, "utf8"),
657
+ truncated,
658
+ exitCode: result.exitCode ?? null,
659
+ durationMs: result.durationMs ?? 0,
660
+ outputFiles: result.outputFiles ?? [],
661
+ networkEvents: result.networkEvents ?? [],
662
+ };
663
+ if (result.errorName !== undefined) {
664
+ merged.errorName = result.errorName;
665
+ }
666
+ if (result.errorMessage !== undefined) {
667
+ merged.errorMessage = result.errorMessage;
668
+ }
669
+ return merged;
670
+ }
671
+ function eryxRunToExecutionResult(run) {
672
+ if (run.status === "timed_out") {
673
+ return {
674
+ status: "timed_out",
675
+ stdout: run.stdout,
676
+ stderr: run.stderr,
677
+ outputFiles: [],
678
+ exitCode: null,
679
+ durationMs: run.durationMs,
680
+ reason: "Eryx Python execution exceeded timeoutMs.",
681
+ };
682
+ }
683
+ if (run.truncated || run.status === "output_limit_exceeded") {
684
+ return {
685
+ status: "failed",
686
+ stdout: run.stdout,
687
+ stderr: run.stderr,
688
+ outputFiles: [],
689
+ exitCode: run.exitCode,
690
+ durationMs: run.durationMs,
691
+ reason: "Eryx Python execution exceeded maxOutputBytes.",
692
+ };
693
+ }
694
+ if (run.status === "error") {
695
+ return {
696
+ status: "failed",
697
+ stdout: run.stdout,
698
+ stderr: run.stderr || run.errorMessage || "Eryx Python execution failed.",
699
+ outputFiles: [],
700
+ exitCode: run.exitCode,
701
+ durationMs: run.durationMs,
702
+ reason: "Eryx Python execution failed.",
703
+ };
704
+ }
705
+ return {
706
+ status: "success",
707
+ stdout: run.stdout,
708
+ stderr: run.stderr,
709
+ outputFiles: [],
710
+ exitCode: 0,
711
+ durationMs: run.durationMs,
712
+ };
713
+ }
714
+ function assessEryxProof(proof, run, timeoutMs, outputBytes) {
715
+ const combined = `${run.stdout ?? ""}\n${run.stderr ?? ""}\n${run.errorMessage ?? ""}`;
716
+ switch (proof) {
717
+ case "env_access_denied":
718
+ return !combined.includes(SECRET_SENTINEL) && !combined.includes(HOST_HOME) && !combined.includes("process.env") && !/AWS_|GITHUB_|TOKEN|SECRET/i.test(combined);
719
+ case "home_access_denied":
720
+ return !combined.includes(HOST_HOME) && !/PRIVATE KEY|FREEFLOW_SANDBOX_SECRET/.test(combined);
721
+ case "repo_access_denied":
722
+ return !combined.includes("@hassangameryt/freeflow") && !combined.includes("@earendil-works/freeflow") && !combined.includes('"scripts"');
723
+ case "vault_access_denied":
724
+ return !combined.includes("ffout_") && !combined.includes("ffrec_");
725
+ case "network_access_denied":
726
+ return run.status !== "success" && networkDeniedByShim(run) && !combined.includes("Example Domain");
727
+ case "input_read_only":
728
+ return run.status !== "success" && !combined.includes("mutated");
729
+ case "output_escape_denied":
730
+ return !combined.includes("/etc/passwd") && run.outputFiles.length === 0;
731
+ case "stdout_stderr_bounded":
732
+ return run.status !== "error" && run.truncated && run.outputBytes <= outputBytes && run.rawStdoutBytes > outputBytes && run.rawStderrBytes > outputBytes;
733
+ case "timeout_enforced":
734
+ return run.status === "timed_out" && run.durationMs < timeoutMs * 5;
735
+ default:
736
+ return false;
737
+ }
738
+ }
739
+ function networkDeniedByShim(run) {
740
+ return run.networkEvents.some((event) => event?.decision === "not-permitted" && (event?.kind === "tcp" || event?.kind === "tls"));
741
+ }
742
+ function pythonPrelude(inputs) {
743
+ const encodedInputs = JSON.stringify(inputs).replaceAll("</", "<\\/");
744
+ return [
745
+ "# Freeflow injected input/output helpers.",
746
+ "import json as __freeflow_json",
747
+ `__freeflow_sources = __freeflow_json.loads(${JSON.stringify(encodedInputs)})`,
748
+ "sources = dict(__freeflow_sources)",
749
+ "def read_text(alias):",
750
+ " return __freeflow_sources[alias]",
751
+ "def write_text(text):",
752
+ " print(str(text), end='')",
753
+ "",
754
+ ].join("\n");
755
+ }
756
+ function cloneProbeResult(result) {
757
+ const cloned = {
758
+ ...result,
759
+ passedProofs: [...result.passedProofs],
760
+ failedProofs: [...result.failedProofs],
761
+ };
762
+ if (result.runtime !== undefined) {
763
+ cloned.runtime = { ...result.runtime };
764
+ }
765
+ return cloned;
766
+ }
767
+ async function runEryxChildCli(argv) {
768
+ const requestPath = argv[0];
769
+ const responsePath = argv[1];
770
+ if (!requestPath || !responsePath) {
771
+ throw new Error("eryx-python child mode requires request and response paths.");
772
+ }
773
+ const payload = JSON.parse(await readFile(requestPath, "utf8"));
774
+ let response;
775
+ try {
776
+ const adapter = await createEryxPythonSandboxAdapter({ packageRoot: payload.packageRoot });
777
+ if (payload.kind === "probe") {
778
+ response = { ok: true, result: await adapter.probe("python", payload.config) };
779
+ }
780
+ else {
781
+ response = { ok: true, result: await adapter.execute(payload.request) };
782
+ }
783
+ }
784
+ catch (error) {
785
+ response = { ok: false, message: errorMessage(error) };
786
+ }
787
+ await writeFile(responsePath, JSON.stringify(response), "utf8");
788
+ }
789
+ function errorMessage(error) {
790
+ return error instanceof Error ? error.message : String(error);
791
+ }
792
+ if (process.argv[2] === ERYX_CHILD_MODE_ARG) {
793
+ runEryxChildCli(process.argv.slice(3)).catch((error) => {
794
+ console.error(errorMessage(error));
795
+ process.exit(1);
796
+ });
797
+ }