@oscharko-dev/keiko 1.1.6 → 1.1.8

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 (300) hide show
  1. package/dist/ui/csp-hashes.json +7 -7
  2. package/dist/ui/static/404.html +1 -1
  3. package/dist/ui/static/__next.__PAGE__.txt +11 -11
  4. package/dist/ui/static/__next._full.txt +13 -13
  5. package/dist/ui/static/__next._tree.txt +2 -2
  6. package/dist/ui/static/_next/static/chunks/02i0eppipph7e.js +1 -0
  7. package/dist/ui/static/_next/static/chunks/09jwu8wa22c1d.js +1 -0
  8. package/dist/ui/static/_next/static/chunks/0hc828_gdkyek.js +3 -0
  9. package/dist/ui/static/_next/static/chunks/0n5hz6kjgs-3-.js +1 -0
  10. package/dist/ui/static/_next/static/chunks/{2l0gyhe4jpu23.css → 0qv7j1mski9o8.css} +1 -1
  11. package/dist/ui/static/_next/static/chunks/{0xq716cs1taxh.js → 0vtnpcijh-jqw.js} +1 -1
  12. package/dist/ui/static/_next/static/chunks/15wj80h_5chl0.js +1 -0
  13. package/dist/ui/static/_next/static/chunks/{0ak5o5brryicl.js → 18grpx4tuavds.js} +1 -1
  14. package/dist/ui/static/_next/static/chunks/1be0-x1mk56wu.js +1 -0
  15. package/dist/ui/static/_next/static/chunks/{0bbv42deftomm.js → 1c1gbfapof8m0.js} +1 -1
  16. package/dist/ui/static/_next/static/chunks/1dwc7p2rl-hs6.js +1 -0
  17. package/dist/ui/static/_next/static/chunks/{1piodmepoegnw.js → 1fgcho9v65crw.js} +1 -1
  18. package/dist/ui/static/_next/static/chunks/{3d-10944zk2o9.js → 1lj5se4u-t-e9.js} +1 -1
  19. package/dist/ui/static/_next/static/chunks/{1th8j9weqwhgx.js → 1o2cihsnxhmh6.js} +1 -1
  20. package/dist/ui/static/_next/static/chunks/22x3pxeia7hk_.js +1 -0
  21. package/dist/ui/static/_next/static/chunks/{0e6em7dg4gr6q.js → 25lgzefdnbdh6.js} +1 -1
  22. package/dist/ui/static/_next/static/chunks/{2t2wftn5liinr.js → 2iz4lys7fa5a6.js} +1 -1
  23. package/dist/ui/static/_next/static/chunks/{21t1klb1thn76.js → 306ima8lxo17-.js} +1 -1
  24. package/dist/ui/static/_next/static/chunks/{2dpxapmdvqlh1.js → 36-tv3y56oau8.js} +1 -1
  25. package/dist/ui/static/_next/static/chunks/{182b4sf3i-krz.js → 39w0xj7fyy7sp.js} +1 -1
  26. package/dist/ui/static/_next/static/chunks/{0ks_l2sa4nfn0.js → 3aizxizjzdddj.js} +1 -1
  27. package/dist/ui/static/_next/static/chunks/3bl12pe7lsanc.js +1 -0
  28. package/dist/ui/static/_next/static/chunks/3k6g7nzn589fs.js +1 -0
  29. package/dist/ui/static/_next/static/chunks/{1j_9b-l0n6u-t.js → 3kmozkq6tgqnz.js} +1 -1
  30. package/dist/ui/static/_next/static/chunks/3q_kn20qriqd7.js +1 -0
  31. package/dist/ui/static/_next/static/chunks/3r4fz572q3l7s.js +1 -0
  32. package/dist/ui/static/_next/static/chunks/{1lb8ju8qa0d9m.js → 40g2_9ovqiuiu.js} +1 -1
  33. package/dist/ui/static/_next/static/chunks/413lmui4la1xe.js +1 -0
  34. package/dist/ui/static/_next/static/chunks/{turbopack-2w1ixcff5kx3x.js → turbopack-2fb3mn6go2p-t.js} +1 -1
  35. package/dist/ui/static/_not-found/__next._full.txt +9 -9
  36. package/dist/ui/static/_not-found/__next._not-found.__PAGE__.txt +8 -8
  37. package/dist/ui/static/_not-found/__next._tree.txt +1 -1
  38. package/dist/ui/static/_not-found.html +1 -1
  39. package/dist/ui/static/_not-found.txt +9 -9
  40. package/dist/ui/static/atlassian-connectors/__next._full.txt +13 -13
  41. package/dist/ui/static/atlassian-connectors/__next._tree.txt +1 -1
  42. package/dist/ui/static/atlassian-connectors/__next.atlassian-connectors.__PAGE__.txt +10 -10
  43. package/dist/ui/static/atlassian-connectors.html +1 -1
  44. package/dist/ui/static/atlassian-connectors.txt +13 -13
  45. package/dist/ui/static/index.html +1 -1
  46. package/dist/ui/static/index.txt +13 -13
  47. package/dist/ui/static/launch/__next._full.txt +13 -13
  48. package/dist/ui/static/launch/__next._tree.txt +2 -2
  49. package/dist/ui/static/launch/__next.launch.__PAGE__.txt +11 -11
  50. package/dist/ui/static/launch.html +1 -1
  51. package/dist/ui/static/launch.txt +13 -13
  52. package/dist/ui/static/local-knowledge/__next._full.txt +13 -13
  53. package/dist/ui/static/local-knowledge/__next._tree.txt +2 -2
  54. package/dist/ui/static/local-knowledge/__next.local-knowledge.__PAGE__.txt +11 -11
  55. package/dist/ui/static/local-knowledge/capsule/__next._full.txt +13 -13
  56. package/dist/ui/static/local-knowledge/capsule/__next._tree.txt +1 -1
  57. package/dist/ui/static/local-knowledge/capsule/__next.local-knowledge.capsule.__PAGE__.txt +10 -10
  58. package/dist/ui/static/local-knowledge/capsule.html +1 -1
  59. package/dist/ui/static/local-knowledge/capsule.txt +13 -13
  60. package/dist/ui/static/local-knowledge.html +1 -1
  61. package/dist/ui/static/local-knowledge.txt +13 -13
  62. package/node_modules/@oscharko-dev/keiko-cli/package.json +14 -14
  63. package/node_modules/@oscharko-dev/keiko-connectors/dist/version.d.ts +1 -1
  64. package/node_modules/@oscharko-dev/keiko-connectors/dist/version.js +1 -1
  65. package/node_modules/@oscharko-dev/keiko-connectors/package.json +3 -3
  66. package/node_modules/@oscharko-dev/keiko-contracts/dist/activity-log-failure-class-contracts.d.ts +55 -15
  67. package/node_modules/@oscharko-dev/keiko-contracts/dist/activity-log-failure-class-contracts.d.ts.map +1 -1
  68. package/node_modules/@oscharko-dev/keiko-contracts/dist/activity-log-failure-class-contracts.js +108 -12
  69. package/node_modules/@oscharko-dev/keiko-contracts/dist/activity-log-registry.generated.d.ts +924 -33
  70. package/node_modules/@oscharko-dev/keiko-contracts/dist/activity-log-registry.generated.d.ts.map +1 -1
  71. package/node_modules/@oscharko-dev/keiko-contracts/dist/activity-log-registry.generated.js +1249 -142
  72. package/node_modules/@oscharko-dev/keiko-contracts/dist/bff-wire.d.ts +7 -2
  73. package/node_modules/@oscharko-dev/keiko-contracts/dist/bff-wire.d.ts.map +1 -1
  74. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench-runtime-api-validation.d.ts.map +1 -1
  75. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench-runtime-api-validation.js +14 -4
  76. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench-runtime-constants.d.ts +1 -1
  77. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench-runtime-constants.d.ts.map +1 -1
  78. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench-runtime.d.ts +7 -0
  79. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench-runtime.d.ts.map +1 -1
  80. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench.d.ts +1 -1
  81. package/node_modules/@oscharko-dev/keiko-contracts/dist/coding-workbench.d.ts.map +1 -1
  82. package/node_modules/@oscharko-dev/keiko-contracts/dist/context-engineering.d.ts +3 -0
  83. package/node_modules/@oscharko-dev/keiko-contracts/dist/context-engineering.d.ts.map +1 -1
  84. package/node_modules/@oscharko-dev/keiko-contracts/dist/context-engineering.js +20 -4
  85. package/node_modules/@oscharko-dev/keiko-contracts/dist/index.d.ts +1 -1
  86. package/node_modules/@oscharko-dev/keiko-contracts/dist/index.d.ts.map +1 -1
  87. package/node_modules/@oscharko-dev/keiko-contracts/dist/verified-commit.d.ts +1 -1
  88. package/node_modules/@oscharko-dev/keiko-contracts/dist/verified-commit.d.ts.map +1 -1
  89. package/node_modules/@oscharko-dev/keiko-contracts/dist/verified-commit.js +8 -1
  90. package/node_modules/@oscharko-dev/keiko-contracts/dist/version.d.ts +2 -2
  91. package/node_modules/@oscharko-dev/keiko-contracts/dist/version.js +2 -2
  92. package/node_modules/@oscharko-dev/keiko-contracts/package.json +1 -1
  93. package/node_modules/@oscharko-dev/keiko-evaluations/package.json +9 -9
  94. package/node_modules/@oscharko-dev/keiko-evidence/dist/version.d.ts +1 -1
  95. package/node_modules/@oscharko-dev/keiko-evidence/dist/version.js +1 -1
  96. package/node_modules/@oscharko-dev/keiko-evidence/package.json +4 -4
  97. package/node_modules/@oscharko-dev/keiko-git/dist/version.d.ts +1 -1
  98. package/node_modules/@oscharko-dev/keiko-git/dist/version.js +1 -1
  99. package/node_modules/@oscharko-dev/keiko-git/package.json +1 -1
  100. package/node_modules/@oscharko-dev/keiko-harness/dist/version.d.ts +1 -1
  101. package/node_modules/@oscharko-dev/keiko-harness/dist/version.js +1 -1
  102. package/node_modules/@oscharko-dev/keiko-harness/package.json +7 -7
  103. package/node_modules/@oscharko-dev/keiko-local-knowledge/dist/version.d.ts +1 -1
  104. package/node_modules/@oscharko-dev/keiko-local-knowledge/dist/version.js +1 -1
  105. package/node_modules/@oscharko-dev/keiko-local-knowledge/package.json +5 -5
  106. package/node_modules/@oscharko-dev/keiko-memory-capture/dist/version.d.ts +1 -1
  107. package/node_modules/@oscharko-dev/keiko-memory-capture/dist/version.js +1 -1
  108. package/node_modules/@oscharko-dev/keiko-memory-capture/package.json +3 -3
  109. package/node_modules/@oscharko-dev/keiko-memory-consolidation/dist/version.d.ts +1 -1
  110. package/node_modules/@oscharko-dev/keiko-memory-consolidation/dist/version.js +1 -1
  111. package/node_modules/@oscharko-dev/keiko-memory-consolidation/package.json +3 -3
  112. package/node_modules/@oscharko-dev/keiko-memory-governance/dist/version.d.ts +1 -1
  113. package/node_modules/@oscharko-dev/keiko-memory-governance/dist/version.js +1 -1
  114. package/node_modules/@oscharko-dev/keiko-memory-governance/package.json +3 -3
  115. package/node_modules/@oscharko-dev/keiko-memory-retrieval/dist/version.d.ts +1 -1
  116. package/node_modules/@oscharko-dev/keiko-memory-retrieval/dist/version.js +1 -1
  117. package/node_modules/@oscharko-dev/keiko-memory-retrieval/package.json +3 -3
  118. package/node_modules/@oscharko-dev/keiko-memory-vault/dist/version.d.ts +1 -1
  119. package/node_modules/@oscharko-dev/keiko-memory-vault/dist/version.js +1 -1
  120. package/node_modules/@oscharko-dev/keiko-memory-vault/package.json +3 -3
  121. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/config.d.ts.map +1 -1
  122. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/config.js +4 -1
  123. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/gateway.d.ts.map +1 -1
  124. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/gateway.js +136 -24
  125. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/index.d.ts +1 -0
  126. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/index.d.ts.map +1 -1
  127. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/index.js +1 -0
  128. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/openai-adapter.d.ts.map +1 -1
  129. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/openai-adapter.js +38 -4
  130. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/openai-embedding-adapter.d.ts.map +1 -1
  131. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/openai-embedding-adapter.js +12 -3
  132. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/realtime-voice-adapter.d.ts.map +1 -1
  133. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/realtime-voice-adapter.js +3 -1
  134. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/rerank-adapter.d.ts.map +1 -1
  135. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/rerank-adapter.js +4 -1
  136. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/resilience.d.ts +9 -1
  137. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/resilience.d.ts.map +1 -1
  138. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/resilience.js +87 -5
  139. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/speech-to-text-adapter.d.ts +1 -0
  140. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/speech-to-text-adapter.d.ts.map +1 -1
  141. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/speech-to-text-adapter.js +158 -10
  142. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/text-to-speech-adapter.d.ts.map +1 -1
  143. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/text-to-speech-adapter.js +176 -12
  144. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/version.d.ts +1 -1
  145. package/node_modules/@oscharko-dev/keiko-model-gateway/dist/version.js +1 -1
  146. package/node_modules/@oscharko-dev/keiko-model-gateway/package.json +4 -4
  147. package/node_modules/@oscharko-dev/keiko-quality-intelligence/package.json +3 -3
  148. package/node_modules/@oscharko-dev/keiko-sandbox/package.json +2 -2
  149. package/node_modules/@oscharko-dev/keiko-sdk/package.json +9 -9
  150. package/node_modules/@oscharko-dev/keiko-security/dist/errors/gateway.d.ts +11 -1
  151. package/node_modules/@oscharko-dev/keiko-security/dist/errors/gateway.d.ts.map +1 -1
  152. package/node_modules/@oscharko-dev/keiko-security/dist/errors/gateway.js +33 -0
  153. package/node_modules/@oscharko-dev/keiko-security/dist/version.d.ts +1 -1
  154. package/node_modules/@oscharko-dev/keiko-security/dist/version.js +1 -1
  155. package/node_modules/@oscharko-dev/keiko-security/package.json +2 -2
  156. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeControlPlane.d.ts +3 -0
  157. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeControlPlane.d.ts.map +1 -1
  158. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeHistory.d.ts +0 -1
  159. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeHistory.d.ts.map +1 -1
  160. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeHistory.js +44 -13
  161. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeManager.d.ts +1 -1
  162. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeManager.d.ts.map +1 -1
  163. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeManager.js +9 -4
  164. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeOrchestrator.d.ts.map +1 -1
  165. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeOrchestrator.js +3 -3
  166. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeOrchestratorTypes.d.ts +2 -1
  167. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeOrchestratorTypes.d.ts.map +1 -1
  168. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeRoutes.d.ts.map +1 -1
  169. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingRuntimeRoutes.js +3 -2
  170. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingSafeActivityProjection.d.ts.map +1 -1
  171. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingSafeActivityProjection.js +143 -32
  172. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolFacade.d.ts +7 -0
  173. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolFacade.d.ts.map +1 -1
  174. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolFacade.js +29 -8
  175. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolFacadePorts.d.ts +16 -0
  176. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolFacadePorts.d.ts.map +1 -1
  177. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolIpc.d.ts +6 -2
  178. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolIpc.d.ts.map +1 -1
  179. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolIpc.js +5 -2
  180. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolReadEditPorts.d.ts +14 -0
  181. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolReadEditPorts.d.ts.map +1 -1
  182. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/codingToolReadEditPorts.js +201 -43
  183. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/launchFailure.d.ts +13 -1
  184. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/launchFailure.d.ts.map +1 -1
  185. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/launchFailure.js +24 -0
  186. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeFunctionalHarness/_governedTools.d.ts +12 -0
  187. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeFunctionalHarness/_governedTools.d.ts.map +1 -1
  188. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeFunctionalHarness/_governedTools.js +31 -0
  189. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeRuntimeAdapter.js +35 -13
  190. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeRuntimeComposition.d.ts +16 -4
  191. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeRuntimeComposition.d.ts.map +1 -1
  192. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeRuntimeComposition.js +74 -14
  193. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeV2ApprovalRequests.d.ts +33 -2
  194. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeV2ApprovalRequests.d.ts.map +1 -1
  195. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/opencodeV2ApprovalRequests.js +277 -44
  196. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionAuxiliaryPorts.d.ts +6 -0
  197. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionAuxiliaryPorts.d.ts.map +1 -1
  198. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionAuxiliaryPorts.js +4 -2
  199. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionDraftDeliveryRuntime.d.ts.map +1 -1
  200. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionDraftDeliveryRuntime.js +3 -0
  201. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionManagedWorktreeTools.d.ts.map +1 -1
  202. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionManagedWorktreeTools.js +72 -26
  203. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionVerifiedCommitRuntime.d.ts.map +1 -1
  204. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/productionVerifiedCommitRuntime.js +6 -0
  205. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/researchApprovalIssuance.d.ts.map +1 -1
  206. package/node_modules/@oscharko-dev/keiko-server/dist/coding-runtime/researchApprovalIssuance.js +2 -0
  207. package/node_modules/@oscharko-dev/keiko-server/dist/coding-sidecar-gateway.d.ts +30 -0
  208. package/node_modules/@oscharko-dev/keiko-server/dist/coding-sidecar-gateway.d.ts.map +1 -1
  209. package/node_modules/@oscharko-dev/keiko-server/dist/coding-sidecar-gateway.js +476 -143
  210. package/node_modules/@oscharko-dev/keiko-server/dist/coding-sidecar-tool-facade.d.ts.map +1 -1
  211. package/node_modules/@oscharko-dev/keiko-server/dist/coding-sidecar-tool-facade.js +32 -5
  212. package/node_modules/@oscharko-dev/keiko-server/dist/deps.d.ts +16 -0
  213. package/node_modules/@oscharko-dev/keiko-server/dist/deps.d.ts.map +1 -1
  214. package/node_modules/@oscharko-dev/keiko-server/dist/deps.js +2 -23
  215. package/node_modules/@oscharko-dev/keiko-server/dist/diagnostics-log.d.ts +5 -1
  216. package/node_modules/@oscharko-dev/keiko-server/dist/diagnostics-log.d.ts.map +1 -1
  217. package/node_modules/@oscharko-dev/keiko-server/dist/diagnostics-log.js +11 -0
  218. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-readiness.d.ts +24 -2
  219. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-readiness.d.ts.map +1 -1
  220. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-readiness.js +215 -38
  221. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-route-deadline.d.ts +6 -0
  222. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-route-deadline.d.ts.map +1 -0
  223. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-route-deadline.js +27 -0
  224. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-setup.d.ts +33 -0
  225. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-setup.d.ts.map +1 -1
  226. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-setup.js +256 -25
  227. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-tool-calling-probe.d.ts +2 -1
  228. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-tool-calling-probe.d.ts.map +1 -1
  229. package/node_modules/@oscharko-dev/keiko-server/dist/gateway-tool-calling-probe.js +11 -4
  230. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/commitRoutes.d.ts +2 -1
  231. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/commitRoutes.d.ts.map +1 -1
  232. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/commitRoutes.js +221 -38
  233. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/verifiedCommitService.d.ts.map +1 -1
  234. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/verifiedCommitService.js +29 -4
  235. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/verifiedCommitTypes.d.ts +2 -1
  236. package/node_modules/@oscharko-dev/keiko-server/dist/gitDelivery/verifiedCommitTypes.d.ts.map +1 -1
  237. package/node_modules/@oscharko-dev/keiko-server/dist/observability/log-redaction.d.ts.map +1 -1
  238. package/node_modules/@oscharko-dev/keiko-server/dist/observability/log-redaction.js +1 -0
  239. package/node_modules/@oscharko-dev/keiko-server/dist/store/codingHistory.d.ts.map +1 -1
  240. package/node_modules/@oscharko-dev/keiko-server/dist/store/codingHistory.js +48 -3
  241. package/node_modules/@oscharko-dev/keiko-server/dist/tool-catalog/catalogToolFacadeBridge.d.ts.map +1 -1
  242. package/node_modules/@oscharko-dev/keiko-server/dist/tool-catalog/catalogToolFacadeBridge.js +54 -8
  243. package/node_modules/@oscharko-dev/keiko-server/package.json +22 -22
  244. package/node_modules/@oscharko-dev/keiko-tool-catalog/package.json +3 -3
  245. package/node_modules/@oscharko-dev/keiko-tools/dist/version.d.ts +1 -1
  246. package/node_modules/@oscharko-dev/keiko-tools/dist/version.js +1 -1
  247. package/node_modules/@oscharko-dev/keiko-tools/package.json +7 -7
  248. package/node_modules/@oscharko-dev/keiko-verification/package.json +5 -5
  249. package/node_modules/@oscharko-dev/keiko-workflows/package.json +10 -10
  250. package/node_modules/@oscharko-dev/keiko-workspace/dist/version.d.ts +1 -1
  251. package/node_modules/@oscharko-dev/keiko-workspace/dist/version.js +1 -1
  252. package/node_modules/@oscharko-dev/keiko-workspace/package.json +3 -3
  253. package/package.json +29 -29
  254. package/release-impact.catalog.json +135 -23
  255. package/vendor/{oscharko-dev-keiko-cli-1.1.6.tgz → oscharko-dev-keiko-cli-1.1.8.tgz} +0 -0
  256. package/vendor/{oscharko-dev-keiko-connectors-1.1.6.tgz → oscharko-dev-keiko-connectors-1.1.8.tgz} +0 -0
  257. package/vendor/oscharko-dev-keiko-contracts-1.1.8.tgz +0 -0
  258. package/vendor/{oscharko-dev-keiko-evaluations-1.1.6.tgz → oscharko-dev-keiko-evaluations-1.1.8.tgz} +0 -0
  259. package/vendor/{oscharko-dev-keiko-evidence-1.1.6.tgz → oscharko-dev-keiko-evidence-1.1.8.tgz} +0 -0
  260. package/vendor/oscharko-dev-keiko-git-1.1.8.tgz +0 -0
  261. package/vendor/{oscharko-dev-keiko-harness-1.1.6.tgz → oscharko-dev-keiko-harness-1.1.8.tgz} +0 -0
  262. package/vendor/{oscharko-dev-keiko-local-knowledge-1.1.6.tgz → oscharko-dev-keiko-local-knowledge-1.1.8.tgz} +0 -0
  263. package/vendor/{oscharko-dev-keiko-memory-capture-1.1.6.tgz → oscharko-dev-keiko-memory-capture-1.1.8.tgz} +0 -0
  264. package/vendor/oscharko-dev-keiko-memory-consolidation-1.1.8.tgz +0 -0
  265. package/vendor/{oscharko-dev-keiko-memory-governance-1.1.6.tgz → oscharko-dev-keiko-memory-governance-1.1.8.tgz} +0 -0
  266. package/vendor/{oscharko-dev-keiko-memory-retrieval-1.1.6.tgz → oscharko-dev-keiko-memory-retrieval-1.1.8.tgz} +0 -0
  267. package/vendor/{oscharko-dev-keiko-memory-vault-1.1.6.tgz → oscharko-dev-keiko-memory-vault-1.1.8.tgz} +0 -0
  268. package/vendor/oscharko-dev-keiko-model-gateway-1.1.8.tgz +0 -0
  269. package/vendor/{oscharko-dev-keiko-quality-intelligence-1.1.6.tgz → oscharko-dev-keiko-quality-intelligence-1.1.8.tgz} +0 -0
  270. package/vendor/{oscharko-dev-keiko-sandbox-1.1.6.tgz → oscharko-dev-keiko-sandbox-1.1.8.tgz} +0 -0
  271. package/vendor/{oscharko-dev-keiko-sdk-1.1.6.tgz → oscharko-dev-keiko-sdk-1.1.8.tgz} +0 -0
  272. package/vendor/oscharko-dev-keiko-security-1.1.8.tgz +0 -0
  273. package/vendor/oscharko-dev-keiko-server-1.1.8.tgz +0 -0
  274. package/vendor/{oscharko-dev-keiko-tool-catalog-1.1.6.tgz → oscharko-dev-keiko-tool-catalog-1.1.8.tgz} +0 -0
  275. package/vendor/{oscharko-dev-keiko-tools-1.1.6.tgz → oscharko-dev-keiko-tools-1.1.8.tgz} +0 -0
  276. package/vendor/{oscharko-dev-keiko-verification-1.1.6.tgz → oscharko-dev-keiko-verification-1.1.8.tgz} +0 -0
  277. package/vendor/{oscharko-dev-keiko-workflows-1.1.6.tgz → oscharko-dev-keiko-workflows-1.1.8.tgz} +0 -0
  278. package/vendor/{oscharko-dev-keiko-workspace-1.1.6.tgz → oscharko-dev-keiko-workspace-1.1.8.tgz} +0 -0
  279. package/dist/ui/static/_next/static/chunks/0c3w2ubla6ek6.js +0 -1
  280. package/dist/ui/static/_next/static/chunks/0gnizm29lk3ll.js +0 -1
  281. package/dist/ui/static/_next/static/chunks/0psckgz_yhen3.js +0 -1
  282. package/dist/ui/static/_next/static/chunks/19fesd0w1gjhp.js +0 -3
  283. package/dist/ui/static/_next/static/chunks/1gb6os129fico.js +0 -1
  284. package/dist/ui/static/_next/static/chunks/1k6f6m-u2q69-.js +0 -1
  285. package/dist/ui/static/_next/static/chunks/1pze64zkvx8vt.js +0 -1
  286. package/dist/ui/static/_next/static/chunks/28mbnwl9g65al.js +0 -1
  287. package/dist/ui/static/_next/static/chunks/2bm60h9zwhm3v.js +0 -1
  288. package/dist/ui/static/_next/static/chunks/2h_zvmb272ozs.js +0 -1
  289. package/dist/ui/static/_next/static/chunks/3-5_wuwrev-8c.js +0 -1
  290. package/dist/ui/static/_next/static/chunks/3-l-8qt163_y6.js +0 -1
  291. package/dist/ui/static/_next/static/chunks/3doin5vaz4zjs.js +0 -1
  292. package/vendor/oscharko-dev-keiko-contracts-1.1.6.tgz +0 -0
  293. package/vendor/oscharko-dev-keiko-git-1.1.6.tgz +0 -0
  294. package/vendor/oscharko-dev-keiko-memory-consolidation-1.1.6.tgz +0 -0
  295. package/vendor/oscharko-dev-keiko-model-gateway-1.1.6.tgz +0 -0
  296. package/vendor/oscharko-dev-keiko-security-1.1.6.tgz +0 -0
  297. package/vendor/oscharko-dev-keiko-server-1.1.6.tgz +0 -0
  298. /package/dist/ui/static/_next/static/{ytpWXc-OicMB-rH2zX5b3 → xFln0tD9J1Dx0Nlv71fAF}/_buildManifest.js +0 -0
  299. /package/dist/ui/static/_next/static/{ytpWXc-OicMB-rH2zX5b3 → xFln0tD9J1Dx0Nlv71fAF}/_clientMiddlewareManifest.js +0 -0
  300. /package/dist/ui/static/_next/static/{ytpWXc-OicMB-rH2zX5b3 → xFln0tD9J1Dx0Nlv71fAF}/_ssgManifest.js +0 -0
@@ -1,10 +1,10 @@
1
1
  import { gatewaySpendRejectionReason } from "./gateway-spend-budget.js";
2
- import { AuthenticationError, CircuitOpenError, ContextOverflowError, ModelRefusalError, ProviderError, RateLimitError, TimeoutError, TransportError, } from "@oscharko-dev/keiko-security/errors/gateway";
2
+ import { CodingRuntimeLaunchRejectedError } from "./coding-runtime/launchFailure.js";
3
+ import { AuthenticationError, CircuitOpenError, ContextOverflowError, MalformedToolCallError, ModelRefusalError, ProviderEmptyAnswerError, ProviderError, ProviderOutputExhaustedError, RateLimitError, TimeoutError, TransportError, } from "@oscharko-dev/keiko-security/errors/gateway";
3
4
  import { createHash, randomUUID } from "node:crypto";
4
- import { findConfiguredCapability, resolveCodingSafeSidecarGatewayProfile, } from "@oscharko-dev/keiko-model-gateway";
5
+ import { findConfiguredCapability, resolveCodingSafeSidecarGatewayProfile, ResponseRedactionError, } from "@oscharko-dev/keiko-model-gateway";
5
6
  import { countGatewayPromptTokens, } from "@oscharko-dev/keiko-model-gateway/internal/prompt-token-accounting";
6
- import { codingWorkbenchProviderTimeoutMs, providerRequestBudgetMs, } from "@oscharko-dev/keiko-model-gateway/internal/resilience";
7
- import { MAX_TIMER_DELAY_MS } from "./abort-race.js";
7
+ import { gatewayRouteDeadlineMs } from "./gateway-route-deadline.js";
8
8
  import { compareStrings } from "@oscharko-dev/keiko-contracts/runtime/comparators";
9
9
  import { CODING_WORKBENCH_MINIMUM_CODING_CONTEXT_PROMPT_TOKENS } from "@oscharko-dev/keiko-contracts/runtime/coding-workbench";
10
10
  import { activityLogEvent, defineActivityLogOperation, } from "@oscharko-dev/keiko-contracts/runtime/observability";
@@ -13,11 +13,13 @@ import { currentGateway, currentGatewayConfig, currentGatewayVerification, } fro
13
13
  import { OPENCODE_RUNTIME_MODEL_ALIAS, OPENCODE_RUNTIME_READINESS_PROMPT, } from "./coding-runtime/opencodeLaunchProfile.js";
14
14
  import { createOpenCodeGatewayToolCatalogAdvertisement, opencodeGatewayOfferLifetimeMs, hasExactOpenCodeVisibleToolContract, OPENCODE_MODEL_VISIBLE_TOOL_NAMES, } from "./coding-runtime/opencodeToolSchemas.js";
15
15
  import { correlationIdOrUnknown, UNKNOWN_CORRELATION_ID } from "./correlation.js";
16
- import { emitServerDiagnostic, serverDiagnosticFromError } from "./diagnostics-log.js";
16
+ import { describeError, emitServerDiagnostic, serverDiagnosticFromError, } from "./diagnostics-log.js";
17
17
  import { readJsonObject } from "./files.js";
18
- import { ensureCodingWorkbenchContextWindows } from "./gateway-readiness.js";
18
+ import { safetyMarginTokensFor } from "@oscharko-dev/keiko-contracts/context-engineering";
19
+ import { ensureCodingWorkbenchContextWindows, isAnyCodingWorkbenchProbePending, isCodingWorkbenchProbePending, } from "./gateway-readiness.js";
19
20
  import { getServerLogger } from "./observability/index.js";
20
21
  import { STREAMING, errorBody } from "./routes.js";
22
+ import { sseBackpressureReporter, writeOrDestroy, } from "./sse-write.js";
21
23
  import { startSseHeartbeat } from "./sse.js";
22
24
  import { createCanonicalOpenCodeHandlerCoverage } from "./tool-catalog/catalogToolFacadeBridge.js";
23
25
  const ENABLE_TOKENS = new Set(["1", "true", "on", "yes", "enabled"]);
@@ -62,31 +64,45 @@ function observedPromptSettlement(outcome, selected, unverified) {
62
64
  * authenticator (not every deployment wires it) is a silent no-op, never a failure.
63
65
  */
64
66
  function settlePromptTokenReservation(deps, reservation, actualPromptTokens) {
67
+ const providerReported = actualPromptTokens !== undefined && actualPromptTokens > 0;
68
+ return applyPromptTokenSettlement(deps, reservation, {
69
+ promptTokens: providerReported ? actualPromptTokens : reservation.reservedPromptTokens,
70
+ source: providerReported ? "provider-reported" : "reserved-estimate",
71
+ });
72
+ }
73
+ /**
74
+ * Releases a reservation whose provider call never ran — the client would not take the opening
75
+ * SSE frame, so the buffered path never called the chat factory and the streamed path never pulled
76
+ * the gateway's generator. The conservative rule above keeps the estimate only for a call that WAS
77
+ * dispatched; here nothing was spent, so the whole reservation goes back to the run's budget
78
+ * (#3602 review). Same idempotence and same not-wired fallback as a settlement.
79
+ */
80
+ function releasePromptTokenReservation(deps, reservation) {
81
+ return applyPromptTokenSettlement(deps, reservation, {
82
+ promptTokens: 0,
83
+ source: "released-unspent",
84
+ });
85
+ }
86
+ function applyPromptTokenSettlement(deps, reservation, usage) {
65
87
  if (reservation.settled) {
66
88
  if (reservation.settlement === undefined)
67
89
  throw new TypeError("missing prompt settlement");
68
90
  return reservation.settlement;
69
91
  }
70
92
  reservation.settled = true;
71
- const providerReported = actualPromptTokens !== undefined && actualPromptTokens > 0;
72
- const promptTokens = providerReported ? actualPromptTokens : reservation.reservedPromptTokens;
73
93
  const unverified = {
74
94
  promptTokens: reservation.reservedPromptTokens,
75
95
  source: "reserved-estimate",
76
96
  status: "unverified",
77
97
  };
78
- const selected = {
79
- promptTokens,
80
- source: providerReported ? "provider-reported" : "reserved-estimate",
81
- status: "settled",
82
- };
98
+ const selected = { ...usage, status: "settled" };
83
99
  reservation.settlement = unverified;
84
100
  const authenticator = runtimeCapabilityAuthenticator(deps);
85
101
  if (authenticator?.settlePromptTokens === undefined) {
86
102
  reservation.settlement = { ...selected, status: "not-wired" };
87
103
  return reservation.settlement;
88
104
  }
89
- const outcome = authenticator.settlePromptTokens(reservation.capability, reservation.reservedPromptTokens, promptTokens);
105
+ const outcome = authenticator.settlePromptTokens(reservation.capability, reservation.reservedPromptTokens, usage.promptTokens);
90
106
  reservation.settlement = observedPromptSettlement(outcome, selected, unverified);
91
107
  return reservation.settlement;
92
108
  }
@@ -114,6 +130,9 @@ const CODING_SIDECAR_GATEWAY_REQUEST_VALIDATED_OPERATION = defineActivityLogOper
114
130
  maxRequestBytes: { type: "integer", dataClass: "count", required: true },
115
131
  maxPromptTokens: { type: "integer", dataClass: "count", required: true },
116
132
  estimatedPromptTokens: { type: "integer", dataClass: "count", required: true },
133
+ // #3591 (1.1.7): the output allowance sent with this request, clamped to the window that
134
+ // remains after the prompt — the value an output-exhausted turn has to be read against.
135
+ maxOutputTokens: { type: "integer", dataClass: "count", required: false },
117
136
  inputMessageCount: { type: "integer", dataClass: "count", required: true },
118
137
  completeness: { type: "string", dataClass: "completeness-state", required: true },
119
138
  loss: { type: "string", dataClass: "loss-state", required: true },
@@ -137,11 +156,21 @@ const CODING_SIDECAR_GATEWAY_READINESS_INSUFFICIENT_OPERATION = defineActivityLo
137
156
  type: "string",
138
157
  dataClass: "closed-enum",
139
158
  required: true,
140
- values: ["model-context-window-insufficient", "no-tool-calling", "tool-calling-unverified"],
159
+ values: [
160
+ "model-context-window-insufficient",
161
+ "no-tool-calling",
162
+ "tool-calling-unverified",
163
+ "model-verification-pending",
164
+ ],
141
165
  },
142
166
  maxPromptTokens: { type: "integer", dataClass: "count", required: false },
143
167
  minimumRequiredPromptTokens: { type: "integer", dataClass: "count", required: false },
144
- probeMode: { type: "string", dataClass: "closed-enum", required: true, values: ["passive"] },
168
+ probeMode: {
169
+ type: "string",
170
+ dataClass: "closed-enum",
171
+ required: true,
172
+ values: ["passive", "pending"],
173
+ },
145
174
  completeness: { type: "string", dataClass: "completeness-state", required: true },
146
175
  loss: { type: "string", dataClass: "loss-state", required: true },
147
176
  },
@@ -240,6 +269,9 @@ const CODING_SIDECAR_GATEWAY_REJECTED_OPERATION = defineActivityLogOperation({
240
269
  },
241
270
  estimatedPromptTokens: { type: "integer", dataClass: "count", required: false },
242
271
  maxPromptTokens: { type: "integer", dataClass: "count", required: false },
272
+ // #3591 review: the bound the prompt was actually admitted against — `maxPromptTokens` less
273
+ // the safety margin and the minimum output allowance (`admissiblePromptTokens`).
274
+ admissiblePromptTokens: { type: "integer", dataClass: "count", required: false },
243
275
  inputMessageCount: { type: "integer", dataClass: "count", required: false },
244
276
  maxInputMessages: { type: "integer", dataClass: "count", required: false },
245
277
  completeness: { type: "string", dataClass: "completeness-state", required: true },
@@ -272,7 +304,14 @@ const CODING_SIDECAR_GATEWAY_TURN_FAILED_OPERATION = defineActivityLogOperation(
272
304
  type: "string",
273
305
  dataClass: "closed-enum",
274
306
  required: true,
275
- values: ["provider-failed", "stream-incomplete", "turn-rejected"],
307
+ values: [
308
+ "provider-failed",
309
+ "stream-incomplete",
310
+ "turn-rejected",
311
+ "output-exhausted",
312
+ "empty-answer",
313
+ "invalid-tool-call",
314
+ ],
276
315
  },
277
316
  published: { type: "boolean", dataClass: "closed-enum", required: true },
278
317
  publicationReason: {
@@ -288,6 +327,23 @@ const CODING_SIDECAR_GATEWAY_TURN_FAILED_OPERATION = defineActivityLogOperation(
288
327
  "capacity-pressure",
289
328
  ],
290
329
  },
330
+ // The Keiko-code frames and cause classes of the failure, when the turn failed on an error
331
+ // (PR #3617 review): a model-answer failure writes no error-level diagnostic, so this line is
332
+ // where its frames live.
333
+ frames: {
334
+ type: "string-array",
335
+ dataClass: "safe-platform-class",
336
+ required: false,
337
+ maxLength: 512,
338
+ maxItems: 8,
339
+ },
340
+ causeChain: {
341
+ type: "string-array",
342
+ dataClass: "error-kind",
343
+ required: false,
344
+ maxLength: 128,
345
+ maxItems: 5,
346
+ },
291
347
  completeness: { type: "string", dataClass: "completeness-state", required: true },
292
348
  loss: { type: "string", dataClass: "loss-state", required: true },
293
349
  },
@@ -309,11 +365,12 @@ const CODING_SIDECAR_GATEWAY_USAGE_SETTLED_OPERATION = defineActivityLogOperatio
309
365
  runId: { type: "string", dataClass: "opaque-id", required: true, maxLength: 128 },
310
366
  completionTokens: { type: "integer", dataClass: "count", required: true },
311
367
  promptTokens: { type: "integer", dataClass: "count", required: true },
368
+ // `released-unspent`: the provider call never ran, the whole reservation went back (#3602).
312
369
  promptSource: {
313
370
  type: "string",
314
371
  dataClass: "closed-enum",
315
372
  required: true,
316
- values: ["provider-reported", "reserved-estimate"],
373
+ values: ["provider-reported", "reserved-estimate", "released-unspent"],
317
374
  },
318
375
  promptSettlementStatus: {
319
376
  type: "string",
@@ -355,6 +412,16 @@ const CODING_SIDECAR_GATEWAY_OUTCOME_OPERATION = defineActivityLogOperation({
355
412
  },
356
413
  completionTokens: { type: "integer", dataClass: "count", required: true },
357
414
  outputBytes: { type: "integer", dataClass: "count", required: true },
415
+ // The route backstop armed for this turn (#3602 review); absent on lines written before 1.1.7.
416
+ deadlineMs: { type: "integer", dataClass: "duration", required: false },
417
+ // On a cancelled outcome only: which armed abort source ended the turn, so a stall that ran into
418
+ // the backstop, or a slow client the shared SSE path killed, never reads like a client that left.
419
+ cancellationCause: {
420
+ type: "string",
421
+ dataClass: "closed-enum",
422
+ required: false,
423
+ values: ["client-disconnect", "route-deadline", "backpressure-killed", "run-stopped"],
424
+ },
358
425
  completeness: { type: "string", dataClass: "completeness-state", required: true },
359
426
  loss: { type: "string", dataClass: "loss-state", required: true },
360
427
  },
@@ -463,6 +530,25 @@ export function _classifyBadRequestReasonForTests(result) {
463
530
  function isModelReasoningEffort(value) {
464
531
  return typeof value === "string" && MODEL_REASONING_EFFORT_SET.has(value);
465
532
  }
533
+ // A run's one pending challenge wait: an observed request ends it with true; a refused request, a
534
+ // clear, the start signal, or a newer wait for the same run ends it with false and disarms the run.
535
+ function pendingChallengeWait(runId, signal, armed, waiters) {
536
+ return new Promise((resolve) => {
537
+ const settle = (result) => {
538
+ signal.removeEventListener("abort", abort);
539
+ if (waiters.get(runId) === settle)
540
+ waiters.delete(runId);
541
+ if (!result)
542
+ armed.delete(runId);
543
+ resolve(result);
544
+ };
545
+ const abort = () => {
546
+ settle(false);
547
+ };
548
+ waiters.set(runId, settle);
549
+ signal.addEventListener("abort", abort, { once: true });
550
+ });
551
+ }
466
552
  export function createOpenCodeGatewayReadinessRegistry() {
467
553
  const observed = new Set();
468
554
  const armed = new Set();
@@ -488,21 +574,12 @@ export function createOpenCodeGatewayReadinessRegistry() {
488
574
  return Promise.resolve(false);
489
575
  waiters.get(runId)?.(false);
490
576
  armed.add(runId);
491
- return new Promise((resolve) => {
492
- const settle = (result) => {
493
- signal.removeEventListener("abort", abort);
494
- if (waiters.get(runId) === settle)
495
- waiters.delete(runId);
496
- if (!result)
497
- armed.delete(runId);
498
- resolve(result);
499
- };
500
- const abort = () => {
501
- settle(false);
502
- };
503
- waiters.set(runId, settle);
504
- signal.addEventListener("abort", abort, { once: true });
505
- });
577
+ return pendingChallengeWait(runId, signal, armed, waiters);
578
+ },
579
+ refuseChallenge: (runId) => {
580
+ if (!armed.delete(runId))
581
+ return;
582
+ waiters.get(runId)?.(false);
506
583
  },
507
584
  noteAdoptionGapDiagnosed: (runId) => {
508
585
  if (adoptionGapDiagnosed.has(runId))
@@ -895,8 +972,25 @@ function emitGatewayEvidenceAggregationDiagnostic(deps, runId) {
895
972
  message: "sidecar-gateway-evidence-aggregation-failed",
896
973
  });
897
974
  }
898
- function recordGatewayOutcome(ctx, deps, runId, outcome, completionTokens, outputBytes) {
899
- getServerLogger().info(activityLogEvent(CODING_SIDECAR_GATEWAY_OUTCOME_OPERATION, { correlationId: correlationIdOrUnknown(ctx.correlationId), parentCorrelationId: runId }, { runId, outcome, completionTokens, outputBytes, completeness: "complete", loss: "none" }));
975
+ // A cancelled outcome names the abort source that ended the turn. Without an aborted source it is
976
+ // the transport path — an SSE or terminal write found the response gone before its `close`
977
+ // listener ran — which is the client leaving, never the deadline.
978
+ function gatewayOutcomeCancellationCause(cancellation, outcome) {
979
+ if (outcome !== "cancelled")
980
+ return undefined;
981
+ return { cancellationCause: cancellation.cause() ?? "client-disconnect" };
982
+ }
983
+ function recordGatewayOutcome(ctx, deps, runId, cancellation, outcome, completionTokens, outputBytes) {
984
+ getServerLogger().info(activityLogEvent(CODING_SIDECAR_GATEWAY_OUTCOME_OPERATION, { correlationId: correlationIdOrUnknown(ctx.correlationId), parentCorrelationId: runId }, {
985
+ runId,
986
+ outcome,
987
+ completionTokens,
988
+ outputBytes,
989
+ deadlineMs: cancellation.deadlineMs,
990
+ ...gatewayOutcomeCancellationCause(cancellation, outcome),
991
+ completeness: "complete",
992
+ loss: "none",
993
+ }));
900
994
  try {
901
995
  void Promise.resolve(evidenceAggregator(deps)?.record({ runId, outcome, completionTokens, outputBytes })).catch(() => {
902
996
  emitGatewayEvidenceAggregationDiagnostic(deps, runId);
@@ -969,8 +1063,9 @@ function budgetValidationError(parsed, runMetadata, estimatedPromptTokens) {
969
1063
  if (parsed.messages.length > runMetadata.maxInputMessages) {
970
1064
  return contextOverflowRequest(`Request body messages exceed profile maxInputMessages (${String(runMetadata.maxInputMessages)}).`);
971
1065
  }
972
- if (estimatedPromptTokens > runMetadata.maxPromptTokens) {
973
- return contextOverflowRequest(`Request body estimated prompt tokens exceed profile maxPromptTokens (${String(runMetadata.maxPromptTokens)}).`);
1066
+ const admissible = admissiblePromptTokens(runMetadata);
1067
+ if (estimatedPromptTokens > admissible) {
1068
+ return contextOverflowRequest(`Request body estimated prompt tokens exceed profile maxPromptTokens (${String(runMetadata.maxPromptTokens)}) less the reserved output allowance (${String(admissible)} admissible).`);
974
1069
  }
975
1070
  return undefined;
976
1071
  }
@@ -1013,7 +1108,26 @@ function gatewayDiagnosticCorrelation(ctx, runId) {
1013
1108
  ? { correlationId }
1014
1109
  : { correlationId, parentCorrelationId: runId };
1015
1110
  }
1016
- function emitGatewayFailureDiagnostic(ctx, deps, error, runId) {
1111
+ // A turn the model ended without a usable answer -- nothing at all, reasoning until its output budget
1112
+ // ran out, or a tool call that never parsed or matched its schema -- is the model's answer, not a
1113
+ // server fault: the warn-level turn-failed line names it. An error-level diagnostic opened a support
1114
+ // incident for every such turn; a lab run of the 1.1.8 candidate behind a LiteLLM hosted_vllm route
1115
+ // opened one on its first empty answer.
1116
+ const MODEL_ANSWER_FAILURES = new Set([
1117
+ "output-exhausted",
1118
+ "empty-answer",
1119
+ "invalid-tool-call",
1120
+ ]);
1121
+ function isModelAnswerFailure(error) {
1122
+ const cause = modelTurnFailureCode(error);
1123
+ return cause !== undefined && MODEL_ANSWER_FAILURES.has(cause);
1124
+ }
1125
+ // A model-answer failure writes no error-level diagnostic only when its warn-level turn-failed line
1126
+ // was written. A run no longer running or paused gets no such line, so the diagnostic keeps the
1127
+ // failure's class and frames (PR #3617 review).
1128
+ function emitGatewayFailureDiagnostic(ctx, deps, error, runId, turnFailureRecorded) {
1129
+ if (turnFailureRecorded && isModelAnswerFailure(error))
1130
+ return;
1017
1131
  emitServerDiagnostic(deps.diagnostics, serverDiagnosticFromError({
1018
1132
  ...gatewayDiagnosticCorrelation(ctx, runId),
1019
1133
  operation: CODING_SIDECAR_GATEWAY_ROUTE,
@@ -1022,19 +1136,23 @@ function emitGatewayFailureDiagnostic(ctx, deps, error, runId) {
1022
1136
  redact: (message) => String(deps.redactor(message)),
1023
1137
  }));
1024
1138
  }
1025
- function reportGatewayTurnFailure(ctx, deps, runId, failureCode) {
1139
+ /** Writes the run's turn-failed line; false for a run that is no longer running or paused. */
1140
+ function reportGatewayTurnFailure(ctx, deps, runId, failureCode, error) {
1026
1141
  const snapshot = deps.codingRuntimeOrchestrator?.getSnapshot(runId);
1027
1142
  if (snapshot?.state !== "running" && snapshot?.state !== "paused")
1028
- return;
1143
+ return false;
1029
1144
  const publicationReason = gatewayTurnFailurePublication(deps, runId, snapshot, failureCode);
1030
- logGatewayTurnFailure(ctx, runId, snapshot.revision, snapshot.state, failureCode, publicationReason);
1145
+ const run = { revision: snapshot.revision, state: snapshot.state };
1146
+ logGatewayTurnFailure(ctx, runId, run, failureCode, publicationReason, error);
1147
+ return true;
1031
1148
  }
1032
1149
  function gatewayTurnFailurePublication(deps, runId, snapshot, failureCode) {
1033
1150
  const publication = deps.codingRuntimeEventHub?.publishTurnFailure(runId, snapshot.state, snapshot.revision, failureCode);
1034
1151
  const publicationReason = publication?.ok === true ? "published" : (publication?.reason ?? "event-hub-unavailable");
1035
1152
  return publicationReason;
1036
1153
  }
1037
- function logGatewayTurnFailure(ctx, runId, revision, state, failureCode, publicationReason) {
1154
+ function logGatewayTurnFailure(ctx, runId, { revision, state }, failureCode, publicationReason, error) {
1155
+ const { frames, causeChain } = describeError(error);
1038
1156
  getServerLogger().warn(activityLogEvent(CODING_SIDECAR_GATEWAY_TURN_FAILED_OPERATION, {
1039
1157
  ...gatewayDiagnosticCorrelation(ctx, runId),
1040
1158
  errorKind: failureCode === "turn-rejected" ? "validation-failed" : "unavailable",
@@ -1045,13 +1163,35 @@ function logGatewayTurnFailure(ctx, runId, revision, state, failureCode, publica
1045
1163
  failureCode,
1046
1164
  published: publicationReason === "published",
1047
1165
  publicationReason,
1166
+ ...(error === undefined || frames === undefined ? {} : { frames }),
1167
+ ...(error === undefined || causeChain === undefined ? {} : { causeChain }),
1048
1168
  completeness: "complete",
1049
1169
  loss: "none",
1050
1170
  }));
1051
1171
  }
1052
- function gatewayTurnFailureCode(error) {
1053
- if (error instanceof ContextOverflowError || error instanceof ModelRefusalError)
1172
+ // The causes the buffered and the streamed path name the same way. Both output-budget exhaustion
1173
+ // and an empty answer are HTTP 200 provider errors, so they are resolved before any status check.
1174
+ function modelTurnFailureCode(error) {
1175
+ // The gateway's own redaction refused an answer nested too deep to walk: a Workbench guard
1176
+ // rejected the turn, whether or not a tool was called, so it is no invalid tool call and keeps
1177
+ // its error-level diagnostic (PR #3617 review). It extends MalformedToolCallError, so it is named
1178
+ // before that check.
1179
+ if (error instanceof ContextOverflowError ||
1180
+ error instanceof ModelRefusalError ||
1181
+ error instanceof ResponseRedactionError)
1054
1182
  return "turn-rejected";
1183
+ if (error instanceof ProviderOutputExhaustedError)
1184
+ return "output-exhausted";
1185
+ if (error instanceof ProviderEmptyAnswerError)
1186
+ return "empty-answer";
1187
+ if (error instanceof MalformedToolCallError)
1188
+ return "invalid-tool-call";
1189
+ return undefined;
1190
+ }
1191
+ function gatewayTurnFailureCode(error) {
1192
+ const modelCause = modelTurnFailureCode(error);
1193
+ if (modelCause !== undefined)
1194
+ return modelCause;
1055
1195
  if (error instanceof TimeoutError ||
1056
1196
  error instanceof TransportError ||
1057
1197
  (error instanceof ProviderError && error.httpStatus === 200))
@@ -1061,8 +1201,9 @@ function gatewayTurnFailureCode(error) {
1061
1201
  function gatewayStreamFailureCode(error) {
1062
1202
  if (gatewaySpendRejectionReason(error) !== undefined)
1063
1203
  return "turn-rejected";
1064
- if (error instanceof ContextOverflowError || error instanceof ModelRefusalError)
1065
- return "turn-rejected";
1204
+ const modelCause = modelTurnFailureCode(error);
1205
+ if (modelCause !== undefined)
1206
+ return modelCause;
1066
1207
  if (error instanceof AuthenticationError ||
1067
1208
  error instanceof RateLimitError ||
1068
1209
  error instanceof CircuitOpenError ||
@@ -1079,7 +1220,9 @@ function gatewayStreamFailureCode(error) {
1079
1220
  * stream died after N deltas". `partialUsage` rides along through `serverDiagnosticFromError`, so an
1080
1221
  * interrupted turn's accumulated token counts stay visible instead of vanishing with the error.
1081
1222
  */
1082
- function emitGatewayStreamFailureDiagnostic(ctx, deps, error, runId) {
1223
+ function emitGatewayStreamFailureDiagnostic(ctx, deps, error, runId, turnFailureRecorded) {
1224
+ if (turnFailureRecorded && isModelAnswerFailure(error))
1225
+ return;
1083
1226
  emitServerDiagnostic(deps.diagnostics, serverDiagnosticFromError({
1084
1227
  ...gatewayDiagnosticCorrelation(ctx, runId),
1085
1228
  operation: CODING_SIDECAR_GATEWAY_ROUTE,
@@ -1183,7 +1326,8 @@ function toolContractMismatch(tools) {
1183
1326
  toolMismatchSha256: toolNameSetDigest([...unexpected, "--missing--", ...missing]),
1184
1327
  };
1185
1328
  }
1186
- function emitGatewayToolContractDiagnostic(ctx, deps, runId, tools) {
1329
+ function emitGatewayToolContractDiagnostic(ctx, deps, runId, parsed) {
1330
+ const { tools } = parsed;
1187
1331
  const { code, reason } = toolContractRejectionReason(tools);
1188
1332
  emitServerDiagnostic(deps.diagnostics, {
1189
1333
  correlationId: ctx.correlationId ?? UNKNOWN_CORRELATION_ID,
@@ -1196,8 +1340,18 @@ function emitGatewayToolContractDiagnostic(ctx, deps, runId, tools) {
1196
1340
  code,
1197
1341
  });
1198
1342
  logGatewayRejection(ctx, runId, 403, reason, toolContractMismatch(tools));
1343
+ refuseReadinessChallenge(deps, runId, parsed);
1199
1344
  reportGatewayTurnFailure(ctx, deps, runId, "turn-rejected");
1200
1345
  }
1346
+ // Ends a pending readiness challenge at once when the route refused the challenge's own request: a
1347
+ // deterministic refusal of the readiness prompt cannot turn into an observed request later (#3603).
1348
+ // A refused side request, a title or compaction call the runtime sends meanwhile, leaves the
1349
+ // challenge waiting for the readiness prompt (PR #3617 review).
1350
+ function refuseReadinessChallenge(deps, runId, parsed) {
1351
+ if (parsed !== undefined && isRuntimeReadinessProbe(parsed)) {
1352
+ gatewayReadinessRegistry(deps)?.refuseChallenge(runId);
1353
+ }
1354
+ }
1201
1355
  /**
1202
1356
  * True when a long managed-tool-set history never invoked one governed keiko_* tool: the model is
1203
1357
  * burning turns without adopting the projected suite. Question/todowrite calls do not count as
@@ -1351,25 +1505,37 @@ function gatewayAuthorityKind(authentication) {
1351
1505
  function unavailableGatewayReason(resolved) {
1352
1506
  return resolved.result.status === "unavailable" ? resolved.result.reason : "missing-provider";
1353
1507
  }
1508
+ function sidecarSseTransport(ctx, deps) {
1509
+ const correlationId = correlationIdOrUnknown(ctx.correlationId);
1510
+ return {
1511
+ backpressure: new AbortController(),
1512
+ onBackpressure: sseBackpressureReporter(deps, "coding-sidecar-gateway", correlationId),
1513
+ correlationId,
1514
+ };
1515
+ }
1516
+ // `AbortSignal.any` carries the reason of the source that aborted first, so a client that leaves
1517
+ // after the deadline already fired still reads as the deadline, never the other way round.
1518
+ function gatewayCancellationCause(signal, sources) {
1519
+ if (!signal.aborted)
1520
+ return undefined;
1521
+ if (signal.reason === sources.client.reason)
1522
+ return "client-disconnect";
1523
+ if (signal.reason === sources.deadline.reason)
1524
+ return "route-deadline";
1525
+ if (signal.reason === sources.backpressure.reason)
1526
+ return "backpressure-killed";
1527
+ return "run-stopped";
1528
+ }
1354
1529
  // The route's deadline is a backstop BEHIND the gateway's own end-to-end budget, never the budget
1355
- // itself. It used to be the provider's per-attempt `timeoutMs`: the first attempt that hung spent
1356
- // it, and this deadline, armed before the gateway started its own clock, aborted the retry the
1357
- // gateway had just scheduled, so a provider timeout surfaced as a cancellation nobody had asked
1358
- // for and failed the run (coding run 23, 2026-09-11). The grace lets the gateway settle its own
1359
- // timeout or exhausted-retry error first.
1360
- const GATEWAY_ROUTE_DEADLINE_GRACE_MS = 1_000;
1530
+ // itself (`gateway-route-deadline.ts`, shared with the commit draft since the #3602 review). It used
1531
+ // to be the provider's per-attempt `timeoutMs`: the first attempt that hung spent it, and this
1532
+ // deadline, armed before the gateway started its own clock, aborted the retry the gateway had just
1533
+ // scheduled, so a provider timeout surfaced as a cancellation nobody had asked for and failed the
1534
+ // run (coding run 23, 2026-09-11).
1361
1535
  export function codingSidecarGatewayRequestDeadlineMs(config, modelId) {
1362
- const provider = config.providers.find((candidate) => candidate.modelId === modelId);
1363
- // An unconfigured model is refused before any provider call; 30 s only bounds that refusal.
1364
- const budget = provider === undefined
1365
- ? 30_000
1366
- : providerRequestBudgetMs({
1367
- ...provider,
1368
- timeoutMs: codingWorkbenchProviderTimeoutMs(provider.timeoutMs),
1369
- });
1370
- // Armed with AbortSignal.timeout, which fires at once past 2^31 - 1 ms: an absurd budget must not
1371
- // turn the backstop into an immediate abort.
1372
- return Math.min(budget + GATEWAY_ROUTE_DEADLINE_GRACE_MS, MAX_TIMER_DELAY_MS);
1536
+ // The sidecar reaches the gateway both ways — a buffered `chat()` answer or a `chatStream()`
1537
+ // read — so its backstop sits behind the longer of the two budgets.
1538
+ return gatewayRouteDeadlineMs(config, modelId, ["buffered", "streamed"]);
1373
1539
  }
1374
1540
  function gatewayRequestCancellation(ctx, deps, config, modelId, runId) {
1375
1541
  const client = new AbortController();
@@ -1378,11 +1544,18 @@ function gatewayRequestCancellation(ctx, deps, config, modelId, runId) {
1378
1544
  };
1379
1545
  ctx.req.once("aborted", abortClient);
1380
1546
  ctx.res.once("close", abortClient);
1381
- const deadline = AbortSignal.timeout(codingSidecarGatewayRequestDeadlineMs(config, modelId));
1547
+ const deadlineMs = codingSidecarGatewayRequestDeadlineMs(config, modelId);
1548
+ const deadline = AbortSignal.timeout(deadlineMs);
1549
+ const transport = sidecarSseTransport(ctx, deps);
1382
1550
  const runSignal = cancellationRegistry(deps)?.signalFor(runId);
1383
- const signals = [client.signal, deadline, runSignal].filter((signal) => signal !== undefined);
1551
+ const signals = [client.signal, deadline, transport.backpressure.signal, runSignal].filter((signal) => signal !== undefined);
1552
+ const signal = AbortSignal.any(signals);
1553
+ const sources = { client: client.signal, deadline, backpressure: transport.backpressure.signal };
1384
1554
  return {
1385
- signal: AbortSignal.any(signals),
1555
+ signal,
1556
+ deadlineMs,
1557
+ transport,
1558
+ cause: () => gatewayCancellationCause(signal, sources),
1386
1559
  dispose: () => {
1387
1560
  ctx.req.removeListener("aborted", abortClient);
1388
1561
  ctx.res.removeListener("close", abortClient);
@@ -1395,23 +1568,23 @@ function requestForGatewayDelivery(ctx, parsed, delivery, signal) {
1395
1568
  async function executeGatewayChat(ctx, deps, binding, parsed, runId, delivery) {
1396
1569
  const cancellation = gatewayRequestCancellation(ctx, deps, binding.config, delivery.modelAlias, runId);
1397
1570
  try {
1398
- return await dispatchGatewayChat(ctx, deps, binding, parsed, runId, delivery, cancellation.signal);
1571
+ return await dispatchGatewayChat(ctx, deps, binding, parsed, runId, delivery, cancellation);
1399
1572
  }
1400
1573
  finally {
1401
1574
  cancellation.dispose();
1402
1575
  }
1403
1576
  }
1404
1577
  // Extracted so `executeGatewayChat` stays under AGENTS.md §6's 50-line ceiling.
1405
- async function dispatchGatewayChat(ctx, deps, binding, parsed, runId, delivery, cancellationSignal) {
1578
+ async function dispatchGatewayChat(ctx, deps, binding, parsed, runId, delivery, cancellation) {
1406
1579
  const { modelAlias, upstreamStreamingSupported } = delivery;
1407
- const request = requestForGatewayDelivery(ctx, parsed, delivery, cancellationSignal);
1580
+ const request = requestForGatewayDelivery(ctx, parsed, delivery, cancellation.signal);
1408
1581
  const dispatch = {
1409
1582
  deps,
1410
1583
  binding,
1411
1584
  modelAlias,
1412
1585
  request,
1413
1586
  runId,
1414
- cancellationSignal,
1587
+ cancellation,
1415
1588
  promptTokenReservation: delivery.promptTokenReservation,
1416
1589
  };
1417
1590
  let bufferedStream;
@@ -1419,21 +1592,41 @@ async function dispatchGatewayChat(ctx, deps, binding, parsed, runId, delivery,
1419
1592
  if (parsed.stream && upstreamStreamingSupported) {
1420
1593
  return await streamGatewayChat(ctx, dispatch);
1421
1594
  }
1422
- if (parsed.stream)
1423
- bufferedStream = beginBufferedOpenAiStream(ctx, modelAlias);
1595
+ if (parsed.stream) {
1596
+ bufferedStream = beginBufferedOpenAiStream(ctx, modelAlias, cancellation.transport);
1597
+ if (bufferedStream === undefined)
1598
+ return settleUndeliverableBufferedStream(ctx, dispatch);
1599
+ }
1424
1600
  return await executeBufferedGatewayChat(ctx, dispatch, bufferedStream);
1425
1601
  }
1426
1602
  catch (error) {
1427
- return settleFailedGatewayChat(ctx, deps, runId, cancellationSignal, error, delivery, bufferedStream);
1603
+ return settleFailedGatewayChat(ctx, deps, runId, cancellation, error, delivery, bufferedStream);
1428
1604
  }
1429
1605
  }
1606
+ // The opening frame never reached the client — the shared path killed the stream, or the response
1607
+ // was already gone — so no provider call is started for an answer nobody can receive, exactly as
1608
+ // `beginGatewayStream` ends the streamed path (#3602 review). The outcome names the abort source.
1609
+ function settleUndeliverableBufferedStream(ctx, dispatch) {
1610
+ const { deps, runId, cancellation, promptTokenReservation } = dispatch;
1611
+ recordGatewayOutcome(ctx, deps, runId, cancellation, "cancelled", 0, 0);
1612
+ releaseUndispatchedPromptBudget(ctx, deps, runId, promptTokenReservation);
1613
+ return STREAMING;
1614
+ }
1615
+ // A handshake the client would not take leaves no provider call behind, so the reservation is
1616
+ // released rather than kept as spent, and the usage line records the release (zero prompt tokens,
1617
+ // `released-unspent`) so the ledger movement is reconstructible from the log.
1618
+ function releaseUndispatchedPromptBudget(ctx, deps, runId, reservation) {
1619
+ const settlement = releasePromptTokenReservation(deps, reservation);
1620
+ logGatewayCompletionUsage(ctx, runId, { completionTokens: 0, outputBytes: 0 }, "output-byte-estimate", settlement);
1621
+ }
1430
1622
  // Extracted so `executeGatewayChat` stays under AGENTS.md §6's 50-line ceiling.
1431
- function settleFailedGatewayChat(ctx, deps, runId, cancellationSignal, error, delivery, bufferedStream) {
1432
- recordGatewayOutcome(ctx, deps, runId, cancellationSignal.aborted ? "cancelled" : "failed", 0, 0);
1433
- emitGatewayFailureDiagnostic(ctx, deps, error, runId);
1623
+ function settleFailedGatewayChat(ctx, deps, runId, cancellation, error, delivery, bufferedStream) {
1624
+ const cancelled = cancellation.signal.aborted;
1625
+ recordGatewayOutcome(ctx, deps, runId, cancellation, cancelled ? "cancelled" : "failed", 0, 0);
1434
1626
  const spendReason = gatewaySpendRejectionReason(error);
1435
- if (!cancellationSignal.aborted)
1436
- reportGatewayTurnFailure(ctx, deps, runId, spendReason === undefined ? gatewayTurnFailureCode(error) : "turn-rejected");
1627
+ const turnFailureRecorded = !cancelled &&
1628
+ reportGatewayTurnFailure(ctx, deps, runId, spendReason === undefined ? gatewayTurnFailureCode(error) : "turn-rejected", error);
1629
+ emitGatewayFailureDiagnostic(ctx, deps, error, runId, turnFailureRecorded);
1437
1630
  settlePromptTokenReservation(deps, delivery.promptTokenReservation);
1438
1631
  if (spendReason !== undefined && bufferedStream === undefined) {
1439
1632
  logGatewayRejection(ctx, runId, 403, spendReason);
@@ -1444,7 +1637,7 @@ function settleFailedGatewayChat(ctx, deps, runId, cancellationSignal, error, de
1444
1637
  : settleBufferedOpenAiStreamError(bufferedStream, "error");
1445
1638
  }
1446
1639
  async function executeBufferedGatewayChat(ctx, dispatch, stream) {
1447
- const { deps, binding, modelAlias, request, runId, cancellationSignal, promptTokenReservation } = dispatch;
1640
+ const { deps, binding, modelAlias, request, runId, cancellation, promptTokenReservation } = dispatch;
1448
1641
  const response = await chatFactoryFor(deps, binding.gateway)(binding.config, modelAlias)(request);
1449
1642
  const promptSettlement = settlePromptTokenReservation(deps, promptTokenReservation, response.usage.promptTokens);
1450
1643
  const output = outputMetrics(response);
@@ -1456,9 +1649,9 @@ async function executeBufferedGatewayChat(ctx, dispatch, stream) {
1456
1649
  };
1457
1650
  logGatewayCompletionUsage(ctx, runId, metrics, usage.source, promptSettlement);
1458
1651
  const record = (outcome) => {
1459
- recordGatewayOutcome(ctx, deps, runId, outcome, metrics.completionTokens, metrics.outputBytes);
1652
+ recordGatewayOutcome(ctx, deps, runId, cancellation, outcome, metrics.completionTokens, metrics.outputBytes);
1460
1653
  };
1461
- if (cancellationSignal.aborted) {
1654
+ if (cancellation.signal.aborted) {
1462
1655
  record("cancelled");
1463
1656
  return stream === undefined
1464
1657
  ? unavailableError()
@@ -1470,15 +1663,15 @@ async function executeBufferedGatewayChat(ctx, dispatch, stream) {
1470
1663
  ? unavailableError()
1471
1664
  : settleBufferedOpenAiStreamError(stream, "length");
1472
1665
  }
1473
- return deliverBufferedGatewayAnswer(ctx, deps, runId, stream, modelAlias, settledResponse, metrics);
1666
+ return deliverBufferedGatewayAnswer(ctx, modelAlias, stream, settledResponse, record);
1474
1667
  }
1475
- function deliverBufferedGatewayAnswer(ctx, deps, runId, stream, modelAlias, response, metrics) {
1668
+ function deliverBufferedGatewayAnswer(ctx, modelAlias, stream, response, record) {
1476
1669
  if (stream === undefined) {
1477
- recordGatewayOutcome(ctx, deps, runId, "accepted", metrics.completionTokens, metrics.outputBytes);
1670
+ record("accepted");
1478
1671
  return openAiResponse(modelAlias, response);
1479
1672
  }
1480
1673
  completeBufferedOpenAiStream(stream, response);
1481
- recordGatewayOutcome(ctx, deps, runId, ctx.res.writableEnded && !ctx.res.destroyed ? "accepted" : "cancelled", metrics.completionTokens, metrics.outputBytes);
1674
+ record(ctx.res.writableEnded && !ctx.res.destroyed ? "accepted" : "cancelled");
1482
1675
  return STREAMING;
1483
1676
  }
1484
1677
  // Closed stream state machine keeps iterator, cancellation, and SSE backpressure transitions together.
@@ -1489,16 +1682,22 @@ async function streamGatewayChat(ctx, dispatch) {
1489
1682
  iterator = chatStreamFactoryFor(deps, binding.gateway)(binding.config, modelAlias)(request)[Symbol.asyncIterator]();
1490
1683
  }
1491
1684
  catch (error) {
1492
- recordGatewayOutcome(ctx, deps, runId, "failed", 0, 0);
1493
- emitGatewayFailureDiagnostic(ctx, deps, error, runId);
1494
- reportGatewayTurnFailure(ctx, deps, runId, gatewayTurnFailureCode(error));
1685
+ recordGatewayOutcome(ctx, deps, runId, dispatch.cancellation, "failed", 0, 0);
1686
+ const turnFailureRecorded = reportGatewayTurnFailure(ctx, deps, runId, gatewayTurnFailureCode(error), error);
1687
+ emitGatewayFailureDiagnostic(ctx, deps, error, runId, turnFailureRecorded);
1495
1688
  settlePromptTokenReservation(deps, promptTokenReservation);
1496
1689
  return unavailableError();
1497
1690
  }
1498
1691
  const session = createGatewayStreamSession(ctx, dispatch, iterator);
1499
1692
  try {
1500
- if (beginGatewayStream(session))
1693
+ if (beginGatewayStream(session)) {
1501
1694
  await pumpGatewayStreamWithCancellation(deps, session);
1695
+ }
1696
+ else {
1697
+ // The gateway's stream is an async generator: nothing was sent before the first pull, and
1698
+ // the handshake failed before it, so the reservation is released, not kept as spent.
1699
+ releaseUndispatchedPromptBudget(ctx, deps, runId, promptTokenReservation);
1700
+ }
1502
1701
  return STREAMING;
1503
1702
  }
1504
1703
  finally {
@@ -1509,7 +1708,8 @@ async function streamGatewayChat(ctx, dispatch) {
1509
1708
  }
1510
1709
  }
1511
1710
  async function pumpGatewayStreamWithCancellation(deps, session) {
1512
- const { cancellationSignal, iterator } = session;
1711
+ const { cancellation, iterator } = session;
1712
+ const cancellationSignal = cancellation.signal;
1513
1713
  const cancelIterator = () => {
1514
1714
  void iterator.return?.();
1515
1715
  };
@@ -1518,10 +1718,9 @@ async function pumpGatewayStreamWithCancellation(deps, session) {
1518
1718
  await pumpGatewayStream(session);
1519
1719
  }
1520
1720
  catch (error) {
1521
- emitGatewayStreamFailureDiagnostic(session.ctx, deps, error, session.runId);
1522
- if (!session.cancellationSignal.aborted) {
1523
- reportGatewayTurnFailure(session.ctx, deps, session.runId, gatewayStreamFailureCode(error));
1524
- }
1721
+ const turnFailureRecorded = !cancellationSignal.aborted &&
1722
+ reportGatewayTurnFailure(session.ctx, deps, session.runId, gatewayStreamFailureCode(error), error);
1723
+ emitGatewayStreamFailureDiagnostic(session.ctx, deps, error, session.runId, turnFailureRecorded);
1525
1724
  settleGatewayStreamError(session);
1526
1725
  }
1527
1726
  finally {
@@ -1529,7 +1728,7 @@ async function pumpGatewayStreamWithCancellation(deps, session) {
1529
1728
  }
1530
1729
  }
1531
1730
  function createGatewayStreamSession(ctx, dispatch, iterator) {
1532
- const { deps, modelAlias, request, runId, cancellationSignal, promptTokenReservation } = dispatch;
1731
+ const { deps, modelAlias, request, runId, cancellation, promptTokenReservation } = dispatch;
1533
1732
  return {
1534
1733
  ctx,
1535
1734
  deps,
@@ -1538,7 +1737,7 @@ function createGatewayStreamSession(ctx, dispatch, iterator) {
1538
1737
  modelId: modelAlias,
1539
1738
  request,
1540
1739
  runId,
1541
- cancellationSignal,
1740
+ cancellation,
1542
1741
  iterator,
1543
1742
  promptTokenReservation,
1544
1743
  metrics: {
@@ -1557,7 +1756,8 @@ function beginGatewayStream(session) {
1557
1756
  "Cache-Control": "no-store",
1558
1757
  Connection: "keep-alive",
1559
1758
  });
1560
- if (!writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { role: "assistant" }, null))) {
1759
+ const opened = writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { role: "assistant" }, null), session.cancellation.transport);
1760
+ if (!opened) {
1561
1761
  ctx.res.destroy();
1562
1762
  recordSessionOutcome(session, "cancelled");
1563
1763
  return false;
@@ -1565,15 +1765,16 @@ function beginGatewayStream(session) {
1565
1765
  return true;
1566
1766
  }
1567
1767
  function recordSessionOutcome(session, outcome) {
1568
- const { ctx, deps, runId, metrics } = session;
1569
- recordGatewayOutcome(ctx, deps, runId, outcome, metrics.completionTokens, metrics.outputBytes);
1768
+ const { ctx, deps, runId, cancellation, metrics } = session;
1769
+ recordGatewayOutcome(ctx, deps, runId, cancellation, outcome, metrics.completionTokens, metrics.outputBytes);
1570
1770
  }
1571
1771
  function writeSessionTerminal(session, finishReason) {
1572
- const { ctx, id, created, modelId, metrics } = session;
1573
- writeStreamTerminal(ctx, id, created, modelId, finishReason, metrics.promptTokens, metrics.completionTokens);
1772
+ const { ctx, id, created, modelId, metrics, cancellation } = session;
1773
+ writeStreamTerminal({ ctx, id, created, modelId, transport: cancellation.transport }, finishReason, metrics.promptTokens, metrics.completionTokens);
1574
1774
  }
1575
1775
  async function pumpGatewayStream(session) {
1576
- const { cancellationSignal, iterator } = session;
1776
+ const { cancellation, iterator } = session;
1777
+ const cancellationSignal = cancellation.signal;
1577
1778
  for (;;) {
1578
1779
  if (isGatewayRequestCancelled(cancellationSignal)) {
1579
1780
  await iterator.return?.();
@@ -1613,7 +1814,8 @@ async function streamGatewayDelta(session, token) {
1613
1814
  writeSessionTerminal(session, "length");
1614
1815
  return false;
1615
1816
  }
1616
- if (!writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { content: token }, null))) {
1817
+ const wrote = writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { content: token }, null), session.cancellation.transport);
1818
+ if (!wrote) {
1617
1819
  ctx.res.destroy();
1618
1820
  await iterator.return?.();
1619
1821
  recordSessionOutcome(session, "cancelled");
@@ -1635,7 +1837,7 @@ async function streamGatewayResponse(session, response) {
1635
1837
  return;
1636
1838
  }
1637
1839
  if (response.toolCalls.length > 0) {
1638
- const wrote = writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { tool_calls: openAiToolCalls(response.toolCalls) }, null));
1840
+ const wrote = writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { tool_calls: openAiToolCalls(response.toolCalls) }, null), session.cancellation.transport);
1639
1841
  if (!wrote) {
1640
1842
  ctx.res.destroy();
1641
1843
  await iterator.return?.();
@@ -1688,7 +1890,7 @@ function logGatewayCompletionUsage(ctx, runId, metrics, source, promptSettlement
1688
1890
  }));
1689
1891
  }
1690
1892
  function settleGatewayStreamError(session) {
1691
- if (!session.cancellationSignal.aborted) {
1893
+ if (!session.cancellation.signal.aborted) {
1692
1894
  recordSessionOutcome(session, "failed");
1693
1895
  writeSessionTerminal(session, "error");
1694
1896
  }
@@ -1699,7 +1901,8 @@ function settleGatewayStreamError(session) {
1699
1901
  function isGatewayRequestCancelled(signal) {
1700
1902
  return signal.aborted;
1701
1903
  }
1702
- function beginBufferedOpenAiStream(ctx, modelId) {
1904
+ /** Returns undefined when the opening SSE frame could not be delivered; the response is destroyed. */
1905
+ function beginBufferedOpenAiStream(ctx, modelId, transport) {
1703
1906
  const id = `chatcmpl-${randomUUID()}`;
1704
1907
  const created = Math.floor(Date.now() / 1000);
1705
1908
  ctx.res.writeHead(200, {
@@ -1707,19 +1910,26 @@ function beginBufferedOpenAiStream(ctx, modelId) {
1707
1910
  "Cache-Control": "no-store",
1708
1911
  Connection: "keep-alive",
1709
1912
  });
1710
- if (!writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { role: "assistant" }, null))) {
1913
+ const opened = writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, { role: "assistant" }, null), transport);
1914
+ if (!opened) {
1711
1915
  ctx.res.destroy();
1916
+ return undefined;
1712
1917
  }
1713
1918
  return {
1714
1919
  ctx,
1715
1920
  id,
1716
1921
  created,
1717
1922
  modelId,
1718
- stopHeartbeat: startSseHeartbeat(ctx.res, BUFFERED_STREAM_HEARTBEAT_MS),
1923
+ transport,
1924
+ stopHeartbeat: startSseHeartbeat(ctx.res, BUFFERED_STREAM_HEARTBEAT_MS, undefined, {
1925
+ controller: transport.backpressure,
1926
+ onBackpressure: transport.onBackpressure,
1927
+ correlationId: transport.correlationId,
1928
+ }),
1719
1929
  };
1720
1930
  }
1721
1931
  function completeBufferedOpenAiStream(session, response) {
1722
- const { ctx, id, created, modelId, stopHeartbeat } = session;
1932
+ const { ctx, id, created, modelId, transport, stopHeartbeat } = session;
1723
1933
  stopHeartbeat();
1724
1934
  if (response.content.length > 0 || response.toolCalls.length > 0) {
1725
1935
  const wrote = writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, {
@@ -1727,32 +1937,35 @@ function completeBufferedOpenAiStream(session, response) {
1727
1937
  ...(response.toolCalls.length === 0
1728
1938
  ? {}
1729
1939
  : { tool_calls: openAiToolCalls(response.toolCalls) }),
1730
- }, null));
1940
+ }, null), transport);
1731
1941
  if (!wrote) {
1732
1942
  ctx.res.destroy();
1733
1943
  return STREAMING;
1734
1944
  }
1735
1945
  }
1736
- writeStreamTerminal(ctx, id, created, modelId, response.finishReason, response.usage.promptTokens, response.usage.completionTokens);
1946
+ writeStreamTerminal(session, response.finishReason, response.usage.promptTokens, response.usage.completionTokens);
1737
1947
  return STREAMING;
1738
1948
  }
1739
1949
  function settleBufferedOpenAiStreamError(session, finishReason) {
1740
- const { ctx, id, created, modelId, stopHeartbeat } = session;
1741
- stopHeartbeat();
1742
- writeStreamTerminal(ctx, id, created, modelId, finishReason, 0, 0);
1950
+ session.stopHeartbeat();
1951
+ writeStreamTerminal(session, finishReason, 0, 0);
1743
1952
  return STREAMING;
1744
1953
  }
1745
- function bufferedOpenAiStream(ctx, modelId, response) {
1746
- return completeBufferedOpenAiStream(beginBufferedOpenAiStream(ctx, modelId), response);
1954
+ function bufferedOpenAiStream(ctx, deps, modelId, response) {
1955
+ const session = beginBufferedOpenAiStream(ctx, modelId, sidecarSseTransport(ctx, deps));
1956
+ return session === undefined ? STREAMING : completeBufferedOpenAiStream(session, response);
1747
1957
  }
1748
- function writeOpenAiSse(ctx, payload) {
1749
- if (!ctx.res.writableEnded && !ctx.res.destroyed) {
1750
- return ctx.res.write(`data: ${JSON.stringify(payload)}\n\n`);
1751
- }
1752
- return false;
1958
+ // False means the frame did not reach the client: either the response was already gone (the
1959
+ // client's own `close` is the abort source that names it) or the shared path killed the stream for
1960
+ // backpressure (its controller is). A caller that sees false stops producing; it never resumes.
1961
+ function writeOpenAiSse(ctx, payload, transport) {
1962
+ if (ctx.res.writableEnded || ctx.res.destroyed)
1963
+ return false;
1964
+ return writeOrDestroy(ctx.res, `data: ${JSON.stringify(payload)}\n\n`, transport.backpressure, transport.onBackpressure, transport.correlationId);
1753
1965
  }
1754
- function writeStreamTerminal(ctx, id, created, modelId, finishReason, promptTokens, completionTokens) {
1755
- writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, {}, finishReason));
1966
+ function writeStreamTerminal(stream, finishReason, promptTokens, completionTokens) {
1967
+ const { ctx, id, created, modelId, transport } = stream;
1968
+ writeOpenAiSse(ctx, openAiStreamChunk(id, created, modelId, {}, finishReason), transport);
1756
1969
  writeOpenAiSse(ctx, {
1757
1970
  id,
1758
1971
  object: "chat.completion.chunk",
@@ -1764,7 +1977,7 @@ function writeStreamTerminal(ctx, id, created, modelId, finishReason, promptToke
1764
1977
  completion_tokens: completionTokens,
1765
1978
  total_tokens: promptTokens + completionTokens,
1766
1979
  },
1767
- });
1980
+ }, transport);
1768
1981
  if (!ctx.res.writableEnded && !ctx.res.destroyed)
1769
1982
  ctx.res.end("data: [DONE]\n\n");
1770
1983
  }
@@ -1788,20 +2001,95 @@ function openAiStreamChunk(id, created, model, delta, finishReason) {
1788
2001
  */
1789
2002
  function gatewayReadinessProjection(ctx, deps) {
1790
2003
  const result = resolveGatewayProfile(deps).result;
1791
- if (result.status === "available" &&
1792
- result.runMetadata.maxPromptTokens >= CODING_WORKBENCH_MINIMUM_CODING_CONTEXT_PROMPT_TOKENS)
2004
+ if (codingContextFits(result))
1793
2005
  return result;
1794
- const reason = result.status === "available" ? "model-context-window-insufficient" : result.reason;
1795
- if (reason !== "model-context-window-insufficient" &&
1796
- reason !== "no-tool-calling" &&
1797
- reason !== "tool-calling-unverified")
2006
+ const shortfall = result.status === "available" ? "model-context-window-insufficient" : result.reason;
2007
+ if (shortfall !== "model-context-window-insufficient" &&
2008
+ shortfall !== "no-tool-calling" &&
2009
+ shortfall !== "tool-calling-unverified")
1798
2010
  return result;
2011
+ // #3591 (1.1.7): while the automatic probe is still running against a slow gateway the shortfall
2012
+ // is not a verdict. The profile says so, and the Workbench re-reads it instead of refusing.
2013
+ const pending = readinessProbePending(deps, result, shortfall);
2014
+ const reason = pending
2015
+ ? "model-verification-pending"
2016
+ : shortfall;
2017
+ logReadinessShortfall(ctx, result, reason, pending);
2018
+ // An open verification replaces the stored shortfall for an unavailable projection too (an
2019
+ // unverified tool-calling proof whose probe is still running), or the Workbench would stop
2020
+ // reading and keep the refusal until an unrelated refresh.
2021
+ if (result.status === "available" || pending)
2022
+ return { status: "unavailable", reason };
2023
+ return result;
2024
+ }
2025
+ // One owner for the rule a coding run's prompt needs: the readiness projection of the default model
2026
+ // and the start of a run with a model chosen in the picker both read it (#3603).
2027
+ function codingContextFits(result) {
2028
+ return (result.status === "available" &&
2029
+ result.runMetadata.maxPromptTokens >= CODING_WORKBENCH_MINIMUM_CODING_CONTEXT_PROMPT_TOKENS);
2030
+ }
2031
+ /**
2032
+ * Why an available coding profile cannot hold a coding run's prompt, or undefined when it can
2033
+ * (#3603). While the automatic probe that could raise the window is still running, the shortfall
2034
+ * is not a verdict yet; the start then names the pending verification instead.
2035
+ */
2036
+ export function codingContextShortfall(config, result) {
2037
+ if (codingContextFits(result))
2038
+ return undefined;
2039
+ return config !== undefined && isCodingWorkbenchProbePending(config, result.modelAlias)
2040
+ ? "model-verification-pending"
2041
+ : "model-context-window-insufficient";
2042
+ }
2043
+ /**
2044
+ * The profile a coding run starts with, for the model chosen in the picker or the default one. A
2045
+ * model the gateway does not admit right now is a typed refusal that names the sidecar's reason
2046
+ * (#3565 Observation 17), never a bare Error the orchestrator can only report as
2047
+ * `authority-resolution-failed`; so is a model whose window cannot hold the run's prompt (#3603).
2048
+ */
2049
+ export function admitCodingRunModel(config, modelId, reasoningEffort) {
2050
+ const resolved = resolveCodingSafeSidecarGatewayProfile(config, {
2051
+ ...(modelId === undefined ? {} : { modelId }),
2052
+ });
2053
+ if (resolved.status !== "available" || config === undefined) {
2054
+ throw new CodingRuntimeLaunchRejectedError("model-unavailable", false, resolved.status === "available" ? "missing-config" : resolved.reason);
2055
+ }
2056
+ const contextShortfall = codingContextShortfall(config, resolved);
2057
+ if (contextShortfall !== undefined) {
2058
+ throw new CodingRuntimeLaunchRejectedError("model-unavailable", false, contextShortfall);
2059
+ }
2060
+ return {
2061
+ profileId: resolved.modelAlias,
2062
+ ...admittedReasoningEffort(config, resolved.modelAlias, reasoningEffort),
2063
+ };
2064
+ }
2065
+ function admittedReasoningEffort(config, modelAlias, reasoningEffort) {
2066
+ if (reasoningEffort === undefined)
2067
+ return {};
2068
+ const efforts = findConfiguredCapability(config, modelAlias)?.reasoningEfforts;
2069
+ if (efforts?.includes(reasoningEffort) !== true) {
2070
+ throw new CodingRuntimeLaunchRejectedError("model-unavailable", false, "reasoning-effort-unavailable");
2071
+ }
2072
+ return { reasoningEffort };
2073
+ }
2074
+ function readinessProbePending(deps, result, shortfall) {
2075
+ if (shortfall === "no-tool-calling")
2076
+ return false;
2077
+ const config = currentGatewayConfig(deps);
2078
+ if (config === undefined)
2079
+ return false;
2080
+ // An unavailable projection (an unverified tool-calling proof) names no model: its verification
2081
+ // is open while any model the Workbench could still elect has an open probe.
2082
+ return result.status === "available"
2083
+ ? isCodingWorkbenchProbePending(config, result.modelAlias)
2084
+ : isAnyCodingWorkbenchProbePending(config);
2085
+ }
2086
+ function logReadinessShortfall(ctx, result, reason, pending) {
1799
2087
  getServerLogger().warn(activityLogEvent(CODING_SIDECAR_GATEWAY_READINESS_INSUFFICIENT_OPERATION, {
1800
2088
  correlationId: correlationIdOrUnknown(ctx.correlationId),
1801
2089
  errorKind: "unavailable",
1802
2090
  }, {
1803
2091
  reason,
1804
- probeMode: "passive",
2092
+ probeMode: pending ? "pending" : "passive",
1805
2093
  ...(result.status === "available"
1806
2094
  ? {
1807
2095
  maxPromptTokens: result.runMetadata.maxPromptTokens,
@@ -1811,9 +2099,6 @@ function gatewayReadinessProjection(ctx, deps) {
1811
2099
  completeness: "complete",
1812
2100
  loss: "none",
1813
2101
  }));
1814
- return result.status === "available"
1815
- ? { status: "unavailable", reason: "model-context-window-insufficient" }
1816
- : result;
1817
2102
  }
1818
2103
  export async function handleCodingSidecarGatewayProfile(ctx, deps) {
1819
2104
  // What Keiko can determine itself it determines itself: a model whose gateway declared no token
@@ -1822,10 +2107,25 @@ export async function handleCodingSidecarGatewayProfile(ctx, deps) {
1822
2107
  // Only while the gateway is the usable source: a subscription source, a disabled policy or a
1823
2108
  // missing configuration must never cause a paid provider probe (ADR-0124 D5).
1824
2109
  if (elected.status === "available" || elected.reason === "tool-calling-unverified") {
1825
- await ensureCodingWorkbenchContextWindows(deps, elected.status === "available" ? elected.modelAlias : undefined, ctx.correlationId);
2110
+ // #3591 (1.1.7): the browser reads this profile with a 15 s deadline while a probe against a
2111
+ // slow gateway may take minutes. Wait a bounded moment for the elected model's proof; past it,
2112
+ // answer with the projection (`model-verification-pending` while the probe runs) and let
2113
+ // the Workbench read again — never leave the read hanging until the browser gives up.
2114
+ await Promise.race([
2115
+ ensureCodingWorkbenchContextWindows(deps, elected.status === "available" ? elected.modelAlias : undefined, ctx.correlationId),
2116
+ boundedProfileWait(),
2117
+ ]);
1826
2118
  }
1827
2119
  return { status: 200, body: gatewayReadinessProjection(ctx, deps) };
1828
2120
  }
2121
+ // Under the browser's 15 s read deadline for this profile (keiko-ui `DEFAULT_READ_TIMEOUT_MS`).
2122
+ export const PROFILE_PROBE_WAIT_MS = 8_000;
2123
+ function boundedProfileWait() {
2124
+ return new Promise((resolve) => {
2125
+ const timer = setTimeout(resolve, PROFILE_PROBE_WAIT_MS);
2126
+ timer.unref();
2127
+ });
2128
+ }
1829
2129
  function upstreamGatewayStreamingSupported(deps, advertisedSupport) {
1830
2130
  return advertisedSupport || deps.codingSidecarGatewayChatStreamFactory !== undefined;
1831
2131
  }
@@ -1837,7 +2137,7 @@ function rejectUnmanagedGatewayToolContract(ctx, deps, parsed, authentication) {
1837
2137
  const declaresTools = parsed.tools !== undefined && parsed.tools.length > 0;
1838
2138
  if (!declaresTools || isExactManagedToolSet(parsed.tools))
1839
2139
  return undefined;
1840
- emitGatewayToolContractDiagnostic(ctx, deps, authentication.runId, parsed.tools);
2140
+ emitGatewayToolContractDiagnostic(ctx, deps, authentication.runId, parsed);
1841
2141
  return forbiddenGatewayRequest();
1842
2142
  }
1843
2143
  function runtimeGatewayAdmissionResponse(ctx, deps, parsed, authentication, modelAlias) {
@@ -1851,7 +2151,7 @@ function runtimeGatewayAdmissionResponse(ctx, deps, parsed, authentication, mode
1851
2151
  function authenticatedGatewayAdmission(ctx, deps, parsed, authentication, modelAlias) {
1852
2152
  const registry = gatewayReadinessRegistry(deps);
1853
2153
  if (!isAdmittedManagedToolSet(parsed.tools, registry, authentication.runId)) {
1854
- emitGatewayToolContractDiagnostic(ctx, deps, authentication.runId, parsed.tools);
2154
+ emitGatewayToolContractDiagnostic(ctx, deps, authentication.runId, parsed);
1855
2155
  return { kind: "handled", result: forbiddenGatewayRequest() };
1856
2156
  }
1857
2157
  if (isExactManagedToolSet(parsed.tools) &&
@@ -1859,7 +2159,7 @@ function authenticatedGatewayAdmission(ctx, deps, parsed, authentication, modelA
1859
2159
  registry?.claim(authentication.runId) === true) {
1860
2160
  return {
1861
2161
  kind: "handled",
1862
- result: fixedReadinessResponse(ctx, modelAlias, parsed.stream === true),
2162
+ result: fixedReadinessResponse(ctx, deps, modelAlias, parsed.stream === true),
1863
2163
  };
1864
2164
  }
1865
2165
  if (isExactManagedToolSet(parsed.tools))
@@ -1893,11 +2193,13 @@ function logChatRequestRejection(ctx, deps, runId, validationError, observed) {
1893
2193
  ? {
1894
2194
  estimatedPromptTokens: observed.estimatedPromptTokens,
1895
2195
  maxPromptTokens: observed.bounds.maxPromptTokens,
2196
+ admissiblePromptTokens: admissiblePromptTokens(observed.bounds),
1896
2197
  inputMessageCount: observed.parsed.messages.length,
1897
2198
  maxInputMessages: observed.bounds.maxInputMessages,
1898
2199
  }
1899
2200
  : undefined;
1900
2201
  logGatewayRejection(ctx, runId, validationError.status, reason, boundedEvidence);
2202
+ refuseReadinessChallenge(deps, runId, observed?.parsed);
1901
2203
  reportGatewayTurnFailure(ctx, deps, runId, "turn-rejected");
1902
2204
  }
1903
2205
  async function readValidatedChatRequest(ctx, deps, resolved, authentication) {
@@ -1933,13 +2235,41 @@ async function runHandleCodingSidecarGatewayChatCompletions(ctx, deps) {
1933
2235
  return admission.result;
1934
2236
  return executeBudgetedGatewayChat(ctx, deps, resolved, parsed, authentication, estimatedPromptTokens, {
1935
2237
  modelAlias: resolved.result.modelAlias,
1936
- maxOutputTokens: resolved.result.runMetadata.maxOutputTokens,
2238
+ maxOutputTokens: admittedOutputTokens(resolved.result.runMetadata, estimatedPromptTokens),
1937
2239
  upstreamStreamingSupported: upstreamGatewayStreamingSupported(deps, resolved.result.supportsStreaming),
1938
2240
  ...(authentication.reasoningEffort === undefined
1939
2241
  ? {}
1940
2242
  : { reasoningEffort: authentication.reasoningEffort }),
1941
2243
  });
1942
2244
  }
2245
+ // #3591 (1.1.7): the run's output reserve (8k for an undeclared limit) is a reserve against the
2246
+ // whole window, and `maxPromptTokens` IS that whole window. A prompt close to the window would
2247
+ // leave the provider a request larger than its window, so the allowance actually sent is what
2248
+ // remains after the prompt and the safety margin — and the least allowance that still lets the
2249
+ // model answer (and report an exhausted budget instead of failing silently) is reserved at
2250
+ // admission, never added on top of a prompt that already fills the window.
2251
+ export const MINIMUM_ADMITTED_OUTPUT_TOKENS = 512;
2252
+ // What a prompt must leave free of `maxPromptTokens`: the window's safety margin plus the minimum
2253
+ // allowance (or the whole reserve, when that is smaller).
2254
+ function reservedWindowTokens(bounds) {
2255
+ return (safetyMarginTokensFor(bounds.maxPromptTokens, bounds.maxOutputTokens) +
2256
+ Math.min(MINIMUM_ADMITTED_OUTPUT_TOKENS, bounds.maxOutputTokens));
2257
+ }
2258
+ // Review of #3591 (P1): admission used to check the prompt against `maxPromptTokens` alone while
2259
+ // the allowance floored at 512, so a prompt just under the window was sent WITH 512 output tokens
2260
+ // — past the window the probe had proven, and the provider refused the turn. The prompt must
2261
+ // leave the margin and the minimum allowance free, or the turn is refused before any call.
2262
+ export function admissiblePromptTokens(bounds) {
2263
+ return bounds.maxPromptTokens - reservedWindowTokens(bounds);
2264
+ }
2265
+ // The allowance an ADMITTED turn sends: the run's reserve, shrunk to what the prompt leaves after
2266
+ // the safety margin. Admission (`admissiblePromptTokens`) guarantees at least the minimum.
2267
+ export function admittedOutputTokens(bounds, estimatedPromptTokens) {
2268
+ const remaining = bounds.maxPromptTokens -
2269
+ estimatedPromptTokens -
2270
+ safetyMarginTokensFor(bounds.maxPromptTokens, bounds.maxOutputTokens);
2271
+ return Math.min(bounds.maxOutputTokens, remaining);
2272
+ }
1943
2273
  function logValidatedRequestBounds(ctx, runId, request, bounds, estimatedPromptTokens) {
1944
2274
  getServerLogger().info(activityLogEvent(CODING_SIDECAR_GATEWAY_REQUEST_VALIDATED_OPERATION, {
1945
2275
  correlationId: correlationIdOrUnknown(ctx.correlationId),
@@ -1949,6 +2279,7 @@ function logValidatedRequestBounds(ctx, runId, request, bounds, estimatedPromptT
1949
2279
  maxRequestBytes: bounds.maxRequestBytes,
1950
2280
  maxPromptTokens: bounds.maxPromptTokens,
1951
2281
  estimatedPromptTokens,
2282
+ maxOutputTokens: admittedOutputTokens(bounds, estimatedPromptTokens),
1952
2283
  inputMessageCount: request.messages.length,
1953
2284
  completeness: "complete",
1954
2285
  loss: "none",
@@ -1968,7 +2299,7 @@ function executeBudgetedGatewayChat(ctx, deps, binding, parsed, authentication,
1968
2299
  offerLifetimeMs: opencodeGatewayOfferLifetimeMs(codingSidecarGatewayRequestDeadlineMs(binding.config, profile.modelAlias)),
1969
2300
  });
1970
2301
  }
1971
- function fixedReadinessResponse(ctx, modelId, stream) {
2302
+ function fixedReadinessResponse(ctx, deps, modelId, stream) {
1972
2303
  const response = {
1973
2304
  modelId,
1974
2305
  content: "",
@@ -1983,5 +2314,7 @@ function fixedReadinessResponse(ctx, modelId, stream) {
1983
2314
  costClass: "low",
1984
2315
  },
1985
2316
  };
1986
- return stream ? bufferedOpenAiStream(ctx, modelId, response) : openAiResponse(modelId, response);
2317
+ return stream
2318
+ ? bufferedOpenAiStream(ctx, deps, modelId, response)
2319
+ : openAiResponse(modelId, response);
1987
2320
  }