@cbartens/bkit 6.0.0 → 6.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (375) hide show
  1. package/.bkit/README.md +154 -151
  2. package/.bkit/core/.agents/context.md +95 -29
  3. package/.bkit/core/.agents/docs.md +17 -15
  4. package/.bkit/core/.agents/plugins/bkit/hooks/bkit-gateguard.cjs +1 -0
  5. package/.bkit/core/.agents/plugins/bkit/hooks/bkit-guardrails.cjs +6 -5
  6. package/.bkit/core/.agents/plugins/bkit/hooks/bkit-readguard.cjs +1 -0
  7. package/.bkit/core/.agents/plugins/bkit/hooks/bkit-rules.cjs +65 -0
  8. package/.bkit/core/.agents/rules.md +108 -107
  9. package/.bkit/core/.agents/skills.md +63 -15
  10. package/.bkit/core/.bkit/README.md +154 -151
  11. package/.bkit/core/.bkit/authoring.md +714 -84
  12. package/.bkit/core/.bkit/cli.md +407 -0
  13. package/.bkit/core/.bkit/cloud.md +281 -193
  14. package/.bkit/core/.bkit/comparison.md +43 -41
  15. package/.bkit/core/.bkit/install.md +154 -84
  16. package/.bkit/core/.bkit/jsonl.md +193 -130
  17. package/.bkit/core/.bkit/pipeline.md +114 -114
  18. package/.bkit/core/.bkit/prompts.md +272 -165
  19. package/.bkit/core/.bkit/providers.md +457 -271
  20. package/.bkit/core/.bkit/router.md +69 -0
  21. package/.bkit/core/.bkit/templates.md +529 -256
  22. package/.bkit/core/.bkit/trackers.md +80 -52
  23. package/.bkit/core/.bkit/workflows.md +589 -504
  24. package/.bkit/core/.claude/hooks/bkit-gateguard.cjs +2 -2
  25. package/.bkit/core/.claude/hooks/bkit-guardrails.cjs +3 -3
  26. package/.bkit/core/.claude/hooks/bkit-preflight.sh +45 -14
  27. package/.bkit/core/.codex/hooks/bkit-guardrails.cjs +3 -3
  28. package/.bkit/core/.codex/hooks/bkit-rules.cjs +57 -0
  29. package/.bkit/core/.cursor/hooks/bkit-guardrails.cjs +7 -7
  30. package/.bkit/core/.cursor/hooks/bkit-preflight.sh +47 -12
  31. package/.bkit/core/.opencode/plugin/bkit-rules.js +89 -0
  32. package/.bkit/core/AGENTS.md +5 -73
  33. package/.bkit/core/docs/templates/bkit-audit-accessibility.md +127 -105
  34. package/.bkit/core/docs/templates/bkit-audit-adoption.md +104 -70
  35. package/.bkit/core/docs/templates/bkit-audit-agents.md +94 -72
  36. package/.bkit/core/docs/templates/bkit-audit-aio-seo.md +99 -78
  37. package/.bkit/core/docs/templates/bkit-audit-analytics.md +140 -82
  38. package/.bkit/core/docs/templates/bkit-audit-annotations.md +98 -72
  39. package/.bkit/core/docs/templates/bkit-audit-app-performance.md +121 -96
  40. package/.bkit/core/docs/templates/bkit-audit-app-store.md +157 -76
  41. package/.bkit/core/docs/templates/bkit-audit-architecture.md +113 -71
  42. package/.bkit/core/docs/templates/bkit-audit-ci-performance.md +75 -72
  43. package/.bkit/core/docs/templates/bkit-audit-config.md +122 -76
  44. package/.bkit/core/docs/templates/bkit-audit-cost.md +85 -74
  45. package/.bkit/core/docs/templates/bkit-audit-coverage.md +75 -68
  46. package/.bkit/core/docs/templates/bkit-audit-database.md +101 -70
  47. package/.bkit/core/docs/templates/bkit-audit-dead-code.md +164 -83
  48. package/.bkit/core/docs/templates/bkit-audit-deferrals.md +115 -86
  49. package/.bkit/core/docs/templates/bkit-audit-dependencies.md +264 -88
  50. package/.bkit/core/docs/templates/bkit-audit-documentation.md +118 -93
  51. package/.bkit/core/docs/templates/bkit-audit-duplication.md +139 -86
  52. package/.bkit/core/docs/templates/bkit-audit-failures.md +110 -81
  53. package/.bkit/core/docs/templates/bkit-audit-flakes.md +99 -74
  54. package/.bkit/core/docs/templates/bkit-audit-gates.md +105 -77
  55. package/.bkit/core/docs/templates/bkit-audit-issues.md +104 -80
  56. package/.bkit/core/docs/templates/bkit-audit-llm.md +131 -58
  57. package/.bkit/core/docs/templates/bkit-audit-logs.md +123 -100
  58. package/.bkit/core/docs/templates/bkit-audit-models.md +98 -53
  59. package/.bkit/core/docs/templates/bkit-audit-privacy.md +56 -52
  60. package/.bkit/core/docs/templates/bkit-audit-prompts.md +63 -54
  61. package/.bkit/core/docs/templates/bkit-audit-red-team.md +124 -68
  62. package/.bkit/core/docs/templates/bkit-audit-reliability.md +95 -70
  63. package/.bkit/core/docs/templates/bkit-audit-repo.md +169 -73
  64. package/.bkit/core/docs/templates/bkit-audit-security.md +206 -96
  65. package/.bkit/core/docs/templates/bkit-audit-slop.md +77 -70
  66. package/.bkit/core/docs/templates/bkit-audit-testing.md +162 -114
  67. package/.bkit/core/docs/templates/bkit-audit-usability.md +113 -71
  68. package/.bkit/core/docs/templates/bkit-plan.md +138 -132
  69. package/.bkit/core/engine/checks/after/auth.ts +40 -37
  70. package/.bkit/core/engine/checks/after/ci-yml.ts +65 -2
  71. package/.bkit/core/engine/checks/after/index.ts +12 -2
  72. package/.bkit/core/engine/checks/after/tools.ts +80 -33
  73. package/.bkit/core/engine/checks/before/context-fetch.ts +227 -33
  74. package/.bkit/core/engine/checks/before/env.ts +1 -1
  75. package/.bkit/core/engine/checks/before/index.ts +7 -0
  76. package/.bkit/core/engine/checks/before/repo.ts +40 -0
  77. package/.bkit/core/engine/checks/status-holes.ts +916 -123
  78. package/.bkit/core/engine/checks/tips/catalogue.ts +68 -17
  79. package/.bkit/core/engine/checks/tips/detect.ts +32 -0
  80. package/.bkit/core/engine/checks/tips/index.ts +47 -6
  81. package/.bkit/core/engine/cli.ts +238 -84
  82. package/.bkit/core/engine/commands/analysis.ts +128 -76
  83. package/.bkit/core/engine/commands/ci/ci.ts +363 -29
  84. package/.bkit/core/engine/commands/ci/pins.ts +172 -0
  85. package/.bkit/core/engine/commands/ci/push-gate.ts +223 -0
  86. package/.bkit/core/engine/commands/install/uninstall.ts +21 -27
  87. package/.bkit/core/engine/commands/install/upgrade.ts +32 -15
  88. package/.bkit/core/engine/commands/land.ts +126 -11
  89. package/.bkit/core/engine/commands/lib.ts +303 -6
  90. package/.bkit/core/engine/commands/registry.ts +6 -2
  91. package/.bkit/core/engine/commands/setup/setup.ts +1 -0
  92. package/.bkit/core/engine/commands/stats.ts +91 -35
  93. package/.bkit/core/engine/commands/workflow/branches.ts +1477 -117
  94. package/.bkit/core/engine/commands/workflow/challenge.ts +114 -39
  95. package/.bkit/core/engine/commands/workflow/completeness.ts +109 -0
  96. package/.bkit/core/engine/commands/workflow/context.ts +433 -55
  97. package/.bkit/core/engine/commands/workflow/counters.ts +244 -39
  98. package/.bkit/core/engine/commands/workflow/decisions.ts +61 -17
  99. package/.bkit/core/engine/commands/workflow/disk.ts +237 -55
  100. package/.bkit/core/engine/commands/workflow/fids.ts +19 -1
  101. package/.bkit/core/engine/commands/workflow/issues.ts +6 -0
  102. package/.bkit/core/engine/commands/workflow/ledgers.ts +458 -18
  103. package/.bkit/core/engine/commands/workflow/locks.ts +2 -0
  104. package/.bkit/core/engine/commands/workflow/plan-blocks.ts +124 -15
  105. package/.bkit/core/engine/commands/workflow/plans.ts +257 -0
  106. package/.bkit/core/engine/commands/workflow/prs.ts +12 -3
  107. package/.bkit/core/engine/commands/workflow/repo.ts +8 -0
  108. package/.bkit/core/engine/commands/workflow/roadmap.ts +56 -15
  109. package/.bkit/core/engine/commands/workflow/run-facts.ts +22 -31
  110. package/.bkit/core/engine/commands/workflow/runs.ts +652 -34
  111. package/.bkit/core/engine/commands/workflow/runtime.ts +3 -0
  112. package/.bkit/core/engine/commands/workflow/skills.ts +1 -0
  113. package/.bkit/core/engine/commands/workflow/status.ts +200 -17
  114. package/.bkit/core/engine/commands/workflow/step-table.ts +75 -33
  115. package/.bkit/core/engine/commands/workflow/steps.ts +635 -24
  116. package/.bkit/core/engine/commands/workflow/summary-findings.ts +177 -0
  117. package/.bkit/core/engine/commands/workflow/summary.ts +192 -46
  118. package/.bkit/core/engine/commands/workflow/tasks.ts +409 -0
  119. package/.bkit/core/engine/commands/workflow/test-blocks.ts +254 -0
  120. package/.bkit/core/engine/commands/workflow/tests-changed.ts +201 -0
  121. package/.bkit/core/engine/commands/workflow/tests.ts +1285 -356
  122. package/.bkit/core/engine/commands/workflow/workflows.ts +228 -43
  123. package/.bkit/core/engine/commands/workflow/worktrees.ts +389 -47
  124. package/.bkit/core/engine/install/cloud-env.ts +251 -121
  125. package/.bkit/core/engine/install/context-ci-workflow.ts +102 -0
  126. package/.bkit/core/engine/install/manifest.ts +60 -12
  127. package/.bkit/core/engine/install/modules/antigravity.ts +3 -2
  128. package/.bkit/core/engine/install/modules/cloud.ts +3 -3
  129. package/.bkit/core/engine/install/modules/codex.ts +15 -2
  130. package/.bkit/core/engine/install/modules/core.ts +956 -52
  131. package/.bkit/core/engine/install/modules/cursor.ts +6 -6
  132. package/.bkit/core/engine/install/modules/opencode.ts +16 -4
  133. package/.bkit/core/engine/install/template-notes.ts +130 -0
  134. package/.bkit/core/engine/install/version-file.ts +3 -2
  135. package/.bkit/core/engine/install/write.ts +5 -0
  136. package/.bkit/core/engine/ledgers/compact.ts +124 -0
  137. package/.bkit/core/engine/ledgers/journal.ts +291 -0
  138. package/.bkit/core/engine/ledgers/status.ts +160 -9
  139. package/.bkit/core/engine/ledgers/telemetry.ts +158 -21
  140. package/.bkit/core/engine/render/hosts.ts +114 -14
  141. package/.bkit/core/engine/shared/bkit-latest.ts +26 -4
  142. package/.bkit/core/engine/shared/dep-updater.ts +60 -0
  143. package/.bkit/core/engine/shared/egress.ts +1 -1
  144. package/.bkit/core/engine/shared/package-manager.ts +9 -0
  145. package/.bkit/core/engine/shared/settings.ts +92 -17
  146. package/.bkit/core/engine/shared/timeouts.ts +2 -0
  147. package/.bkit/core/engine/tools/cli-doc.ts +111 -0
  148. package/.bkit/core/engine/tools/grammar-density.ts +1 -1
  149. package/.bkit/core/engine/tools/instructions-sources.ts +15 -17
  150. package/.bkit/core/engine/tools/instructions.ts +94 -43
  151. package/.bkit/core/hooks/antigravity-hook.js +25 -10
  152. package/.bkit/core/hooks/gateguard-engine.js +53 -29
  153. package/.bkit/core/hooks/guardrails-shell.js +4 -4
  154. package/.bkit/core/hooks/guardrails-tiers.js +3 -3
  155. package/.bkit/core/hooks/preflight-linear.sh +9 -9
  156. package/.bkit/core/hooks/preflight-rules.sh +8 -0
  157. package/.bkit/core/hooks/readguard-engine.js +29 -10
  158. package/.bkit/core/hooks/session-rules-advisory.js +61 -0
  159. package/.bkit/core/prompts/bkit-audit-generate.md +187 -116
  160. package/.bkit/core/prompts/bkit-build-challenge-1.md +56 -37
  161. package/.bkit/core/prompts/bkit-build-challenge-2.md +65 -45
  162. package/.bkit/core/prompts/bkit-build-install-skills.md +38 -29
  163. package/.bkit/core/prompts/bkit-build-plan-convert.md +43 -32
  164. package/.bkit/core/prompts/bkit-build-plan-current.md +49 -39
  165. package/.bkit/core/prompts/bkit-build-plan-gaps-remediate.md +79 -55
  166. package/.bkit/core/prompts/bkit-build-plan-tasks.md +149 -0
  167. package/.bkit/core/prompts/bkit-build-plan-tests.md +120 -0
  168. package/.bkit/core/prompts/bkit-build-plan-verify.md +87 -50
  169. package/.bkit/core/prompts/bkit-debug-abort-handover.md +36 -35
  170. package/.bkit/core/prompts/bkit-debug-failure-diagnose.md +96 -64
  171. package/.bkit/core/prompts/bkit-debug-failure-remediate.md +54 -45
  172. package/.bkit/core/prompts/bkit-debug-failure-reproduce.md +42 -33
  173. package/.bkit/core/prompts/bkit-debug-failure-soak.md +53 -37
  174. package/.bkit/core/prompts/bkit-design-brainstorm-ideate-1.md +35 -30
  175. package/.bkit/core/prompts/bkit-design-brainstorm-ideate-2.md +34 -29
  176. package/.bkit/core/prompts/bkit-design-brainstorm-ideate-3.md +32 -28
  177. package/.bkit/core/prompts/bkit-design-generate.md +68 -52
  178. package/.bkit/core/prompts/bkit-design-organise.md +73 -35
  179. package/.bkit/core/prompts/bkit-design-review-1.md +58 -54
  180. package/.bkit/core/prompts/bkit-design-review-2.md +54 -50
  181. package/.bkit/core/prompts/bkit-design-spike.md +54 -47
  182. package/.bkit/core/prompts/bkit-design-update.md +42 -31
  183. package/.bkit/core/prompts/bkit-plan-generate.md +76 -58
  184. package/.bkit/core/prompts/bkit-plan-review-1.md +63 -49
  185. package/.bkit/core/prompts/bkit-plan-review-2.md +60 -48
  186. package/.bkit/core/prompts/bkit-plan-scope-check.md +53 -46
  187. package/.bkit/core/prompts/bkit-plan-update.md +70 -52
  188. package/.bkit/core/prompts/bkit-ship-conflict-resolve.md +82 -50
  189. package/.bkit/core/prompts/bkit-ship-deferred-sweep.md +80 -45
  190. package/.bkit/core/prompts/bkit-ship-update-docs.md +35 -29
  191. package/.bkit/core/prompts/bkit-utility-compress-markdown.md +55 -49
  192. package/.bkit/core/prompts/bkit-utility-update-decisions.md +76 -49
  193. package/.bkit/core/prompts/bkit-utility-update-deferrals.md +70 -41
  194. package/.bkit/core/prompts/bkit-utility-update-documentation.md +33 -30
  195. package/.bkit/core/prompts/bkit-utility-update-failures.md +73 -43
  196. package/.bkit/core/prompts/bkit-utility-update-knowledge.md +77 -50
  197. package/.bkit/core/prompts/bkit-utility-update-roadmap.md +77 -45
  198. package/.bkit/core/prompts/bkit-utility-update-tracker.md +144 -68
  199. package/.bkit/core/workflows/bkit-auto.md +53 -30
  200. package/.bkit/core/workflows/bkit-build.md +635 -304
  201. package/.bkit/core/workflows/bkit-debug.md +301 -148
  202. package/.bkit/core/workflows/bkit-design.md +167 -95
  203. package/.bkit/core/workflows/bkit-plan.md +172 -84
  204. package/.bkit/core/workflows/bkit-ship.md +299 -152
  205. package/.bkit/shared/.agents/plugins/bkit/hooks.json +10 -0
  206. package/.bkit/shared/.claude/bkit-cloud.md +12 -10
  207. package/.bkit/shared/.claude/settings.json +3 -0
  208. package/.bkit/shared/.codex/bkit-cloud.md +98 -0
  209. package/.bkit/shared/.codex/hooks.json +20 -0
  210. package/.bkit/shared/.cursor/bkit-cloud.md +3 -3
  211. package/.bkit/shared/.opencode/opencode.json +1 -1
  212. package/.bkit/shared/CLAUDE.md +5 -1
  213. package/.bkit/shared/GEMINI.md +4 -1
  214. package/.bkit/shared/docs/architecture.md +4 -7
  215. package/.bkit/shared/docs/data-schema.md +4 -7
  216. package/.bkit/shared/docs/deployments.md +239 -166
  217. package/.bkit/shared/docs/design.md +11 -12
  218. package/.bkit/shared/docs/noise.md +8 -7
  219. package/.bkit/shared/docs/observability.md +5 -8
  220. package/.bkit/shared/docs/security.md +35 -17
  221. package/.bkit/shared/docs/setup.md +29 -18
  222. package/.bkit/shared/docs/technology.md +3 -6
  223. package/.bkit/shared/docs/testing.md +282 -176
  224. package/.bkit/shared/gitignore +6 -2
  225. package/.bkit/templates/.claude/hooks/bkit-cloud-bootstrap.sh +15 -16
  226. package/.bkit/templates/.cursor/Dockerfile +12 -8
  227. package/.bkit/templates/.cursor/environment.json +1 -1
  228. package/.bkit/templates/.github/ISSUE_TEMPLATE/bkit-bug.yml +10 -8
  229. package/.bkit/templates/.github/ISSUE_TEMPLATE/bkit-change.yml +24 -12
  230. package/.bkit/templates/.github/ISSUE_TEMPLATE/bkit-idea.yml +3 -3
  231. package/.bkit/templates/.github/renovate.json5 +34 -6
  232. package/.bkit/templates/.github/workflows/bkit-ci.yml +53 -50
  233. package/.bkit/templates/.github/workflows/bkit-cloud.yml +13 -14
  234. package/.bkit/templates/.github/workflows/bkit-health.yml +97 -0
  235. package/.bkit/templates/.github/workflows/bkit-renovate.yml +5 -4
  236. package/.bkit/templates/.github/workflows/bkit-trivy.yml +5 -4
  237. package/.bkit/templates/.husky/pre-commit +2 -2
  238. package/.bkit/templates/.husky/pre-push +95 -35
  239. package/.bkit/templates/.prettierignore +6 -7
  240. package/.bkit/templates/.vale.ini +27 -27
  241. package/.bkit/templates/context/bkit-cloudrun.md +23 -15
  242. package/.bkit/templates/context/bkit-expo.md +70 -52
  243. package/.bkit/templates/context/bkit-github.md +93 -56
  244. package/.bkit/templates/context/bkit-inngest.md +21 -15
  245. package/.bkit/templates/context/bkit-maestro.md +22 -16
  246. package/.bkit/templates/context/bkit-playwright.md +23 -17
  247. package/.bkit/templates/context/bkit-resend.md +12 -10
  248. package/.bkit/templates/context/bkit-sentry.md +30 -20
  249. package/.bkit/templates/context/bkit-serena.md +8 -7
  250. package/.bkit/templates/context/bkit-stripe.md +14 -13
  251. package/.bkit/templates/context/bkit-supabase.md +16 -13
  252. package/.bkit/templates/context/bkit-tavily.md +14 -11
  253. package/.bkit/templates/context/bkit-vercel.md +27 -18
  254. package/.bkit/templates/docs/capabilities.md +15 -14
  255. package/.bkit/templates/scripts/ci/{check-accepted-divergences.mjs → bkit-check-accepted-divergences.mjs} +6 -3
  256. package/.bkit/templates/scripts/ci/{check-accepted-divergences.test.mjs → bkit-check-accepted-divergences.test.mjs} +3 -3
  257. package/.bkit/templates/scripts/ci/{check-capstone.mjs → bkit-check-capstone.mjs} +2 -2
  258. package/.bkit/templates/scripts/ci/{check-capstone.test.mjs → bkit-check-capstone.test.mjs} +4 -4
  259. package/.bkit/templates/scripts/ci/{check-dependency-audit.mjs → bkit-check-dependency-audit.mjs} +103 -18
  260. package/.bkit/templates/scripts/ci/{check-dependency-audit.test.mjs → bkit-check-dependency-audit.test.mjs} +31 -3
  261. package/.bkit/templates/scripts/ci/bkit-machine-lock.sh +387 -0
  262. package/.bkit/templates/scripts/ci/bkit-test-awake.mjs +75 -0
  263. package/.bkit/templates/scripts/ci/bkit-test-awake.test.mjs +160 -0
  264. package/.bkit/templates/scripts/ci/lanes/{resolve-push-base.mjs → bkit-resolve-push-base.mjs} +3 -3
  265. package/.bkit/templates/scripts/ci/lanes/{resolve-push-base.test.mjs → bkit-resolve-push-base.test.mjs} +4 -4
  266. package/.bkit/templates/scripts/cleanup/bkit-disk-prune.mjs +9 -1
  267. package/.bkit/templates/scripts/graphify/bkit-graph-drift.py +1 -1
  268. package/.bkit/templates/scripts/graphify/bkit-graph-eager.py +3 -3
  269. package/.bkit/templates/scripts/graphify/bkit-graph-embed.py +63 -0
  270. package/.bkit/templates/scripts/graphify/bkit-graph-imports.py +2 -2
  271. package/.bkit/templates/scripts/graphify/bkit-graph-pre-push.sh +69 -18
  272. package/.bkit/templates/scripts/graphify/bkit-graph-repair.py +3 -3
  273. package/.bkit/templates/scripts/graphify/bkit-graph-surface.py +1 -1
  274. package/.bkit/templates/scripts/graphify/bkit-graphify-graphml.py +1 -1
  275. package/.bkit/templates/scripts/graphify/bkit-graphify-rollup.py +197 -126
  276. package/.bkit/templates/scripts/graphify/bkit-graphify.sh +64 -41
  277. package/.bkit/templates/scripts/slop/bkit-slop-type-suppression.mjs +23 -7
  278. package/.bkit/templates/slop-scan.config.json +9 -2
  279. package/CHANGELOG.md +119 -0
  280. package/README.md +250 -76
  281. package/dist/checks/after/auth.js +36 -33
  282. package/dist/checks/after/ci-yml.js +52 -2
  283. package/dist/checks/after/index.js +12 -4
  284. package/dist/checks/after/tools.js +73 -33
  285. package/dist/checks/before/context-fetch.js +218 -33
  286. package/dist/checks/before/env.js +1 -1
  287. package/dist/checks/before/index.js +6 -2
  288. package/dist/checks/before/repo.js +35 -1
  289. package/dist/checks/status-holes.js +739 -114
  290. package/dist/checks/tips/catalogue.js +59 -17
  291. package/dist/checks/tips/detect.js +30 -0
  292. package/dist/checks/tips/index.js +48 -6
  293. package/dist/cli.js +222 -84
  294. package/dist/commands/analysis.js +77 -53
  295. package/dist/commands/ci/ci.js +302 -25
  296. package/dist/commands/ci/pins.js +142 -0
  297. package/dist/commands/ci/push-gate.js +194 -0
  298. package/dist/commands/install/uninstall.js +23 -27
  299. package/dist/commands/install/upgrade.js +21 -11
  300. package/dist/commands/land.js +126 -10
  301. package/dist/commands/lib.js +244 -5
  302. package/dist/commands/registry.js +6 -2
  303. package/dist/commands/setup/setup.js +1 -0
  304. package/dist/commands/stats.js +69 -30
  305. package/dist/commands/workflow/branches.js +1232 -111
  306. package/dist/commands/workflow/challenge.js +90 -37
  307. package/dist/commands/workflow/completeness.js +97 -0
  308. package/dist/commands/workflow/context.js +371 -51
  309. package/dist/commands/workflow/counters.js +206 -40
  310. package/dist/commands/workflow/decisions.js +60 -17
  311. package/dist/commands/workflow/disk.js +213 -55
  312. package/dist/commands/workflow/fids.js +18 -1
  313. package/dist/commands/workflow/issues.js +6 -0
  314. package/dist/commands/workflow/ledgers.js +372 -21
  315. package/dist/commands/workflow/locks.js +2 -0
  316. package/dist/commands/workflow/plan-blocks.js +78 -14
  317. package/dist/commands/workflow/plans.js +205 -0
  318. package/dist/commands/workflow/prs.js +12 -3
  319. package/dist/commands/workflow/repo.js +6 -0
  320. package/dist/commands/workflow/roadmap.js +50 -17
  321. package/dist/commands/workflow/run-facts.js +23 -28
  322. package/dist/commands/workflow/runs.js +530 -33
  323. package/dist/commands/workflow/runtime.js +3 -0
  324. package/dist/commands/workflow/skills.js +1 -0
  325. package/dist/commands/workflow/status.js +166 -19
  326. package/dist/commands/workflow/step-table.js +61 -29
  327. package/dist/commands/workflow/steps.js +542 -26
  328. package/dist/commands/workflow/summary-findings.js +117 -0
  329. package/dist/commands/workflow/summary.js +154 -41
  330. package/dist/commands/workflow/tasks.js +378 -0
  331. package/dist/commands/workflow/test-blocks.js +231 -0
  332. package/dist/commands/workflow/tests-changed.js +172 -0
  333. package/dist/commands/workflow/tests.js +1013 -321
  334. package/dist/commands/workflow/workflows.js +191 -38
  335. package/dist/commands/workflow/worktrees.js +310 -42
  336. package/dist/install/cloud-env.js +233 -99
  337. package/dist/install/context-ci-workflow.js +85 -0
  338. package/dist/install/manifest.js +57 -12
  339. package/dist/install/modules/antigravity.js +3 -2
  340. package/dist/install/modules/cloud.js +3 -3
  341. package/dist/install/modules/codex.js +12 -2
  342. package/dist/install/modules/core.js +851 -53
  343. package/dist/install/modules/cursor.js +6 -6
  344. package/dist/install/modules/opencode.js +11 -5
  345. package/dist/install/template-notes.js +116 -0
  346. package/dist/install/version-file.js +3 -2
  347. package/dist/ledgers/compact.js +87 -0
  348. package/dist/ledgers/journal.js +233 -0
  349. package/dist/ledgers/status.js +115 -5
  350. package/dist/ledgers/telemetry.js +121 -17
  351. package/dist/render/hosts.js +100 -14
  352. package/dist/shared/bkit-latest.js +26 -4
  353. package/dist/shared/dep-updater.js +49 -0
  354. package/dist/shared/egress.js +1 -1
  355. package/dist/shared/package-manager.js +6 -0
  356. package/dist/shared/settings.js +59 -10
  357. package/dist/shared/timeouts.js +2 -0
  358. package/dist/tools/cli-doc.js +97 -0
  359. package/dist/tools/grammar-density.js +1 -1
  360. package/dist/tools/instructions-sources.js +15 -16
  361. package/dist/tools/instructions.js +83 -39
  362. package/package.json +58 -42
  363. package/.bkit/core/.bkit/lite.md +0 -73
  364. package/.bkit/core/engine/commands/workflow/workstreams.ts +0 -172
  365. package/.bkit/core/prompts/bkit-build-plan-execute.md +0 -124
  366. package/.bkit/templates/scripts/ci/bkit-local-stack-lock.sh +0 -194
  367. package/dist/commands/workflow/workstreams.js +0 -149
  368. /package/.bkit/templates/.vale/styles/{House → Bkit}/Filler.yml +0 -0
  369. /package/.bkit/templates/.vale/styles/{House → Bkit}/Hedging.yml +0 -0
  370. /package/.bkit/templates/.vale/styles/{House → Bkit}/LLMSlop.yml +0 -0
  371. /package/.bkit/templates/.vale/styles/{House → Bkit}/Verbose.yml +0 -0
  372. /package/.bkit/templates/scripts/ci/{accepted-advisories.json → bkit-accepted-advisories.json} +0 -0
  373. /package/.bkit/templates/scripts/ci/{accepted-divergences.json → bkit-accepted-divergences.json} +0 -0
  374. /package/.bkit/templates/scripts/launchd/{launchd.env.example → bkit-launchd.env.example} +0 -0
  375. /package/.bkit/templates/scripts/launchd/{ai.bkit.launchd.plist → bkit-launchd.plist} +0 -0
@@ -5,18 +5,17 @@
5
5
  # Agent Prompts Index
6
6
 
7
7
  > [!CAUTION]
8
- > **Purpose:** Index of `.bkit/core/prompts/*.md` — description, frontmatter, rationale, settings. Audit template catalogue under § Audit (Ad-hoc).
8
+ > **Purpose:** The index of `.bkit/core/prompts/*.md`: each agent's description, frontmatter, rationale and settings. The audit template catalogue is under § Audit (Ad-hoc).
9
9
  >
10
10
  > **Rules:**
11
11
  >
12
- > - Use dense token-efficient phrasing; no filler / hedging / restating; tables > prose, bullets > paragraphs.
13
- > - Describe current state / rules, not reasons / history; state what to do, not why.
14
- > - Check `.agents/docs.md` index first and confirm topic belongs in this doc, not another.
15
- > - Ships to consumers: describe the current setup only. NEVER an internal identifier (tracker issue number, ledger key, review-finding id) and NEVER project history — decisions belong in `docs/decisions.jsonl`, incidents in `docs/knowledge.jsonl`.
16
- > - Subagents are leaves, never invoke other subagents.
17
- > - One row per agent — `Name | Scope | maxTurns | Recommended | Rationale`.
18
- > - Authoring rules for `.bkit/core/prompts/*.md` → `.bkit/authoring.md`; explanations behind them stay here.
19
- > - Update when agent tools / effort / step count changes materially.
12
+ > - Follow `.bkit/authoring.md` § All Agent-Read Files and § Docs when you edit this file.
13
+ > - This file ships to every project that installs bkit. Describe the current setup and the trade-offs behind it.
14
+ > - Leave out internal labels, such as tracker issue numbers, ledger keys and review-finding ids, and leave out project history. Decisions go in `docs/decisions.jsonl`, and incidents go in `docs/knowledge.jsonl`.
15
+ > - Subagents are leaves: a subagent never invokes another subagent.
16
+ > - Give each agent one row: `Name | Scope | maxTurns | Recommended | Rationale`.
17
+ > - The authoring rules for `.bkit/core/prompts/*.md` live in `.bkit/authoring.md`. The explanations behind them stay here.
18
+ > - Update this file when an agent's tools, effort or step count changes materially.
20
19
 
21
20
  ## Frontmatter Conventions
22
21
 
@@ -41,23 +40,42 @@ autoCompactWindow: 300000 # Claude-only compaction window; plain integer (`300k`
41
40
  | `low` | `claude-haiku-5-5` | Default Haiku rung. No agent pins Haiku; the tier stays available, with its per-provider pins in § Models (`low` → `high` buys +11.4 CursorBench points for $0.24/task). |
42
41
  | `medium` | `claude-sonnet-5-5` | Source modifiers with test loops, single-shot writers, scope audits, light utilities (Anthropic: `medium` for well-specified agentic tasks, `high` for harder or longer ones — a plan task or a diagnosis is well specified). Multi-doc writers pin `claude-sonnet-5-5` at `high`. |
43
42
  | `high` | `claude-opus-5-5` | Synthesizers, adversarial review, diagnosis. Highest rung any agent runs at. 1M context. |
44
- | `xhigh` | `claude-opus-5-5` | Reserved — no agent runs here. Same model ID as `high`; differs by `effort: xhigh` thinking-budget allocation. Escalate from `high` only on a measured quality gain (Anthropic: reserve `xhigh` and `max` for work where you've measured one). |
45
- | `max` | `claude-opus-5-5` | Same model as `xhigh`; differs by `effort: max` thinking-budget allocation. Reserved tier — escalate from `xhigh` only after observed truncation. |
43
+ | `xhigh` | `claude-opus-5-5` | Reserved: no agent runs here. Same model ID as `high`, and differs by the `effort: xhigh` thinking-budget allocation. Escalate from `high` only on a measured quality gain (Anthropic: reserve `xhigh` and `max` for work where you've measured one). |
44
+ | `max` | `claude-opus-5-5` | Same model as `xhigh`, and differs by the `effort: max` thinking-budget allocation. Reserved tier: escalate from `xhigh` only after observed truncation. |
46
45
 
47
- **Fable — not pinned.** `claude-fable-5-1` is the tier above Opus 5.5 ($10/$50 vs $4/$20). Cursor Cloud lists `claude-fable-5-thinking-high`/`-xhigh`. no bkit agent pins Fable. `max` stays on `claude-opus-5-5`.
46
+ **Fable — not pinned.** `claude-fable-5-1` is the tier above Opus 5.5 ($10/$50 against $4/$20). Cursor Cloud lists `claude-fable-5-thinking-high`/`-xhigh`. No bkit agent pins Fable, and `max` stays on `claude-opus-5-5`.
48
47
 
49
- Pinning `model:` locks agent model regardless of orchestrator session. Without pin, subagents inherit session default — `/bkit-plan` from Sonnet silently downgrades every Opus-tier agent.
48
+ Pinning `model:` locks an agent's model, whatever model the orchestrator session runs. Without a pin, subagents inherit the session default, so `/bkit-plan` started from Sonnet silently downgrades every Opus-tier agent.
50
49
 
51
- **Effort → host thinking (canonical):** Prompt `effort:` = row key in **Default**. **Keyed on HOST, not provider** — field name + ladder come from the host config format (`effort:` claude/cursor, a `variant:` field opencode, `model_reasoning_effort` codex), not the model → every opencode provider shares the one **OpenCode** column. A provider key here THROWS (axis confusion, not a fallback). Codex ceilings `max`→`xhigh` (no higher rung). Install parses this table — no hardcoded ladders in TS. **opencode columns ceiling out, they do not downshift:** a provider below its tier target spends **more** thinking to close the gap, and can afford to — every non-Claude pin costs $0.03-$1.76 per agentic task vs Claude's $2.14-$5.34 (`providers.md` § Agentic-coding parity board). Every rung from `medium` up runs at OpenCode `max`, so no opencode agent thinks below the heavy-thinking setting its board score was measured at. Haiku-tier agents run at `high` (OpenCode `max`): every non-Claude board score was measured with heavy thinking, so the haiku target compares like with like, and haiku-cell pins cost $0.03-$0.11 per task. **Effort and model are independent** — multi-doc writers run `claude-sonnet-5-5` at `high` and the ledger agents run `claude-opus-5-5` at `medium`; the map above is the default model per rung, not a constraint. **Cursor stays identity with Default** — its thinking lever is the model-ID suffix (`cursor-grok-4.5-high`), not `effort:`, so ceiling-out happens at the pin.
50
+ **Effort → host thinking (canonical):** a prompt's `effort:` is the row key in the Default column.
51
+
52
+ - The table is keyed on host, not provider. The field name and its ladder come from the host's config format, not from the model: `effort:` for claude / cursor, a `variant:` field for OpenCode, and `model_reasoning_effort` for codex, so every OpenCode provider shares the one OpenCode column.
53
+ - A provider key in this table throws an error. It is an axis confusion, and nothing falls back.
54
+ - Codex caps `max` at `xhigh`, because it has no higher rung.
55
+ - Install parses this table, and no ladder is hardcoded in TS.
56
+ - OpenCode effort targets the Claude bar, never past it. A model under the sonnet bar thinks as hard as it can to close the gap, so every rung from `medium` up runs at OpenCode `max`. A model at or above the sonnet bar (`OPENCODE_AT_BAR_MODELS` in `.bkit/core/engine/render/hosts.ts`: `gpt-6.1-sol`, `gpt-6-astra` and the Claude models) skips the OpenCode column and keeps the agent's own `Claude` effort. It matches the Claude agent it stands in for, instead of out-spending it.
57
+ - Effort and model are independent. Multi-doc writers run `claude-sonnet-5-5` at `high`, and the ledger agents run `claude-opus-5-5` at `medium`. The map above gives the default model per rung, not a constraint.
58
+ - The Cursor column stays identical to Default, because Cursor's thinking lever is the model-ID suffix (`cursor-grok-4.5-high`), not `effort:`. The cap is applied at the pin.
52
59
 
53
60
  <!-- prettier-ignore -->
54
- | Default | Claude | Cursor | OpenCode | Codex | Antigravity |
61
+ | Default | Claude | Codex | Antigravity | Cursor | OpenCode |
55
62
  | --- | --- | --- | --- | --- | --- |
56
- | `low` | `low` | `low` | `none` | `low` | `low` |
57
- | `medium` | `medium` | `medium` | `max` | `medium` | `medium` |
58
- | `high` | `high` | `high` | `max` | `high` | `high` |
63
+ | `low` | `low` | `low` | `low` | `low` | `none` |
64
+ | `medium` | `medium` | `medium` | `medium` | `medium` | `max` |
65
+ | `high` | `high` | `high` | `high` | `high` | `max` |
59
66
  | `xhigh` | `xhigh` | `xhigh` | `max` | `xhigh` | `max` |
60
- | `max` | `max` | `max` | `max` | `xhigh` | `max` |
67
+ | `max` | `max` | `xhigh` | `max` | `max` | `max` |
68
+
69
+ **OpenCode `variant:` is per model per gateway, not per host.** The ladder above is what bkit asks for. The renderer (`clampOpencodeVariant`) clamps it to what the pinned model exposes on that gateway (`reasoning_options` in models.dev), because OpenCode 2.0 hard-errors on a variant the model does not carry there. On the current pins:
70
+
71
+ - `deepseek-v4.1-flash` takes `max`. It exposes `none/low/high/max` on `openrouter` and `low/high/max` on Zen and Go, so a `low` agent clamps `none` → `low` there.
72
+ - `gpt-6.1-sol` exposes `low…max`, so it keeps the agent's own `high` / `medium`.
73
+ - `gemini-4-argon` exposes `low/medium/high`, and clamps to `high`.
74
+ - `ollama` gets no variant on any model.
75
+
76
+ Subscription hosts keep the eager mapping, because on a plan you already paid for, spending the reasoning budget is the point.
77
+
78
+ **`Google` is one model, many effort levels.** Antigravity pins `gemini-4-argon` on all 40 agents. The per-agent `effort` is the only dial, because Google's catalogue offers no rung between argon (61.2) and `gemini-3.8-flash` (16.5). Argon exposes `low/medium/high`, so an `xhigh`/`max` agent clamps to `high`. No agent uses those tiers today.
61
79
 
62
80
  ### Field rationale
63
81
 
@@ -65,13 +83,13 @@ Pinning `model:` locks agent model regardless of orchestrator session. Without p
65
83
  | Field | Why this default | When to deviate |
66
84
  | --- | --- | --- |
67
85
  | `name` | `@name` invocation in `commands/*.md` requires exact match | Never deviate. |
68
- | `description` | Surfaces in picker and `Agent` tool catalog; one-sentence scope statement only — H1 inside the markdown body carries the deliverable name (`# Build Skills Installed`, `# Plan Review`, etc.). | Keep terse; never duplicate the H1. |
86
+ | `description` | Surfaces in the picker and the `Agent` tool catalog. A one-sentence scope statement only: the H1 inside the markdown body carries the deliverable name (`# Build Skills Installed`, `# Plan Review`, etc.). | Keep it terse, and never duplicate the H1. |
69
87
  | `effort` | Maps to a Claude model via the table above, with two groups pinning opus at `medium` effort — a depth-sensitive lens (`bkit-plan-scope-check`) and the ledger / tracker agents that check entries against the codebase — and multi-doc writers pinning sonnet at `high`. Parallel fan-outs (triplet `-1`/`-2`/`-3` for ideate; pairs `-1`/`-2` for design review / plan review / challenge) mix `high`/`medium` effort to spread reasoning depth on Claude. On Cursor, the same fan-outs use different `model:` values for cross-family diversity. | `high` for adversarial / generative / diagnostic / cross-codebase work and multi-doc writers. `medium` for source modifiers with test loops and ledger upkeep. `xhigh` only on a measured quality gain. |
70
- | `model` | Explicit Claude model pin per the effort mapping table above. Locks the model independent of the orchestrator's session default — without it, subagents inherit user's Claude Code default and Opus thinking depth + 1M context disappear if the user runs Sonnet. | Brittle by version (must update on every model rev), but explicit and discoverable. Never use family alias (`opus` / `sonnet`) — they don't lock context-window variant. |
88
+ | `model` | Explicit Claude model pin, per the effort mapping table above. It locks the model whatever the orchestrator's session default is. Without it, subagents inherit the user's Claude default, and Opus thinking depth and 1M context disappear if the user runs Sonnet. | Brittle by version, because it needs an update on every model revision, but explicit and discoverable. Never use a family alias (`opus` / `sonnet`), because an alias does not lock the context-window variant. |
71
89
  | `tools` | Minimum capability sandbox. Enforced on Claude (native). Cursor: `bkit install` T3 transform passes the field through verbatim, but Cursor CLI 2026.10.01 does not enforce it (measured: a `tools: [Read]` subagent still has `Shell`, `Write`, `StrReplace`, and every MCP tool via `CallDynamicTool`). Removing `Edit`/`Bash` makes an agent provably non-mutating on Claude only. | Add `WebFetch`/`WebSearch` only for genuine ecosystem research (`bkit-design-review-1/2`) or to pin a probe against vendor docs before running it (`bkit-design-spike`). |
72
- | `permissionMode: acceptEdits` | End-to-end runs of `/design`/`/plan`/`/build`/`/debug`/`/ship` would break if mid-workflow prompt blocked an agent. Bash auto-approve surface lives in `.claude/settings.json` (`workflows.md` § Permissions) | Never deviate inside the pipeline; one-off ad-hoc invocations may use stricter modes. |
73
- | `maxTurns` | Pure runaway-loop safety cap, not a workload budget. 4 tiers (S=50 / M=100 / L=200 / XL=400). Hitting the cap means the agent is definitely stuck (tight retry loop, infinite re-read pattern, etc.), not "tight on a complex run." | Cap by **tier** (see § `maxTurns` calibration). Per-agent assignment in § Models. |
74
- | `autoCompactWindow: 300000` | Safety net, not the cap: the agent summarises its context at 300k tokens at the latest. Only lowers the window — a lower repo / user setting or `CLAUDE_CODE_AUTO_COMPACT_WINDOW` still wins. Claude Code only; other hosts drop it (`providers.md` § Context compaction). | Never deviate; every `bkit-*` agent carries it (`hosts.test.ts`). |
90
+ | `permissionMode: acceptEdits` | End-to-end runs of `/design`/`/plan`/`/build`/`/debug`/`/ship` would break if a mid-workflow prompt blocked an agent. The Bash auto-approve surface lives in `.claude/settings.json` (`workflows.md` § Permissions). | Never deviate inside the pipeline. One-off ad-hoc invocations may use stricter modes. |
91
+ | `maxTurns` | Pure runaway-loop safety cap, not a workload budget. 4 tiers (S=50 / M=100 / L=200 / XL=400). Hitting the cap means the agent is definitely stuck (tight retry loop, infinite re-read pattern, etc.), not "tight on a complex run." | Cap by tier (§ `maxTurns` calibration). The per-agent assignment is in § Models. |
92
+ | `autoCompactWindow: 300000` | Safety net, not the cap: the agent summarises its context at 300k tokens at the latest. Claude only; precedence and other hosts: `providers.md` § Context compaction. | Never deviate; every `bkit-*` agent carries it (`hosts.test.ts`). |
75
93
 
76
94
  ### Effort tiers
77
95
 
@@ -79,7 +97,7 @@ Pinning `model:` locks agent model regardless of orchestrator session. Without p
79
97
  | Tier | Use for | Example agents |
80
98
  | --- | --- | --- |
81
99
  | `low` | Pure synthesis with no verdict riding on it. | (none currently) |
82
- | `medium` | Source modifiers with test loops — implementers and fix executors (sonnet model) — single-shot writers, scope audits, skill curators, and ledger / tracker / roadmap upkeep against the codebase (opus model). | `bkit-design-brainstorm-ideate-2`, `bkit-design-review-2`, `bkit-plan-scope-check`, `bkit-build-plan-convert`, `bkit-build-plan-execute`, `bkit-build-plan-gaps-remediate`, `bkit-build-install-skills`, `bkit-debug-failure-remediate`, `bkit-debug-abort-handover`, `bkit-ship-conflict-resolve`, `bkit-utility-compress-markdown`, `bkit-utility-update-deferrals`, `bkit-utility-update-failures`, `bkit-utility-update-decisions`, `bkit-utility-update-knowledge`, `bkit-utility-update-roadmap`, `bkit-utility-update-tracker` |
100
+ | `medium` | Source modifiers with test loops — implementers and fix executors (sonnet model) — single-shot writers, scope audits, skill curators, and ledger / tracker / roadmap upkeep against the codebase (opus model). | `bkit-design-brainstorm-ideate-2`, `bkit-design-review-2`, `bkit-plan-scope-check`, `bkit-build-plan-convert`, `bkit-build-plan-tests`, `bkit-build-plan-tasks`, `bkit-build-plan-gaps-remediate`, `bkit-build-install-skills`, `bkit-debug-failure-remediate`, `bkit-debug-abort-handover`, `bkit-ship-conflict-resolve`, `bkit-utility-compress-markdown`, `bkit-utility-update-deferrals`, `bkit-utility-update-failures`, `bkit-utility-update-decisions`, `bkit-utility-update-knowledge`, `bkit-utility-update-roadmap`, `bkit-utility-update-tracker` |
83
101
  | `high` | Top-of-pipeline synthesizers and their critique mergers, adversarial review, parallel-fan-out high-effort slots, failure diagnosis, observability instrumentation, feasibility research, audits, and multi-doc writers (sonnet model). | `bkit-design-brainstorm-ideate-1`, `bkit-design-brainstorm-ideate-3`, `bkit-design-generate`, `bkit-design-update`, `bkit-design-review-1`, `bkit-design-spike`, `bkit-design-organise`, `bkit-plan-generate`, `bkit-plan-update`, `bkit-plan-review-1`, `bkit-plan-review-2`, `bkit-build-plan-current`, `bkit-build-plan-verify`, `bkit-build-challenge-1`, `bkit-build-challenge-2`, `bkit-debug-failure-diagnose`, `bkit-debug-failure-soak`, `bkit-debug-failure-reproduce`, `bkit-ship-deferred-sweep`, `bkit-ship-update-docs`, `bkit-audit-generate`, `bkit-utility-update-documentation` |
84
102
  | `xhigh` | Reserved — only on a measured quality gain over `high` for a recurring failure mode. Model-gated (Opus-only on current models). A stuck bug goes to Fable by hand (§ Orchestrators), not to an `xhigh` agent. | (none currently) |
85
103
  | `max` | Reserved — only if `xhigh` proves insufficient on a recurring failure mode. Costliest tier; one-shot orchestrator-level work only. Lazy-escalate from `xhigh` based on observed truncation, never preemptively. | (none currently) |
@@ -95,7 +113,7 @@ Pinning `model:` locks agent model regardless of orchestrator session. Without p
95
113
  | Stub-replace + log-append reviewer | `Read, Grep, Glob, Write, Edit, Bash` | `bkit-build-plan-current`, `bkit-build-plan-verify`, `bkit-build-challenge-1`, `bkit-build-challenge-2`, `bkit-ship-deferred-sweep` |
96
114
  | Doc rewriter (cp + Edit) | `Read, Grep, Glob, Edit, Bash` | `bkit-design-update`, `bkit-plan-update` |
97
115
  | Doc writer (with shell) | `Read, Grep, Glob, Write, Edit, Bash` | `bkit-design-generate`, `bkit-design-organise`, `bkit-plan-generate`, `bkit-build-plan-convert`, `bkit-ship-update-docs`, `bkit-utility-update-documentation` |
98
- | Source modifier (with test gate) | `Read, Grep, Glob, Write, Edit, Bash` | `bkit-build-plan-execute`, `bkit-build-plan-gaps-remediate`, `bkit-debug-failure-reproduce`, `bkit-debug-failure-remediate`, `bkit-ship-conflict-resolve` |
116
+ | Source modifier (with test gate) | `Read, Grep, Glob, Write, Edit, Bash` | `bkit-build-plan-tests`, `bkit-build-plan-tasks`, `bkit-build-plan-gaps-remediate`, `bkit-debug-failure-reproduce`, `bkit-debug-failure-remediate`, `bkit-ship-conflict-resolve` |
99
117
  | Skill-installer | `Read, Grep, Glob, Write, Bash` | `bkit-build-install-skills` |
100
118
  | Tracker reader (read-only MCP) | `Read, Grep, Glob, Write, Edit, Bash` + per tracker: github `mcp__github__list_issues, mcp__github__issue_read`; linear `mcp__linear__list_issues, mcp__linear__get_issue`; jira `mcp__jira__searchJiraIssuesUsingJql, mcp__jira__getJiraIssue`; none — | `bkit-utility-update-roadmap` |
101
119
  | Tracker writer (read-write MCP) | `Read, Grep, Glob, Bash` + per tracker: github `mcp__github__list_issues, mcp__github__issue_read, mcp__github__issue_write, mcp__github__sub_issue_write, mcp__github__list_label, mcp__github__projects_list, mcp__github__projects_write`; linear `mcp__linear__list_issues, mcp__linear__get_issue, mcp__linear__save_issue, mcp__linear__list_issue_labels, mcp__linear__list_issue_statuses, mcp__linear__list_milestones`; jira `mcp__jira__searchJiraIssuesUsingJql, mcp__jira__getJiraIssue, mcp__jira__editJiraIssue, mcp__jira__transitionJiraIssue, mcp__jira__executeRead, mcp__jira__executeWrite`; none — | `bkit-utility-update-tracker` |
@@ -103,86 +121,109 @@ Pinning `model:` locks agent model regardless of orchestrator session. Without p
103
121
  | Investigator (web access) | `Read, Grep, Glob, Write, Edit, Bash, WebFetch, WebSearch` | `bkit-design-review-1`, `bkit-design-review-2`, `bkit-design-spike`, `bkit-debug-failure-diagnose`, `bkit-debug-failure-soak` |
104
122
  | Investigator + tracker comment | `Read, Grep, Glob, Write, Edit, Bash, WebFetch, WebSearch` + per tracker: github `mcp__github__add_issue_comment, mcp__github__search_issues`; linear `mcp__linear__save_comment, mcp__linear__list_issues`; jira `mcp__jira__addOrEditJiraIssueComment, mcp__jira__searchJiraIssuesUsingJql`; none — | `bkit-audit-generate` |
105
123
 
106
- **Language-server lookups.** Every agent carrying `Grep` except `bkit-utility-compress-markdown` and `bkit-build-install-skills` (neither reads source) also lists `mcp__serena__*` directly after `Glob`. A Claude Code subagent with a `tools:` list reaches an MCP tool only when the list names it; an entry for an unwired server is dropped silently and the agent still starts (probed Claude Code 2.1.295 + Serena 1.7.0: explicit names, `mcp__serena__*`, and bare `mcp__serena` all grant; omitted → none). Wildcard grants every tool Serena advertises, so the template `.serena/project.yml` `excluded_tools` list is what keeps Serena lookup-only — Serena edits bypass GateGuard (`Edit|Write|MultiEdit` matcher). Measured with that template: `claude-code` and `ide` contexts advertise lookups only; `codex` advertises the excluded tools but refuses each call. Product-named only here — the template ships with bkit (`.bkit/templates.md` § Serena); prompt bodies and the `.agents/rules.md` `Language server first` / `Code reads` rules stay capability-worded. Codex ignores `tools:`, so a wired server is reachable regardless. Cursor passes the entry through verbatim and ignores `tools:`, so a wired server is reachable regardless (probed Cursor CLI 2026.10.01; with the template `.serena/project.yml`, `--context ide` offers lookups only). OpenCode respells `mcp__<server>__<tool>` as `<server>_<tool>` (its MCP tool ids) in its `tools` map → `serena_*`.
124
+ **Language-server lookups.** Every agent carrying `Grep` also lists `mcp__serena__*` directly after `Glob`, except `bkit-utility-compress-markdown` and `bkit-build-install-skills`, because neither reads source.
107
125
 
108
- **Maintainer-local release agents.** `utility-new-release` and `utility-pre-release-check` live only in the bkit repo's `.claude/agents/` and `.opencode/agents/` — hand-authored, no `.bkit/core/prompts/` source, not shipped in the package. They are intentionally absent from this catalog and from the Models / tool tables.
126
+ - A Claude subagent with a `tools:` list reaches an MCP tool only when the list names it. An entry for an unwired server is dropped silently, and the agent still starts (probed Claude 2.1.295 + Serena 1.7.0: explicit names, `mcp__serena__*`, and bare `mcp__serena` all grant; omitted → none).
127
+ - The wildcard grants every tool Serena advertises, so the template `.serena/project.yml` `excluded_tools` list is what keeps Serena lookup-only. That matters because Serena edits bypass GateGuard (`Edit|Write|MultiEdit` matcher). Measured with that template, the `claude-code` and `ide` contexts advertise lookups only, and `codex` advertises the excluded tools but refuses each call.
128
+ - Serena is named by product only here, because the template ships with bkit (`.bkit/templates.md` § Serena). Prompt bodies and the `.agents/rules.md` `Language server first` / `Code reads` rules stay capability-worded.
129
+ - Codex ignores `tools:`, so a wired server is reachable regardless.
130
+ - Cursor passes the entry through verbatim and ignores `tools:`, so a wired server is reachable regardless (probed Cursor CLI 2026.10.01; with the template `.serena/project.yml`, `--context ide` offers lookups only).
131
+ - OpenCode respells `mcp__<server>__<tool>` as `<server>_<tool>` (its MCP tool ids) in its `tools` map → `serena_*`.
132
+
133
+ **Maintainer-local release agents.** `utility-new-release` and `utility-pre-release-check` live only in the bkit repo's `.claude/agents/`, `.codex/agents/` and `.opencode/agents/`. They are hand-authored, have no `.bkit/core/prompts/` source and are not shipped in the package. They are intentionally absent from this catalog and from the Models / tool tables.
109
134
 
110
135
  ### `maxTurns` calibration
111
136
 
112
- `maxTurns` = max **agent turns** (assistant messages); each turn may emit multiple parallel tool calls. **Runaway-loop safety cap, not workload budget.** Runtime bounded by universal 3600s step timeout (`workflows.md` § Execution Protocol).
137
+ `maxTurns` = the maximum number of agent turns (assistant messages). Each turn may emit several parallel tool calls. It is a runaway-loop safety cap, not a workload or runtime budget. bkit sets no universal workflow-step deadline; a command's own bounded wait and the host's execution limits still apply (`.bkit/providers.md` § Long-running shell commands).
113
138
 
114
- **Same philosophy as timeout:** 3600s ~2x typical 30 min runtime — hit means something wrong. `maxTurns` tiers are S=50 / M=100 / L=200 / XL=400, per-agent in § Models:
139
+ The `maxTurns` tiers are S=50 / M=100 / L=200 / XL=400, assigned per agent in § Models:
115
140
 
116
141
  <!-- prettier-ignore -->
117
142
  | Tier | Cap | Typical work | Use for | Examples |
118
143
  | --- | --- | --- | --- | --- |
119
144
  | **S** | **50** | 3-10 turns | Pure synthesizers reading 1-3 run artifacts and writing one verdict; single-doc reformatters with one check command. | `bkit-build-plan-convert` |
120
- | **M** | **100** | 5-25 turns | Reviewers and generators reading a few docs and writing one, scope checks, single-pass utilities, install, handover writers; one plan group's executor, verifier and gap fixer (<= `bkit.workflows.tasks.maxPerGroup` tasks each, default 3). | `bkit-design-brainstorm-ideate-*`, `bkit-design-generate`, `bkit-design-review-*`, `bkit-design-organise`, `bkit-plan-review-*`, `bkit-plan-scope-check`, `bkit-build-install-skills`, `bkit-build-plan-execute`, `bkit-build-plan-verify`, `bkit-build-plan-gaps-remediate`, `bkit-debug-abort-handover`, `bkit-utility-compress-markdown` |
145
+ | **M** | **100** | 5-25 turns | Reviewers and generators reading a few docs and writing one, scope checks, single-pass utilities, install, handover writers; one task group's tests writer, executor, verifier and gap fixer (<= `bkit.workflows.tasks.maxPerGroup` tasks each, default 5; the tests writer carries the group's tests, so the executor carries none). | `bkit-design-brainstorm-ideate-*`, `bkit-design-generate`, `bkit-design-review-*`, `bkit-design-organise`, `bkit-plan-review-*`, `bkit-plan-scope-check`, `bkit-build-install-skills`, `bkit-build-plan-tests`, `bkit-build-plan-tasks`, `bkit-build-plan-verify`, `bkit-build-plan-gaps-remediate`, `bkit-debug-abort-handover`, `bkit-utility-compress-markdown` |
121
146
  | **L** | **200** | 30-60 turns | Source modifiers with test loops — iterative edit→`pnpm bkit:gate:static-unit`→edit cycles; probe-code runners; investigators sweeping a diff, its callers or the codebase; whole-doc rewriters; per-cycle doc sync. | `bkit-design-spike`, `bkit-design-update`, `bkit-plan-update`, `bkit-build-plan-current`, `bkit-build-challenge-1`, `bkit-build-challenge-2`, `bkit-debug-failure-reproduce`, `bkit-debug-failure-diagnose`, `bkit-debug-failure-remediate`, `bkit-debug-failure-soak`, `bkit-ship-conflict-resolve`, `bkit-ship-deferred-sweep`, `bkit-ship-update-docs` |
122
147
  | **XL** | **400** | 60-180 turns | Whole-surface sweeps: every probe step of one audit template against repo and live systems; every row of a jsonl ledger, the roadmap board or the doc catalog re-checked against code, docs and tracker; every citation in a plan's design, audit and issue inputs re-checked against code before the plan is written. | `bkit-plan-generate`, `bkit-audit-generate`, `bkit-utility-update-*` |
123
148
 
124
- **Tier assignment is canonical** — per-agent rows in § Models. Cap hit at any tier → agent stuck (retry loop, infinite re-read) — orchestrator verdict-validation retry handles partial recovery.
149
+ **Tier assignment is canonical:** the per-agent rows are in § Models. A cap hit at any tier means the agent is stuck (retry loop, infinite re-read). The orchestrator's verdict-validation retry handles partial recovery.
125
150
 
126
151
  ## Models
127
152
 
128
- Canonical per-agent metadata — **single source of truth for model per agent**. **Keyed on PROVIDER, not host** — a model ID belongs to the gateway serving it → shared provider = one column, never two; table never grows as hosts x providers. Columns: `Anthropic` (Claude Code's pinned provider — locked ID, session-independent), `Cursor` (Cursor's pinned provider; cross-family diversity), `OpenAI` (Codex), `Google` (Antigravity), one per opencode provider (`OpenRouter`, `Ollama`, …). `bkit install --opencode=<provider>` reads the matching column; every allowlisted provider ships one.
153
+ Canonical per-agent metadata, and the single source of truth for each agent's model. The table is keyed on provider, not host, because a model ID belongs to the gateway serving it, so a shared provider is one column, never two, and the table never grows as hosts x providers. The columns, subscription providers first: `Anthropic` (Claude's pinned provider, with a locked ID that does not depend on the session), `OpenAI` (Codex), `Google` (Antigravity), then `Cursor` (Cursor's pinned provider, for cross-family diversity), then one per OpenCode provider (`OpenRouter`, `Ollama`, `OpenCode Zen`, `OpenCode Go`). `bkit install --opencode=<provider>` reads the matching column, and every allowlisted provider ships one.
154
+
155
+ **Cells = bare provider-scoped model ID** (`deepseek-v4.1-flash`, `moonshotai/kimi-k3`), never with a provider prefix. `formatModelPin` adds the host-format prefixes at render, and also maps the key `opencode-zen` → OpenCode's own id `opencode`. A prefixed cell (`opencode/deepseek-v4.1-flash`) gets a double prefix.
156
+
157
+ > **`OpenAI` column — provisional, not parity-proven.** Every other gateway column = a live catalog probe plus a per-model tool-calling smoke test. The `OpenAI` cells follow value inside the plan: `gpt-6-astra` (77.5) only on the 12 judges (Judges note below), and `gpt-6.1-sol` (68.7) on the 27 builders. Sol scores 96.0 Coding against astra's 93.7, so on code-writing agents it is both the better model and 5.9x cheaper per LiveBench task ($116.78 against $690.82). `gpt-6-luna` (30.9, $17.42/task) is the catalog's best value, and stays unused until a haiku tier exists. `gpt-6-astra` is OpenAI's opus analogue, not its fable: it costs the same $691 per task as `claude-opus-5.5`'s $684, for 77.5 against 72.8, and both sit at ~0.11 points per dollar. The fable-shaped tier (`gpt-daybreak-red`, $12.5/$75) is unmeasured on 2.0 and unpinned, exactly as `claude-fable-5.1` is unpinned on the Claude side. Re-pin from a real probe before relying on them.
129
158
 
130
- **Cells = BARE provider-scoped model ID** (`deepseek-v4.1-flash`, `moonshotai/kimi-k3`) — never a provider prefix. Host-format prefixes are added at render by `formatModelPin`, which also maps key `opencode-zen` → opencode's own id `opencode`. A prefixed cell (`opencode/deepseek-v4.1-flash`) double-prefixes.
159
+ > **Haiku stays unpinned — deliberate.** `claude-haiku-5-5` is the cheapest Claude rung (40.0 Long Agentic, $188.84/task against opus's $684.16), and no agent runs it. The 17 medium-effort agents are not boilerplate: four are source modifiers with test loops, three are judges, six reconcile ledgers and the roadmap against the codebase, and the rest read a whole run before writing. Haiku's documented fit is a well-defined task, and a role prompt cannot promise that its input is well-defined. The quota dial is `Effort`, which is reversible per agent, and the model rung is not.
131
160
 
132
- > **`OpenAI` column — PROVISIONAL, not parity-proven.** Every other gateway column = live catalog probe + per-model tool-calling smoke. `OpenAI` cells mapped from documented role only (`gpt-6-astra` deep reasoning, `gpt-5.6-terra` balanced default). Haiku cells are board-picked: `gpt-6-luna` (LiveBench 51.2; `gpt-5.6-luna` measures 48.4), 2026-10-09, existence via models.dev only — no OpenAI key to probe. Re-pin from a real probe before relying on them. `gpt-5.4` / `gpt-5.4-mini` retired 2026-08-31.
161
+ > **`Google` column — measured pin.** Every cell runs `gemini-4-argon`, which scores 61.2 Long Agentic on the 2.0 board and is served on the `google` provider (models.dev), so Antigravity's free tier and a Google subscription both reach it. `gemini-3.8-pro` is unmeasured on 2.0 and is not pinned, and `gemini-3.8-flash` 16.5 is not a candidate.
133
162
 
134
- `Effort` + `Max Turns` cross-host: `Effort` = Claude `effort:` → opencode `variant:` field / Codex `model_reasoning_effort` at install; `Max Turns` = runaway cap (Claude today). `Anthropic` / `Effort` / `Max Turns` restate prompt-file frontmatter — parity enforced by `.bkit/core/engine/audits/workflows/models-table-parity.test.ts`. See § `maxTurns` calibration for workload rationale.
163
+ `Effort` and `Max Turns` across hosts: `Effort` = the Claude `effort:`, which install maps to the OpenCode `variant:` field / Codex `model_reasoning_effort`. `Max Turns` = the runaway cap (Claude today). The `Anthropic` / `Effort` / `Max Turns` columns restate the prompt-file frontmatter, and `.bkit/core/engine/audits/workflows/models-table-parity.test.ts` enforces that parity. The workload rationale is in § `maxTurns` calibration.
135
164
 
136
- > **Context-window note:** Claude Code `model:` is the API ID. `claude-opus-5-5` and `claude-sonnet-5-5` are 1M context. Haiku pins `claude-haiku-5-5` (1M context).
165
+ > **Context-window note:** Claude `model:` is the API ID. `claude-opus-5-5` and `claude-sonnet-5-5` are 1M context. Haiku pins `claude-haiku-5-5` (1M context).
166
+ >
167
+ > **OpenCode model IDs:** one column per provider. Each cell holds the bare model ID that provider serves (e.g. `moonshotai/kimi-k3`, `deepseek-v4.1-flash:1017`). `bkit install --opencode=<provider>` reads the column and adds the OpenCode provider prefix (`formatModelPin`), so `moonshotai/kimi-k3` is written as `openrouter/moonshotai/kimi-k3`. Allowlisted: `openrouter`, `ollama`, `opencode-zen` + `opencode-go` (both `BKIT_OPENCODE_API_KEY`). All four ship curated columns below. Zen's models.dev id is `opencode/<id>`, and Go's is `opencode-go/<id>`. Both gateways are built-ins (apiKey only). The `/zen/v1/models` REST probe cannot see Go-only models, so models.dev is the source of truth. See `.bkit/providers.md` § Provider models.
168
+ >
169
+ > **Metered columns target the Claude bar; subscription columns spend the plan.** The OpenCode columns use Claude as the yardstick: each cell is the cheapest model closest to its tier's Claude bar (`providers.md` § Selecting pins). `Anthropic`, `OpenAI` and `Google` instead take each vendor's best value inside the plan. Any cell can be re-pinned.
137
170
  >
138
- > **opencode model IDs:** Per-provider column; full provider-prefixed ID (`<provider>/<model-id>`, e.g. `openrouter/moonshotai/kimi-k3`, `ollama/deepseek-v4.1-flash`). `bkit install --opencode=<provider>` installs column verbatim. Allowlisted: `openrouter`, `ollama`, `opencode-zen` + `opencode-go` (both `BKIT_OPENCODE_API_KEY`). **All four ship curated columns below.** Zen: models.dev id `opencode/<id>`; Go: `opencode-go/<id>`. Both gateways = built-ins (apiKey only). **`/zen/v1/models` REST probe cannot see Go-only models — models.dev is source of truth.** See `.bkit/providers.md` § Models.
171
+ > **Judges — 12 agents:** `bkit-design-brainstorm-ideate-3`, `bkit-design-review-1`, `bkit-plan-review-1`, `bkit-plan-review-2`, `bkit-plan-scope-check`, `bkit-build-plan-current`, `bkit-build-plan-verify`, `bkit-build-challenge-1`, `bkit-build-challenge-2`, `bkit-debug-failure-diagnose`, `bkit-debug-failure-soak`, `bkit-ship-deferred-sweep`. They review, challenge, verify or escalate, and the other 27 are builders. **Key gates — 4 of the 12**, where a miss passes downstream unchecked: `bkit-build-plan-verify` (decides a build is done), `bkit-ship-deferred-sweep` (last check before merge), `bkit-debug-failure-diagnose` (a wrong root cause sends the fix loop after the wrong bug), `bkit-build-challenge-2` (a missed security finding ships).
139
172
  >
140
- > **Columns target scores, not tiers.** opencode columns do **not** mirror the Claude ladder: each cell is the cheapest model reaching that tier's target on the parity board (`providers.md` § Selecting pins); chain-of-thought reasoner for convergent agents (risk / diagnosis lens). Per-row columns; any cell re-pinnable.
173
+ > **Per-tier quality targets; cheapest sufficient pin.** Each OpenCode cell is the cheapest model closest to its tier's Claude bar on LiveBench 2.0 Long Agentic: opus 72.8, sonnet 63.4 and haiku 40.0, the `claude-opus-5.5` / `claude-sonnet-5.5` / `claude-haiku-5.5` scores on the 2026-10-25 release (`providers.md` § LiveBench 2.0 board; process in § Selecting pins). A cell under its bar records the gap. **Any vendor may pin on an OpenCode column; cost is the limit** (`providers.md` § Budget band), and cost is what keeps Claude models off them.
141
174
  >
142
- > **Per-tier quality targets; cheapest sufficient pin.** Tier cells are not the same rung of each provider's ladder — each is the **cheapest model that reaches that tier's Claude score** on LiveBench Agentic Coding (opus target **65.2**, sonnet target **59.4**, haiku target **~51** — `claude-haiku-5-5` 51.4 at `xhigh`, run at `high`; opus and sonnet targets stay the `claude-opus-5` / `claude-sonnet-5` scores because LiveBench's 5.5 rows land level (opus) or don't read cleanly (sonnet) — `providers.md` § Agentic-coding parity board; process in § Selecting pins). Two consequences pull opposite ways: a provider whose ceiling sits under the sonnet target **collapses opus+sonnet onto that ceiling** (spreading them buys nothing and costs quality on the sonnet half), while **haiku cells stay cheap** because a low-cost model already clears the haiku bar. **Claude is never a pin on an opencode column** — those columns exist to carry work when the Claude quota is spent, so a Claude model there defeats the purpose regardless of score. **opus** → `openrouter` `deepseek-v4.1-flash` (77.3); `ollama` `deepseek-v4.1-flash` (77.3); `opencode-zen` `deepseek-v4.1-flash` (77.3 — re-pinned from `kimi-k2.7-code` because 2.0 rejects a `variant:` the model does not expose, and `kimi-k2.7-code` exposes none; Zen bills V4.1 Flash at $0.30/$1.20, so this cell costs more than the old one); `opencode-go` `deepseek-v4.1-flash` (77.3). **sonnet** → same `deepseek-v4.1-flash` on every opencode column (clears 59.4). **haiku** → `openrouter` + `opencode-zen` `glm-5.3-flash` (56.8); `ollama` + `opencode-go` `deepseek-v4.1-flash` (77.3 — cheaper than GLM on those two); `OpenAI` `gpt-6-luna` (51.2). **convergent** → `kimi-k3` (62.2) on every opencode column; on both gateways it bills $3/$15 per Mtok, the column's most expensive pin. **gpt-oss pinned nowhere** — `gpt-oss-120b` measures **~10 est**, below every tier target including haiku. **`ollama` subscription:** every tier needs a **paid plan** (`deepseek-v4.1-flash` / `kimi-k3`); the free `qwen3.5:397b` / `gpt-oss:20b` are unmeasured or far below target — free-tier users re-pin in § Model Overrides and accept unverified quality. **`opencode-go`** carries no `gemini-*` (it lists `gpt-6-luna`, `gpt-5.6-luna` and `claude-haiku-5-5`; Claude is still never pinned); its DeepSeek models need the workspace privacy setting on "Global regions"; prefer Go 5h window pins. Re-pin via § Model Overrides or models.dev / provider `/v1/models` (`providers.md`).
175
+ > - **builders** → `openrouter` `deepseek-v4.1-flash-1017` + `ollama` `deepseek-v4.1-flash:1017` (40.9, which is 31.9 under opus and 22.5 under sonnet); `opencode-zen` + `opencode-go` `deepseek-v4.1-flash` (not measured, because the undated slug serves the 2026-09-10 snapshot, not the `1017` row; Zen bills it $0.30/$1.20). Opus and sonnet collapse onto it, because the only model over the sonnet bar that the budget carries is `gpt-6.1-sol`, and it carries four agents.
176
+ > - **key gates** → `openrouter` `openai/gpt-6.1-sol` + `opencode-zen` `gpt-6.1-sol` (68.7, 4.1 under opus); `opencode-go` `gemini-4-argon` (61.2, because Go serves no sol); `ollama` `deepseek-v4.1-flash:1017` (nothing stronger in its catalog).
177
+ > - **The other eight judges** → the column's DeepSeek pin.
178
+ > - **haiku** → no agent runs the tier (Haiku note above). `providers.md` § Per-tier lineup across providers holds its cells: `ollama` `deepseek-v4.1-flash:1017` (40.9, +0.9), `opencode-go` `deepseek-v4.1-flash` (not measured), `openrouter` + `opencode-zen` `glm-5.3-flash` (19.0, 21.0 under the bar). `OpenAI` `gpt-6-luna` (30.9) waits for the tier.
179
+ > - **`kimi-k3` pinned nowhere:** 26.4 at a $0.19 cache read.
180
+ > - **gpt-oss pinned nowhere:** `gpt-oss-120b` is not on the 2.0 board, so nothing measures it against a target.
181
+ >
182
+ > The plan and account requirements per provider (paid Ollama plan, Go privacy setting and 5h window) are in `providers.md` § Per-tier lineup across providers. Re-pin through § Model Overrides, or from models.dev / the provider's `/v1/models` (`providers.md`).
143
183
 
144
184
  <!-- prettier-ignore -->
145
- | Prompt | Anthropic | Cursor | OpenRouter | Ollama | OpenCode Zen | OpenCode Go | OpenAI | Google | Effort | Max Turns | Role |
185
+ | Prompt | Anthropic | OpenAI | Google | Cursor | OpenRouter | Ollama | OpenCode Zen | OpenCode Go | Effort | Max Turns | Role |
146
186
  | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
147
- | `bkit-design-brainstorm-ideate-1` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | parallel slot 1 (pragmatic lens) |
148
- | `bkit-design-brainstorm-ideate-2` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | parallel slot 2 (optimal-tradeoff lens; Composer vs Grok siblings on Cursor; cross-tier on Claude) |
149
- | `bkit-design-brainstorm-ideate-3` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `moonshotai/kimi-k3` | `kimi-k3` | `kimi-k3` | `kimi-k3` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | parallel slot 3 (risk-premortem lens) |
150
- | `bkit-design-generate` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | generate design from brainstorms (top-of-pipeline synthesizer; whole-doc rewrite) |
151
- | `bkit-design-review-1` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | design review slot 1 (web search + feasibility analysis) |
152
- | `bkit-design-review-2` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | design review slot 2 (Composer vs Grok sibling on Cursor; cross-tier on Claude) |
153
- | `bkit-design-spike` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | settle review-named unknowns with throwaway probe code (conditional step) |
154
- | `bkit-design-update` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | merge review critiques into design (whole-doc rewrite — needs depth) |
155
- | `bkit-design-organise` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | decompose reviewed design into dependency-annotated workstreams; coverage self-verify + DAG cycle detection |
156
- | `bkit-plan-generate` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 400 | plan draft generator (top-of-pipeline synthesizer) |
157
- | `bkit-plan-review-1` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | review slot 1 (tightness + dependency order + design-coverage; reads the plan) |
158
- | `bkit-plan-review-2` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `moonshotai/kimi-k3` | `kimi-k3` | `kimi-k3` | `kimi-k3` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | review slot 2 (regression-risk + design-coverage; reads the code) |
159
- | `bkit-plan-scope-check` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 100 | scope audit (template, criteria, verification; mechanical audit — medium effort, depth-sensitive — opus) |
160
- | `bkit-plan-update` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | merge review + scope critiques into plan (whole-doc rewrite — needs depth) |
161
- | `bkit-build-plan-convert` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 50 | write an issue or a free-form plan as task blocks (single-doc writer with one check command) |
162
- | `bkit-build-plan-current` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | judge plan vs changes landed since its base commit |
163
- | `bkit-build-install-skills` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | skill identifier |
164
- | `bkit-build-plan-execute` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | task executor for one plan group (source modifier with test loop) |
165
- | `bkit-build-plan-verify` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 100 | verify one group, or the whole plan once, vs the plan |
166
- | `bkit-build-plan-gaps-remediate` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | fix one group's verification gaps (source modifier with test loop) |
167
- | `bkit-build-challenge-1` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | parallel slot 1 (bug-hunting lens) |
168
- | `bkit-build-challenge-2` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `moonshotai/kimi-k3` | `kimi-k3` | `kimi-k3` | `kimi-k3` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | parallel slot 2 (security lens; depth-sensitive — both opus) |
169
- | `bkit-debug-failure-reproduce` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | failing tests from a bug report, no fix; opus over the sonnet test-writer tier — an off-symptom test sends the fix loop after the wrong bug |
170
- | `bkit-debug-failure-diagnose` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `moonshotai/kimi-k3` | `kimi-k3` | `kimi-k3` | `kimi-k3` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | failure diagnosis (Final-strategy attempts especially benefit from depth before human handover) |
171
- | `bkit-debug-failure-remediate` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 200 | implement diagnosis fixes (source modifier with test loop) |
172
- | `bkit-debug-failure-soak` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `moonshotai/kimi-k3` | `kimi-k3` | `kimi-k3` | `kimi-k3` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | soak fix — one accumulating-context instrument-or-fix step per iteration against the live boundary (orchestrator re-runs tests + loops) |
173
- | `bkit-debug-abort-handover` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | handover writer |
174
- | `bkit-ship-conflict-resolve` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 200 | mechanical rebase-conflict resolver (source modifier with test loop) |
175
- | `bkit-ship-deferred-sweep` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 200 | pre-merge completeness gate — intended-vs-actual + declared-deferral resurface |
176
- | `bkit-ship-update-docs` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `high` | 200 | sync cataloged docs (multi-doc writer) |
177
- | `bkit-audit-generate` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `high` | 400 | audit report from one template (top-of-pipeline synthesizer) |
178
- | `bkit-utility-update-deferrals` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 400 | deferrals update |
179
- | `bkit-utility-update-failures` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 400 | failures update |
180
- | `bkit-utility-update-decisions` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 400 | decisions update |
181
- | `bkit-utility-update-knowledge` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 400 | knowledge update |
182
- | `bkit-utility-update-roadmap` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 400 | roadmap analysis |
183
- | `bkit-utility-update-tracker` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | `medium` | 400 | tracker update |
184
- | `bkit-utility-compress-markdown` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `medium` | 100 | token-optimize markdown |
185
- | `bkit-utility-update-documentation` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | `high` | 400 | doc sync (multi-doc writer) |
187
+ | `bkit-design-brainstorm-ideate-1` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | parallel slot 1 (pragmatic lens) |
188
+ | `bkit-design-brainstorm-ideate-2` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | parallel slot 2 (optimal trade-off lens; Composer vs Grok siblings on Cursor; cross-tier on Claude) |
189
+ | `bkit-design-brainstorm-ideate-3` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | parallel slot 3 (risk-premortem lens) |
190
+ | `bkit-design-generate` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | generate design from brainstorms (top-of-pipeline synthesizer; whole-doc rewrite) |
191
+ | `bkit-design-review-1` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | design review slot 1 (web search + feasibility analysis) |
192
+ | `bkit-design-review-2` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | design review slot 2 (Composer vs Grok sibling on Cursor; cross-tier on Claude) |
193
+ | `bkit-design-spike` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | settle review-named unknowns with throwaway probe code (conditional step) |
194
+ | `bkit-design-update` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | merge review critiques into design (whole-doc rewrite — needs depth) |
195
+ | `bkit-design-organise` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | decompose reviewed design into dependency-annotated workstreams grouped into plans; coverage self-verify + DAG cycle detection |
196
+ | `bkit-plan-generate` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 400 | plan draft generator (top-of-pipeline synthesizer) |
197
+ | `bkit-plan-review-1` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | review slot 1 (tightness + dependency order + design-coverage; reads the plan) |
198
+ | `bkit-plan-review-2` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 100 | review slot 2 (regression-risk + design-coverage; reads the code) |
199
+ | `bkit-plan-scope-check` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | scope audit (template, criteria, verification; mechanical audit — medium effort, depth-sensitive — opus) |
200
+ | `bkit-plan-update` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | merge review + scope critiques into plan (whole-doc rewrite — needs depth) |
201
+ | `bkit-build-plan-convert` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 50 | write an issue or a free-form plan as task blocks (single-doc writer with one check command) |
202
+ | `bkit-build-plan-current` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | judge plan vs changes landed since its base commit |
203
+ | `bkit-build-install-skills` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | installs matching skills under the registry's activation boundaries |
204
+ | `bkit-build-plan-tests` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | writes one task group's planned tests before its code (test files only); uses the machine lock for shared services |
205
+ | `bkit-build-plan-tasks` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | task executor for one task group (source modifier with test loop); uses the machine lock for shared services; re-runs tests-lock waits |
206
+ | `bkit-build-plan-verify` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `openai/gpt-6.1-sol` | `deepseek-v4.1-flash:1017` | `gpt-6.1-sol` | `gemini-4-argon` | `high` | 100 | verify one group, or the whole plan once, vs the plan; re-runs tests-lock waits |
207
+ | `bkit-build-plan-gaps-remediate` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | fix one group's verification gaps (source modifier with test loop); uses the machine lock for shared services; re-runs group-gate and changed-test lock waits |
208
+ | `bkit-build-challenge-1` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | parallel slot 1 (bug-hunting lens) |
209
+ | `bkit-build-challenge-2` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `openai/gpt-6.1-sol` | `deepseek-v4.1-flash:1017` | `gpt-6.1-sol` | `gemini-4-argon` | `high` | 200 | parallel slot 2 (security lens; depth-sensitive — both opus) |
210
+ | `bkit-debug-failure-reproduce` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | failing tests from a bug report, no fix; opus over the sonnet test-writer tier — an off-symptom test sends the fix loop after the wrong bug |
211
+ | `bkit-debug-failure-diagnose` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `openai/gpt-6.1-sol` | `deepseek-v4.1-flash:1017` | `gpt-6.1-sol` | `gemini-4-argon` | `high` | 200 | failure diagnosis (Final-strategy attempts especially benefit from depth before human handover) |
212
+ | `bkit-debug-failure-remediate` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 200 | implement diagnosis fixes (source modifier with test loop); re-runs tests-lock waits |
213
+ | `bkit-debug-failure-soak` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | soak fix — one accumulating-context instrument-or-fix step per iteration against the live boundary (orchestrator re-runs tests + loops) |
214
+ | `bkit-debug-abort-handover` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | handover writer |
215
+ | `bkit-ship-conflict-resolve` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 200 | mechanical merge-conflict resolver (source modifier with test loop) |
216
+ | `bkit-ship-deferred-sweep` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `openai/gpt-6.1-sol` | `deepseek-v4.1-flash:1017` | `gpt-6.1-sol` | `gemini-4-argon` | `high` | 200 | pre-merge completeness gate — intended-vs-actual + declared-deferral resurface |
217
+ | `bkit-ship-update-docs` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 200 | sync cataloged docs (multi-doc writer) |
218
+ | `bkit-audit-generate` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 400 | audit report from one template (top-of-pipeline synthesizer) |
219
+ | `bkit-utility-update-deferrals` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 400 | deferrals update |
220
+ | `bkit-utility-update-failures` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 400 | failures update |
221
+ | `bkit-utility-update-decisions` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 400 | decisions update |
222
+ | `bkit-utility-update-knowledge` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 400 | knowledge update |
223
+ | `bkit-utility-update-roadmap` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 400 | roadmap analysis |
224
+ | `bkit-utility-update-tracker` | `claude-opus-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 400 | tracker update |
225
+ | `bkit-utility-compress-markdown` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `medium` | 100 | token-optimize markdown |
226
+ | `bkit-utility-update-documentation` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `high` | 400 | doc sync (multi-doc writer) |
186
227
 
187
228
  **Effort vs maxTurns vs model — three independent knobs:**
188
229
 
@@ -195,54 +236,67 @@ Canonical per-agent metadata — **single source of truth for model per agent**.
195
236
 
196
237
  ### Live model-ID verification
197
238
 
198
- > **Verified against live provider APIs** (`providers.md` § Per-provider live model lookup). Last verification: 2026-10-19 (Ollama column — existence via `/api/tags`, no tool-call smoke: no Ollama key; OpenCode Zen + Go opus/sonnet/convergent — existence via models.dev); 2026-10-09 (Claude; haiku cells of every column — existence only, no tool-call smoke: Ollama needs a paid plan, Zen balance zero, no Go subscription, no OpenAI key); 2026-10-06 (Cursor); 2026-09-16 (OpenRouter column); 2026-07-25 (remaining columns).
239
+ > **Verified against live provider APIs** (`providers.md` § Per-provider live model lookup). Last verification: 2026-10-28 (all four OpenCode columns — existence via OpenRouter `/api/v1/models`, Ollama `/v1/models` and models.dev; tool-call smoke refused: the OpenRouter key is past its monthly limit, no Ollama or OpenCode key); 2026-10-09 (Claude; haiku cells of every column — existence only, no tool-call smoke: Ollama needs a paid plan, Zen balance zero, no Go subscription, no OpenAI key); 2026-10-06 (Cursor); 2026-07-25 (remaining columns).
199
240
  >
200
241
  > - **Claude** (`/v1/models`) — pins `claude-opus-5-5` / `claude-sonnet-5-5` / `claude-haiku-5-5`. Opus 5, Sonnet 5 and Haiku 4.5 remain listed; not pinned.
201
- > - **Cursor** (`/v0/models`) — **defaults = first-party pool only:** `cursor-grok-4.5-high` (depth / synthesizers / parallel diversity) + `composer-2.5` (volume / build+ship orchestrators). Catalog has **no** `cursor-grok-4.7` and **no** Opus 5.5 / Sonnet 5.5 IDs. API-pool IDs present and **not pinned**: `claude-opus-5-thinking-high`, `claude-sonnet-5-thinking-high`/`-xhigh`, `gpt-5.6-*`, `claude-fable-5-thinking-high`/`-xhigh`. Opt in via § Model Overrides.
202
- > - **Google** — Antigravity pins `gemini-3.8-pro` for reasoning/synthesis, `gemini-3.8-flash` for the haiku tier.
242
+ > - **Cursor** (`/v0/models`) — the defaults come from the first-party pool only: `cursor-grok-4.5-high` (depth / synthesizers / parallel diversity) + `composer-2.5` (volume / build+ship orchestrators). The catalog has no `cursor-grok-4.7` and no Opus 5.5 / Sonnet 5.5 IDs. API-pool IDs that are present but not pinned: `claude-opus-5-thinking-high`, `claude-sonnet-5-thinking-high`/`-xhigh`, `gpt-5.6-*`, `claude-fable-5-thinking-high`/`-xhigh`. Opt in through § Model Overrides.
243
+ > - **Google** — Antigravity pins `gemini-4-argon` on every agent (§ Models `Google` column).
203
244
  >
204
- > **OpenAI-compatible providers curated + tool-call-smoked** (`tools:` chat / completions round-trip per pin, confirming `tool_calls` — `/models` existence != tool-calls):
245
+ > **OpenAI-compatible providers curated** (tool-call smoke = a `tools:` chat / completions round-trip per pin that confirms `tool_calls`, because `/models` existence != tool-calls; the 2026-10-28 pins are existence-only):
205
246
  >
206
- > - **OpenRouter** (`/api/v1/models`, public) — `deepseek/deepseek-v4.1-flash` (opus+sonnet, 77.3, 16+ endpoints), `z-ai/glm-5.3-flash` (haiku, 56.8, 27 endpoints), `moonshotai/kimi-k3` (convergent, 62.2, 17 endpoints). All three auto-cache — `providers.md` § Selecting pins. (Tool-call smoke 403 `Key limit exceeded` — existence-verified; smoke deferred.)
207
- > - **Ollama** (`/v1/models`) — `deepseek-v4.1-flash` (opus+sonnet+haiku, 77.3), `kimi-k3` (convergent, 62.2). Every tier needs a paid plan. (Existence-verified via `/api/tags` 2026-10-19; smoke deferred.)
247
+ > - **OpenRouter** (`/api/v1/models`, public) — the overflow default in § Model Presets: `deepseek/deepseek-v4.1-flash-1017` (builders and eight judges, 40.9, 36 endpoints) and `openai/gpt-6.1-sol` (the four key gates, 68.7, 8 endpoints). Option 2 adds five more judges on `openai/gpt-6.1-sol` against the A$140 cap. DeepSeek auto-caches. (Existence via `/api/v1/models` 2026-10-28.)
248
+ > - **Ollama** (`/v1/models`) — `deepseek-v4.1-flash:1017` on all 40 (40.9, because nothing stronger is in the catalog). Every tier needs a paid plan, and the A$ caps do not apply to a plan-billed gateway.
208
249
  >
209
- > **opencode gateways — TWO providers; catalogs from models.dev (not `/zen/v1/models`).** `/zen/v1/models` = Zen only; CANNOT see Go-only models — models.dev source of truth (`opencode` = Zen, `opencode-go` = Go).
250
+ > **OpenCode gateways — two providers; catalogs from models.dev (not `/zen/v1/models`).** `/zen/v1/models` = Zen only, and it cannot see Go-only models, so models.dev is the source of truth (`opencode` = Zen, `opencode-go` = Go).
210
251
  >
211
- > - **`opencode-zen`** (`https://opencode.ai/zen/v1`) — full gateway; `deepseek-v4.1-flash` (opus+sonnet), `glm-5.3-flash` (haiku), `kimi-k3` (convergent, 62.2). `kimi-k2.7-code` / `qwen3.6-plus` not pinned — dominated by V4.1 Flash (77.3 at $0.06/task vs 45.7 at $0.39 / 41.4 at $1.55). models.dev validated (Zen balance zero; smoke on Go).
212
- > - **`opencode-go`** (`https://opencode.ai/zen/**go**/v1` — nested under `/zen/`) — open-weight budget plan. `deepseek-v4.1-flash` (opus+sonnet+haiku, 77.3), `kimi-k3` (convergent, 62.2 — $3/$15 per Mtok, the heaviest draw on the Go 5h window). `minimax-m3` / `qwen3.7-plus` not pinned — both score under V4.1 Flash at higher cost (40.7 at $0.42 for `minimax-m3`; `qwen3.7-plus` unmeasured). Prefer Go 5h window pins. No `gemini-*`; never pin its `gpt-*` / `claude-*`.
252
+ > - **`opencode-zen`** (`https://opencode.ai/zen/v1`) — `deepseek-v4.1-flash` (builders; the 2026-09-10 snapshot, not measured on 2.0, because the 40.9 row is the `1017` refresh), `gpt-6.1-sol` (the four key gates, 68.7). Validated on models.dev.
253
+ > - **`opencode-go`** (`https://opencode.ai/zen/**go**/v1`) — `deepseek-v4.1-flash` (builders; the same 2026-09-10 snapshot, not measured), `gemini-4-argon` (the four key gates, 61.2, because Go serves no sol). Prefer pins inside the Go 5h window. DeepSeek models need the workspace privacy setting on "Global regions".
213
254
 
214
255
  ### Orchestrators (slash commands)
215
256
 
216
- `.bkit/core/workflows/*.md` = workflow orchestrators — dispatch subagents, synthesize results. **Not subagents** (no `effort:` / `maxTurns:` frontmatter); run on session model. Pin session model to table below when invoking.
257
+ The files in `.bkit/core/workflows/*.md` are the workflow orchestrators: they dispatch subagents and synthesize the results. They are not subagents (no `effort:` / `maxTurns:` frontmatter), and they run on the session model. Pin the session model to the table below when you invoke one.
217
258
 
218
259
  <!-- prettier-ignore -->
219
- | Orchestrator | Anthropic | Cursor | OpenRouter | Ollama | OpenCode Zen | OpenCode Go | OpenAI | Google | Role |
260
+ | Orchestrator | Anthropic | OpenAI | Google | Cursor | OpenRouter | Ollama | OpenCode Zen | OpenCode Go | Role |
220
261
  | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
221
- | `/design` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | Synthesizes conflicting brainstorms; resolves design ambiguity. |
222
- | `/plan` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | Merges review (pair) + scope-check critiques in one pass; fixes what `tests plan-check` refuses after each plan write. |
223
- | `/build` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | Fixed pipeline: dispatch → verify → challenge → tests → summary. Subagents own the heavy reasoning (`bkit-build-challenge-1`/`-2`, `bkit-build-plan-verify` are `high` effort). |
224
- | `/debug` | `claude-opus-5-5` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-6-astra` | `gemini-3.8-pro` | Picks remediation path under uncertainty; Standard / Pivot / Final escalation. |
225
- | `/ship` | `claude-sonnet-5-5` | `composer-2.5` | `deepseek/deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | `gpt-5.6-terra` | `gemini-3.8-pro` | Mostly mechanical: merge + archive + parallel docs / summary. |
262
+ | `/design` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | Synthesizes conflicting brainstorms; resolves design ambiguity. |
263
+ | `/plan` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | Merges review (pair) + scope-check critiques in one pass; fixes what `tests plan-check` refuses after each plan write. |
264
+ | `/build` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | Fixed pipeline: dispatch → verify → challenge → tests → summary. Subagents own the heavy reasoning (`bkit-build-challenge-1`/`-2`, `bkit-build-plan-verify` are `high` effort). |
265
+ | `/debug` | `claude-opus-5-5` | `gpt-6-astra` | `gemini-4-argon` | `cursor-grok-4.5-high` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | Picks remediation path under uncertainty; Standard / Pivot / Final escalation. |
266
+ | `/ship` | `claude-sonnet-5-5` | `gpt-6.1-sol` | `gemini-4-argon` | `composer-2.5` | `deepseek/deepseek-v4.1-flash-1017` | `deepseek-v4.1-flash:1017` | `deepseek-v4.1-flash` | `deepseek-v4.1-flash` | Mostly mechanical: merge + archive + parallel docs / summary. |
267
+
268
+ **Rule:** Opus where the orchestrator resolves ambiguity, and Sonnet for a defined sequence.
269
+
270
+ - **If** a Sonnet orchestrator skips Failure Path branches or mis-routes verify gaps → escalate to Opus.
226
271
 
227
- **Rule:** Opus where orchestrator resolves ambiguity; Sonnet for defined sequence. **If** Sonnet orchestrator skips Failure Path branches or mis-routes verify gaps → escalate Opus.
272
+ **Reasoning depth — default `medium`** for all orchestrators. Orchestration is procedural (verdict extraction, worst-of-N, gate verification, failure-path branching), not deep reasoning. Higher Sonnet thinking worsens inlining drift ("I could just do this myself"), so the drift escalation is `sonnet/medium → opus/medium`, never `sonnet/medium → sonnet/high`. Reserve `high` for one-off debugging of steps that mis-extract verdicts or malform `status.jsonl`.
228
273
 
229
- **Reasoning depth — default `medium`** for all orchestrators. Orchestration is procedural (verdict extraction, worst-of-N, gate verification, failure-path branching) — not deep reasoning. Higher Sonnet thinking _worsens_ inlining drift ("I could just do this myself"): drift escalation → `sonnet/medium → opus/medium`, never `sonnet/medium → sonnet/high`. Reserve `high` for one-off debugging of steps that mis-extract verdicts or malform `status.jsonl`.
274
+ **Inlining-drift risk.** Sonnet is more prone than Opus to inlining subagent work when the work is tractable (review / scope-check). `rules.md` § Orchestrators and the per-command CAUTION box mitigate it.
230
275
 
231
- **Inlining-drift risk.** Sonnet more prone than Opus to inlining subagent work when tractable (review / scope-check); `rules.md` § Orchestrators + per-command CAUTION mitigate. Observed → bump command to Opus (not higher Sonnet reasoning). Cursor Cloud automation: `composer-2.5` for metadata / API plumbing; launched agent runs orchestrator model from table.
276
+ - **If** drift is observed → bump the command to Opus, not to higher Sonnet reasoning.
232
277
 
233
- **Claude Code session setup** — orchestrator runs on the session model + effort; subagents ignore both (`model:` / `effort:` frontmatter wins over session level).
278
+ Cursor Cloud automation uses `composer-2.5` for metadata / API plumbing, and the launched agent runs the orchestrator model from the table.
279
+
280
+ **Claude session setup:** the orchestrator runs on the session model and effort. Subagents ignore both, because `model:` / `effort:` frontmatter wins over the session level.
234
281
 
235
282
  <!-- prettier-ignore -->
236
283
  | Goal | How |
237
284
  | --- | --- |
238
285
  | Model per phase | Launch on the `Anthropic` cell above before the first prompt: `claude --model claude-opus-5-5` → `/bkit-design`, `/bkit-plan`, `/bkit-debug`; `claude --model claude-sonnet-5-5` → `/bkit-build`, `/bkit-ship`. Mid-session `/model` re-reads the whole conversation uncached (cache is per model). |
239
286
  | Fable, by hand | `claude --model claude-fable-5-1` (2.5x `claude-opus-5-5`), outside any `/bkit-*` run: (1) refine a fuzzy idea into a brief before `/bkit-design` → start `/bkit-design` in a fresh `claude-opus-5-5` session; (2) `/bkit-debug` stuck on a bug → work it from `docs/plans/{YYYYMM}/logs/{FID}/debug-handover.md`, `/effort xhigh`. No orchestrator or agent pins Fable. |
240
- | Effort `medium` | Claude Code default for `claude-opus-5-5`, `claude-sonnet-5-5`, `claude-haiku-5-5` — set nothing. Top-level `effortLevel` in `~/.claude/settings.json` does NOT apply to these models; a level `/effort` saves lands in `~/.claude/settings.json` `modelSettings.<model>.effortLevel` and does. bkit ships no `effortLevel`. |
287
+ | Effort `medium` | Claude default for `claude-opus-5-5`, `claude-sonnet-5-5`, `claude-haiku-5-5` — set nothing. Top-level `effortLevel` in `~/.claude/settings.json` does not apply to these models; a level `/effort` saves lands in `~/.claude/settings.json` `modelSettings.<model>.effortLevel` and does. bkit ships no `effortLevel`. |
241
288
  | One session | `claude --effort <level>` at launch, or `/effort <level>` in-session. |
242
289
  | One repo | `"effortLevel": "<level>"` in `.claude/settings.local.json` (you only) or `.claude/settings.json` (whole team) — applies to every model, outranks `~/.claude/settings.json`. |
243
290
 
244
- - NEVER set `CLAUDE_CODE_EFFORT_LEVEL` — outranks subagent `effort:` frontmatter → every subagent runs one level, § Effort tiers ladder gone.
245
- - NEVER set `maxEffortLevel` below `high` — caps subagent `effort:` frontmatter → `high` agents (§ Effort tiers) run at the cap.
291
+ - Never set `CLAUDE_CODE_EFFORT_LEVEL`, because it outranks subagent `effort:` frontmatter. Every subagent then runs at one level, and the § Effort tiers ladder is gone.
292
+ - Never set `maxEffortLevel` below `high`, because it caps subagent `effort:` frontmatter. The `high` agents (§ Effort tiers) then run at the cap.
293
+
294
+ **Codex session setup:** the orchestrator runs on the session model and effort. Subagents ignore both, because each `.codex/agents/*.toml` sets its own `model` and `model_reasoning_effort`.
295
+
296
+ <!-- prettier-ignore -->
297
+ | Goal | How |
298
+ | --- | --- |
299
+ | Model per phase | Launch on the `OpenAI` cell above before the first prompt: `codex -m gpt-6-astra -c model_reasoning_effort=medium` → `$bkit-design`, `$bkit-plan`, `$bkit-debug`; `codex -m gpt-6.1-sol -c model_reasoning_effort=medium` → `$bkit-build`, `$bkit-ship`. The `-c` flag sets the effort, because `.codex/config.toml` sets none. |
246
300
 
247
301
  ## Design Phase
248
302
 
@@ -250,41 +304,42 @@ Canonical per-agent metadata — **single source of truth for model per agent**.
250
304
  | Agent | Scope |
251
305
  | --- | --- |
252
306
  | `bkit-design-brainstorm-ideate-1` | Independent design exploration — pragmatic lens; effort slot 1 (`high`). Runs parallel with `-2` and `-3`; writes `design-brainstorm-ideate-1.md`. Never references sibling output (circular reasoning guard). |
253
- | `bkit-design-brainstorm-ideate-2` | Same as `-1` — optimal-tradeoff lens; effort slot 2 (`medium`). Effort-diversity within the parallel triplet (Sonnet on Claude, Composer on Cursor vs Grok siblings). |
307
+ | `bkit-design-brainstorm-ideate-2` | Same as `-1` — optimal trade-off lens; effort slot 2 (`medium`). Effort-diversity within the parallel triplet (Sonnet on Claude, Composer on Cursor vs Grok siblings). |
254
308
  | `bkit-design-brainstorm-ideate-3` | Same as `-1` — risk-premortem lens; effort slot 3 (`high`). Restores Opus depth for the failure-mode side of the brainstorm triplet. |
255
- | `bkit-design-generate` | Generate unified design draft from brainstorms and the design input; the input's `Grill decisions` section (design Step 3) holds the human's answers — `Settled` entries are strong pointers, not hard constraints — followed unless a brainstorm shows the answer fails; a departure gets a `Decisions` row and a `Risks & Open Questions` item asking the human to confirm. `Assumed, not confirmed` entries are defaults it may override with a stated reason; each `Open` entry becomes a `Risks & Open Questions` item. The section's `Draft Acceptance Scenario`, labelled `DRAFT only`, is the starting point for the design's `Acceptance Scenario`; every change to it gets a `Decisions` row, no confirmation item, because the human confirms the scenario again at the plan's `Acceptance Scenario confirmed` box. It reads the input directly because the ideate triplet may not carry every decision through. Emits `Reuse & Prior Art` (default `EXTEND`): duplication is cheapest to decide at design — net-new committed here is expensive to unwind at plan or build. Every net-new `Architecture` component traces to a `NEW` verdict; plan phase starts `Reuse Inventory` from it. Verdict only — sources, absence proof and rejected `NEW` reasons are checked by `bkit-design-review-1` and evidenced in the plan's `Search performed`. Appends `docs/deferrals.jsonl` if applicable. Does not write `docs/decisions.jsonl` — the design phase reads ADRs but never authors them (decisions land via debug remediation or build / ship doc-integration). Open questions in `Risks & Open Questions` are never bare — each carries 2-3 concrete options with trade-offs, a `Recommended` option and a `Default if unanswered`; the design phase states this requirement itself and does not read the plan template. |
309
+ | `bkit-design-generate` | Generate unified design draft from brainstorms and the design input; the input's `Grill decisions` section (design Step 3) holds the human's answers — `Settled` entries are strong pointers, not hard constraints — followed unless a brainstorm shows the answer fails; a departure gets a `Decisions` row and a `Risks & Open Questions` item asking the human to confirm. `Assumed, not confirmed` entries are defaults it may override with a stated reason; each `Open` entry becomes a `Risks & Open Questions` item. The section's `Draft Acceptance Scenario`, labelled `DRAFT only`, is the starting point for the design's `Acceptance Scenario`; every change to it gets a `Decisions` row, no confirmation item, because the human confirms the scenario again at the plan's `Acceptance Scenario confirmed` box. It reads the input directly because the ideate triplet may not carry every decision through. Emits `Reuse & Prior Art` (default `EXTEND`): duplication is cheapest to decide at design — net-new committed here is expensive to unwind at plan or build. Every net-new `Architecture` component traces to a `NEW` verdict; plan phase starts `Reuse Inventory` from it. Verdict only — sources, absence proof and rejected `NEW` reasons are checked by `bkit-design-review-1` and evidenced in the plan's `Search performed`. Appends `docs/deferrals.jsonl` if applicable. Does not write `docs/decisions.jsonl` — the design phase reads ADRs but never authors them (decisions land via debug remediation or build / ship doc-integration). Open questions in `Risks & Open Questions` are never bare — each carries 2-3 concrete options with trade-offs as the lettered `Options: A) {option} — {trade-off} · B) …`, a `Recommended` option and a `Default if unanswered`; the design phase states this requirement itself and does not read the plan template. |
256
310
  | `bkit-design-review-1` | Feasibility research with `WebFetch`/`WebSearch`; prior-art search per capability — repo sweep first (in-repo candidate outranks an ecosystem one; an ecosystem-only sweep cannot see the repo's own near-miss helpers), then dependency built-ins and ecosystem practice. Repo sweep — worded identically here and in plan template Pre-Flight — adds language-server symbol references and committed capability map `docs/capabilities.md` to text search. Tools named by capability, never product: prompts stay stack-independent and work with none installed. Both only add candidates — empty language-server answer = unresolved query; code map confirms presence, not absence — so absence (`NEW` reason (a)) rests on recorded text search. `EXTEND-EXISTING` heads the recommendation enum; unrebutted `NEW` verdict over live candidate = `CONCERNS` finding; writes review log. Sole emitter of the `Spike` header field (`true` when an `Unknowns & Blockers` item is settleable only by running code) — gates workflow Step 7. |
257
- | `bkit-design-review-2` | Same as `-1`; runs in parallel with cross-tier model diversity. Also audits `Risks & Open Questions` format — question missing options, `Recommended` or `Default if unanswered` → OPTIONAL. Does not emit `Spike`. |
311
+ | `bkit-design-review-2` | Same as `-1`; runs in parallel with cross-tier model diversity. Also audits `Risks & Open Questions` format — question missing options, `Recommended` or `Default if unanswered` (the lettered `Options: A) …`) → OPTIONAL. Does not emit `Spike`. |
258
312
  | `bkit-design-spike` | Conditional — runs only when `bkit-design-review-1` sets `Spike` = true. Restates each unknown as a falsifiable claim, writes the smallest probe that could disprove it under `docs/plans/{YYYYMM}/logs/{FID}/spike/`, runs it, records captured output verbatim. Settles claims by evidence only — never infers a result from source reading, never simulates an unavailable service. PASS / CONCERNS / FAIL where FAIL = a design assumption is disproven (Step 8 revises against it) and CONCERNS = unsettleable this run (assumption carried forward unproven); per-claim `SETTLED` / `DISPROVEN` / `BLOCKED` stays in the body. Disproof needs evidence that an honest attempt failed — "could not get it working" is `BLOCKED`, never `DISPROVEN`, and no claim may be weakened or stubbed to make a probe pass. Probe code is throwaway and confined to the artifact tree; never touches tracked source or a repo manifest. |
259
- | `bkit-design-update` | Apply CRITICAL items and OPTIONAL items judged on merit from both reviews; whole-doc rewrite of design. Open questions a finding adds or rewords carry options, `Recommended` and `Default if unanswered`; untouched regions stay verbatim. |
260
- | `bkit-design-organise` | Decompose `design-v2-reviewed.md` into dependency-annotated workstreams (`WS-1`, `WS-2`, …); inline `Requires` per workstream; FAIL if cycles detected. `## Suggested Workstreams` is the suggestion. Workflow Step 11 posts that list on the parent issue. It never creates tracker issues. |
313
+ | `bkit-design-update` | Apply CRITICAL items and OPTIONAL items judged on merit from both reviews; whole-doc rewrite of design. Open questions a finding adds or rewords carry options as the lettered `Options: A) …`, `Recommended` and `Default if unanswered`; untouched regions stay verbatim. Given **Update** instead of **Create**, it skips the seed and edits that design in place. An `## Owner answers` input (the design orchestrator after close) is applied as CRITICAL items: the answered question leaves `Risks & Open Questions`, a `Decision: {key}` becomes a `Decisions` item, and overridden statements follow the answer. |
314
+ | `bkit-design-organise` | Decompose `design-v2-reviewed.md` into dependency-annotated workstreams (`WS-1`, `WS-2`, …) with inline `Requires`, then group them into plans (`P-1`, …) by the deliverable test: a plan splits off only for an independent deliverable, its own rollout / rollback, its own risk or owner, or independent release; a same-change mandate keeps workstreams together. No task count or task cap. Each plan states `Deliverable`, `Covers` (every workstream in exactly one plan) and `Requires` (plan ids; cross-plan workstream edges lift to plan edges, intra-plan edges drop). FAIL if either graph has a cycle. `Suggested Plans` is the suggestion; a plan is the filed unit. Workflow Step 11 posts the plan list on the parent issue. It never creates tracker issues. |
261
315
 
262
316
  ## Plan Phase
263
317
 
264
318
  <!-- prettier-ignore -->
265
319
  | Agent | Scope |
266
320
  | --- | --- |
267
- | `bkit-plan-generate` | Read design and / or audit reports + template + docs → write new plan draft. Design absent and audits present is valid. Neither present → the issue is the source: a change-shape issue's `Narrative`, `Success Criteria`, `Edge Cases`, matrix rows and tasks carry over verbatim (Contract `Acceptance Scenario` and `Scenario detail`). Audit contract: if input includes audit-scope directives, plan covers only those findings; the rest of each report is context only, not backlog. Workstream contract: if input includes `**Workstream:**` directive, plan covers only that workstream's `## Scope` content unless stated otherwise; surface `**Requires:**` at top of plan. Process = one step per major template section in **dependency (authoring) order, not template document order** — derived sections (Breaking Changes, Supersession, Security, Docs, Rollback) follow Architecture rather than preceding it: Context → Acceptance Scenario → Derived Test Matrix → Open Questions → Pre-Flight → Alternatives → Data Schema → Reuse Inventory → Architecture (+ `decisions.jsonl`) → Breaking Changes → Supersession → Security → Docs → Rollback → Out of Scope → Suggested Scope Expansion → Agent Execution Runbook (groups of task blocks in build order, each group and task within the `bkit.workflows.tasks` caps the workflow passes in Context; pick tests from matrix only; each row → exactly one task; every task has `DONE WHEN`; `READ` / `UPDATE` / `DELETE` only for symbols it opened; never writes its own `TEST EXCEPTION`, copies an input's word for word) → Verification. The workflow runs `bkit tests plan-check --existence=refuse` on the draft straight after; exit 1 re-invokes it once to fix only the lines the report names. Human-confirmation checkboxes stay unchecked at generate. Each section group lands in the draft before reading for the next group starts, so a turn-cap stop leaves finished groups on disk, not an all-`PENDING` stub. Every `Open Questions & Assumptions` question carries 2-3 concrete options with trade-offs, a `Recommended` option and a `Default if unanswered` (design's answer carried forward where it has one); a bare question is not acceptable output. `Out of Scope` entries must be `Exempt` (genuinely-additive), else a runbook task. `Suggested Scope Expansion` holds candidates, not tasks; no tracker / ledger scan here — surface overlaps come from the orchestrator Fetch Context step via `bkit-plan-review-1`. |
268
- | `bkit-plan-review-1` | Read-only critique of the plan text — tightness + dependency-order lens; effort slot 1 (`high`). Parallel with `-2` and `bkit-plan-scope-check`. Drift from source, design-coverage gaps and unjustified expansion → CRITICAL. Tasks declare `DEPENDS ON`, which `bkit tests plan-check` checks mechanically (a missing task, the task itself, a task in a later group); review-1 derives the graph regardless: what each task produces and consumes, read against runbook order (build order), because a declared field lists only the edges its author saw. A consumer before its producer, or a prerequisite no task produces and the repo lacks, is CRITICAL. Surfaces `Suggested Scope Expansion` candidates from the `related-*.md` context. FAIL only for a task-order cycle no reorder resolves, or a plan that contradicts its source design — every other gap is CRITICAL → CONCERNS so `bkit-plan-update` repairs it. |
269
- | `bkit-plan-review-2` | Same scope — regression-risk lens, reads the code; effort slot 2 (`high`). Traces each task's files, symbols, schemas and endpoints to their consumers. `TEST FIRST` / `DONE WHEN` check current code cannot pass, no task changing it → CRITICAL, cites the ruling `file:line`. FAIL only for a regression vector the source design itself requires and no plan edit mitigates — every other gap is CRITICAL → CONCERNS so `bkit-plan-update` repairs it. |
270
- | `bkit-plan-scope-check` | Internal-completeness scope audit; one slot, parallel with both reviewers on the draft (`medium` effort, opus model — depth-sensitive lens). Owns the template checks: sections present; `Acceptance Scenario` / `Derived Test Matrix` missing or vague → CRITICAL; `Open Questions & Assumptions` question missing options, `Recommended` or `Default if unanswered` → OPTIONAL (mirrors `bkit-design-review-2` on the design side). Per-task `DONE WHEN` / `TEST FIRST` are not this lens: `bkit tests plan-check` refuses a task missing either after every plan write. `Blocking` out-of-scope deferral → CRITICAL. Duplication guard is bidirectional: `Supersession Inventory` catches code the plan replaces but never deletes; `Reuse Inventory` catches capability the plan rebuilds instead of extending. Recorded search (exact commands, not a claim) is the evidence — empty `Search performed` = unsearched, not absence; untraced `[NEW]` → CRITICAL. FAIL only for a matrix-to-runbook mapping no plan edit recovers — a broken template and every other gap is CRITICAL → CONCERNS so `bkit-plan-update` repairs it. |
271
- | `bkit-plan-update` | Merge review / scope critique notes → in-place plan edits + record log. Invariant: every task's `**DONE WHEN:**` and `**TEST FIRST:**` fields survive critique-merge (may be tightened, `n/a` → named tests included; never removed / weakened). Writes review's `Suggested Scope Expansion` candidates into the plan section (advisory; never tasks). Open questions a finding adds or rewords carry options, `Recommended` and `Default if unanswered`; untouched regions stay verbatim. Human-confirmation checkboxes survive the merge as found — an unconditional invariant beside `DONE WHEN`, not a clause of the open-question rule. Mirrors to `docs/deferrals.jsonl` only the `Out of Scope` `Exempt` rows its merge adds; `Suggested Scope Expansion` candidates are never appended — most were read from that ledger, so appending duplicates them. Writes a `docs/decisions.jsonl` row per decision its merge adds or changes, same `key` when changed, and keeps the plan's `**Ledger rows:**` line current for plan close's `bkit ledgers check`. |
321
+ | `bkit-plan-generate` | Read design and / or audit reports + template + docs → write new plan draft. Design absent and audits present is valid. Neither present → the issue is the source: a change-shape issue's `Narrative`, `Success Criteria`, `Edge Cases`, matrix rows and tasks carry over verbatim (Contract `Acceptance Scenario` and `Scenario detail`). Audit contract: if input includes audit-scope directives, plan covers only those findings; the rest of each report is context only, not backlog. Workstream contract: if input includes `**Workstream:**` directive, plan covers only that workstream's `Scope` content unless stated otherwise; surface `**Requires:**` at top of plan. Process = one step per major template section in **dependency (authoring) order, not template document order** — derived sections (Breaking Changes, Supersession, Security, Docs, Rollback) follow Architecture rather than preceding it: Context → Acceptance Scenario → Derived Test Matrix → Open Questions → Pre-Flight → Alternatives → Data Schema → Reuse Inventory → Architecture (+ `decisions.jsonl`) → Breaking Changes → Supersession → Security → Docs → Rollback → Out of Scope → Suggested Scope Expansion → Agent Execution Runbook (groups of task blocks in build order, each group and task within the `bkit.workflows.tasks` caps the workflow passes in Context; pick tests from matrix only; each row → exactly one task; every task has `DONE WHEN`; `READ` / `UPDATE` / `DELETE` only for symbols it opened; never writes its own `TEST EXCEPTION`, copies an input's word for word) → Verification. The workflow runs `bkit tests plan-check --existence=refuse` on the draft straight after; exit 1 re-invokes it once to fix only the lines the report names. Human-confirmation checkboxes stay unchecked at generate. Each section group lands in the draft before reading for the next group starts, so a turn-cap stop leaves finished groups on disk, not an all-`PENDING` stub. Every `Open Questions & Assumptions` question carries 2-3 concrete options with trade-offs as the lettered `Options: A) {option} — {trade-off} · B) …`, a `Recommended` option and a `Default if unanswered` (design's answer carried forward where it has one); a bare question is not acceptable output. `Out of Scope` entries must be `Exempt` (genuinely-additive), else a runbook task. `Suggested Scope Expansion` holds candidates, not tasks; no tracker / ledger scan here — surface overlaps come from the orchestrator Fetch Context step via `bkit-plan-review-1`. |
322
+ | `bkit-plan-review-1` | Read-only critique of the plan text — tightness + dependency-order lens; effort slot 1 (`high`). Parallel with `-2` and `bkit-plan-scope-check`. Drift from source, design-coverage gaps and unjustified expansion → CRITICAL. Tasks declare `DEPENDS ON`, which `bkit tests plan-check` checks mechanically (a missing task, the task itself, a task in a later group); review-1 derives the graph regardless: what each task produces and consumes, read against runbook order (build order), because a declared field lists only the edges its author saw. A consumer before its producer, or a prerequisite no task produces and the repo lacks, is CRITICAL. Surfaces `Suggested Scope Expansion` candidates from the `related-*.md` context. A group that leaves a test failing which only a later group's task fixes is CRITICAL, because the build checks each group before the next. FAIL only for a task-order cycle no reorder resolves, or a plan that contradicts its source design — every other gap is CRITICAL → CONCERNS so `bkit-plan-update` repairs it. Marks each CRITICAL `(plan edit)` or `(design change)`; header `Redesign: true` when any is a design change, which sends a `FAIL` review straight to the Failure Path instead of an apply round and its one re-review. |
323
+ | `bkit-plan-review-2` | Same scope — regression-risk lens, reads the code; effort slot 2 (`high`). Traces each task's files, symbols, schemas and endpoints to their consumers. `TEST FIRST` / `DONE WHEN` check current code cannot pass, no task changing it → CRITICAL, cites the ruling `file:line`. A `TEST FIRST` test whose logic already passes on current code or an earlier task's code → CRITICAL, citing the code that satisfies it: new behaviour needs a new test, judged by reading the test, since no test runs at plan time. A planned test whose fake or fixture value differs from what the cited producing code (`file:line`) returns is CRITICAL, and so is a `TEST FIRST` test that only compares against a regenerated baseline (snapshot, coverage list, generated doc), which regenerating makes pass. FAIL only for a regression vector the source design itself requires and no plan edit mitigates — every other gap is CRITICAL → CONCERNS so `bkit-plan-update` repairs it. Marks each CRITICAL `(plan edit)` or `(design change)`; header `Redesign: true` when any is a design change, which sends a `FAIL` review straight to the Failure Path instead of an apply round and its one re-review. |
324
+ | `bkit-plan-scope-check` | Internal-completeness scope audit; one slot, parallel with both reviewers on the draft (`medium` effort, opus model — depth-sensitive lens). Owns the template checks: sections present; `Acceptance Scenario` / `Derived Test Matrix` missing or vague → CRITICAL; `Open Questions & Assumptions` question missing options, `Recommended` or `Default if unanswered` (the lettered `Options: A) …`) → OPTIONAL (mirrors `bkit-design-review-2` on the design side). Per-task `DONE WHEN` / `TEST FIRST`, the matrix rows' task mapping and each `DELETE`'s `Supersession Inventory` row are not this lens: `bkit tests plan-check` refuses each after every plan write. `Blocking` out-of-scope deferral → CRITICAL. Duplication guard is bidirectional: `Supersession Inventory` catches code the plan replaces but never deletes; `Reuse Inventory` catches capability the plan rebuilds instead of extending. Recorded search (exact commands, not a claim) is the evidence — empty `Search performed` = unsearched, not absence; untraced `[NEW]` → CRITICAL. FAIL only for a `Blocking` out-of-scope deferral no plan edit recovers — a broken template and every other gap is CRITICAL → CONCERNS so `bkit-plan-update` repairs it. Marks each CRITICAL `(plan edit)` or `(design change)`; header `Redesign: true` when any is a design change, which sends a `FAIL` review straight to the Failure Path instead of an apply round and its one re-review. |
325
+ | `bkit-plan-update` | Merge review / scope critique notes → in-place plan edits + record log. Invariant: every task's `**DONE WHEN:**` and `**TEST FIRST:**` fields survive critique-merge (may be tightened, `n/a` → named tests included; never removed / weakened). Writes review's `Suggested Scope Expansion` candidates into the plan section (advisory; never tasks). Open questions a finding adds or rewords carry options as the lettered `Options: A) …`, `Recommended` and `Default if unanswered`; untouched regions stay verbatim. Human-confirmation checkboxes survive the merge as found — an unconditional invariant beside `DONE WHEN`, not a clause of the open-question rule. Mirrors to `docs/deferrals.jsonl` only the `Out of Scope` `Exempt` rows its merge adds; `Suggested Scope Expansion` candidates are never appended — most were read from that ledger, so appending duplicates them. Writes a `docs/decisions.jsonl` row per decision its merge adds or changes, same `key` when changed, and keeps the plan's `**Ledger rows:**` line current for plan close's `bkit ledgers check`. Given **Update** instead of **Create** (the re-review round, or the plan orchestrator after close), it skips the seed and edits that plan in place. An `## Owner answers` input is applied as CRITICAL items: the answered question leaves `Open Questions & Assumptions`, a `Decision: {key}` joins the `**Ledger rows:**` line, and overridden tasks and statements follow the answer. |
272
326
 
273
327
  ## Build Phase
274
328
 
275
329
  <!-- prettier-ignore -->
276
330
  | Agent | Scope |
277
331
  | --- | --- |
278
- | `bkit-build-plan-convert` | Write the plan's `Agent Execution Runbook` as groups of task blocks when `bkit tests plan-check` finds none, from whatever input arrives in `build-input-{RUN}.md` (session prompt, issue text, plan as it arrived) and the plan file. Keeps every section outside the runbook and every end state and test the inputs name, tightened and never weakened. Same scope: adds no work the issue or plan does not ask for, drops none. Group and task caps: the `bkit.workflows.tasks` values the workflow passes in Context. Names only files it opened; a symbol it did not find → the file alone. A `NEW` `Reuse Inventory` row per `CREATE` file. Never writes `TEST EXCEPTION` of its own and never authors a `bkit:anchor` block; copies each `TEST EXCEPTION` an input writes word for word (a request the owner confirms at the build's Decision Pause). Where the plan and the issue disagree, the plan wins. Runs the check command the orchestrator supplies, at most 3 times. No verdict header. |
279
- | `bkit-build-plan-current` | Judge whether changes landed on `origin/dev` since the plan's `bkit:anchor` `baseSha` leave its tasks executable. Derives its own changed-path list from `BASE_SHA` (orchestrator-supplied via Context), intersects it against the paths each task depends on, and reads per-path diffs for the intersection only — no whole-range diff, no path outside it. Same output shape as `bkit-build-challenge-*`: header count + one append-only findings section. `Executable` tasks are not findings and are not logged; `Reword` / `Broken` are, and `Affected` = their count. PASS = no affected task; CONCERNS = any `Reword`; FAIL = any `Broken`. All three are advisory — build proceeds on every verdict and `@bkit-build-plan-execute` reads the findings as its currency input. Judges the plan against the codebase — the design doc is not an input and a superseded design is not a finding. Commit distance and elapsed time never carry the verdict. Does not revise the plan: remediation is a `/bkit-plan` re-run, never `bkit-plan-update` from build. |
332
+ | `bkit-build-plan-convert` | Write the plan's `Agent Execution Runbook` as groups of task blocks when `bkit tests plan-check` finds none, from whatever input arrives in `build-input-{RUN}.md` (session prompt, issue text, plan as it arrived) and the plan file. Keeps every section outside the runbook and every end state and test the inputs name, tightened and never weakened. Same scope: adds no work the issue or plan does not ask for, drops none. Group and task caps: the `bkit.workflows.tasks` values the workflow passes in Context. The plan's `Derived Test Matrix` keeps a `Task` column beside `Test description`, `Type` and `Target file`, each cell naming the one runbook task that lands that row's test. Names only files it opened; a symbol it did not find → the file alone. A `NEW` `Reuse Inventory` row per `CREATE` file. Never writes `TEST EXCEPTION` of its own and never authors a `bkit:anchor` block; copies each `TEST EXCEPTION` an input writes word for word (a request the owner confirms at the build's Decision Pause). Where the plan and the issue disagree, the plan wins. Runs the check command the orchestrator supplies, at most 3 times. No verdict header. |
333
+ | `bkit-build-plan-current` | Judge whether changes landed on `origin/dev` since the plan's `bkit:anchor` `baseSha` leave its tasks executable. Runs `bkit tasks currency --plan= --base={BASE_SHA} --against=origin/dev` — the engine records each task's in-scope paths and symbols in the plan block and prints only the tasks whose path or symbol moved in `{BASE_SHA}...origin/dev`, each with its matched paths, needles and diff hunks — and reads those hunks only, no whole-range diff, no path outside a matched set. Same output shape as `bkit-build-challenge-*`: header count + one append-only findings section. `Executable` tasks are not findings and are not logged; `Reword` / `Broken` are, and `Affected` = their count. PASS = no affected task; CONCERNS = any `Reword`; FAIL = any `Broken`. All three are advisory — build proceeds on every verdict and `@bkit-build-plan-tasks` reads the findings as its currency input. Judges the plan against the codebase — the design doc is not an input and a superseded design is not a finding. Commit distance and elapsed time never carry the verdict. Does not revise the plan: remediation is a `/bkit-plan` re-run, never `bkit-plan-update` from build. |
280
334
  | `bkit-build-install-skills` | Identify and install relevant skills via `npx skills find`/`add`: quality-hurdle the source (official publisher, or 1000+ installs / 100+ stars), install one named skill per gap, report provenance. |
281
- | `bkit-build-plan-execute` | Execute one plan group — the task blocks in the brief `bkit context brief --next` writes; earlier tasks in the brief are context only; the brief's `Deviations` section, when present, is the owner's run instruction, applied to the tasks it names. An existing log is kept and its `Verdict` reset to `PENDING`. Verdict PASS / HANDOVER / FAIL: `HANDOVER` = turn cap near with tasks neither `DONE` nor `BLOCKED` → stop after the current task's log line; the workflow re-runs the group once with a fresh brief, not as a failed attempt. Per task naming `TEST FIRST` tests: tests + stubs first, then the red command the orchestrator supplies in Context (`bkit tests red`) once per test — never the task's code before every red is recorded; skipped for tasks logged `DONE — TASK {N}: already landed`; refused as `passed` → `BLOCKED` (never undo code to force a red); no red for tests a task's `TEST EXCEPTION` field names, never touches one; other refusals → fix test / stub and re-run; `bkit:test:file` not declared → `red not enforced`; could not run → `BLOCKED` (`.bkit/workflows.md` § Test-first). Branches on the outcome the command prints, not its exit code; honours the brief's `Reuse Inventory` rows and `Architecture` entries at the only point the decision can still be quietly discarded — `EXTEND` row modifies named artifact, never a parallel file; rows or entries the brief left out (8 KB work cap) → reads the plan section it names; reusable artifact found mid-build folds in, never forks; reads each task's `DONE WHEN` field as the completion criterion to satisfy; modify source; write stub artifact first then incremental `## Task Log` entries — one `DONE — TASK {N}:` or `BLOCKED — TASK {N}:` line per task at line start, which the next brief reads to skip finished tasks. Reads `build-plan-current.md` when present: findings are fact, plan is intent — `Reword` tasks build at the paths the report names as current, `Broken` tasks whose work already landed log `DONE — TASK {N}: already landed`, the rest build the intent against current code. Never edits the plan file; tasks absent from the findings follow plan wording verbatim. Does not tick plan checkboxes (verify owns that). Does not write `docs/deferrals.jsonl` — a blocked task is a `BLOCKED` log line for verify to catch, not a deferral. |
282
- | `bkit-build-plan-verify` | Verify implementation vs plan; writes Plan Verification log. `Group` mode judges one group's briefs against the group diff (changes since the group's start commit); `Holistic` mode runs once after the last group, across every group, plus integration. Ticks plan `- [ ] **TASK {N}:**` → `[x]` only when the task's `DONE WHEN` field is satisfied in the codebase and the orchestrator's test-first audit report (`bkit tests green`) shows no gap for it — `not-green` `Fixable: true`; `test-changed` / `red-mismatch` judged from the diff; `added lines only` diff read, a line added inside a planned test or its setup → `test-changed`; `red-mismatch` for a planned test renamed or moved and passing now → no gap; `no-red` on an already-landed task (plan-currency `Broken` + execute `DONE — TASK {N}: already landed`) → no gap; `delete-still-referenced` → each hit read, a different thing with the same name → no gap, else `Fixable: true`; other `no-red`, `late-red`, `exception-refused` and the plan-defect kinds `Fixable: false`; each gap logged `TASK {N}` with `— Fixable: true` / `— Fixable: false`, header `Fixable: true` when any gap is (build → fixable-only remediate, then Decision Pause); test-order gaps logged with task, test file and test name; a test a task's `TEST EXCEPTION` field names is no gap; `Late-red check` tasks → `Notes` — a `BLOCKED` log line or a deferral row never discharges a task, so unfinished work routes to `@bkit-build-plan-gaps-remediate` in-loop instead of surfacing at ship. Missing `DONE WHEN` on any task → FAIL (template structure broken; remediation cannot recover). |
283
- | `bkit-build-plan-gaps-remediate` | Fix specific gaps from the group's latest verify log + group gate log failures (the tests of the groups so far; that gate MUST pass too); reads the group's briefs for its task blocks; Standard / Pivot / Final escalation. Context `Scope: fixable only` → `— Fixable: false` gaps out of scope. Header `Fixed` = gaps `DONE` this attempt (0 → build Decision Pause). A supplied decision record → the owner's `Decision:` words steer the round. No strategy records a red by undoing or swapping out code; test-order gap → `BLOCKED`; never touches a `TEST EXCEPTION` field (`.bkit/workflows.md` § Test-first). |
284
- | `bkit-build-challenge-1` | Adversarial review of `git diff origin/dev...HEAD` — bug-hunting lens; effort slot 1 (`high`). Parallel with `-2`. Writes each finding as one task block — `Severity`, `Class` (`mechanical\|architectural`), `Fixable`, `Location`, `DONE WHEN`, `TEST FIRST` within the per-task `bkit.workflows.tasks` caps the workflow passes in Context — that `bkit challenge merge` appends to the plan as a new last group. Never references sibling output (circular reasoning guard). |
335
+ | `bkit-build-plan-tests` | Write one task group's planned tests — every `TEST FIRST` test of the task blocks in the brief `bkit context brief --group --tests-only --file` writes, which leaves out the tests the group's red rows already proved. Each test fails on the behaviour its `fails on` text names, is lint-clean, and never names a test the task's `TEST EXCEPTION` names; a stub only where a test cannot load without one; no source code (`steps finish --step=build-plan-tests --plan` refuses a diff touching any other file, exit 15). Flat log `build-plan-tests-{GROUP}-log.md`: ``WROTE — TASK {N}: `{file}` `{name}` ``, ``PASSES — TASK {N}: `{file}` `{name}`: {reason}`` (the behaviour already exists; `bkit tests red` records the test `passed-accepted`), ``BLOCKED — TASK {N}: `{file}` `{name}` — {reason}``, `SOURCE — TASK {N}: {value} from {file}:{line}` (where a value the test checks or fakes comes from). Verdict PASS / HANDOVER / FAIL: `HANDOVER` = turn cap near with tests left; the workflow re-runs the tests half once with a fresh brief. Re-invoked with a red report → rewrites only the tests it lists as `vacuous`, `missing`, `no-report` or `lint`. Never runs `bkit tests red` itself; the orchestrator runs it once per group after the step. |
336
+ | `bkit-build-plan-tasks` | Execute one task group — the task blocks in the brief `bkit context brief --next` writes; earlier tasks in the brief are context only; the brief's `Deviations` section, when present, is the owner's run instruction, applied to the tasks it names. An existing log is kept and its `Verdict` reset to `PENDING`. Verdict PASS / HANDOVER / FAIL: `HANDOVER` = turn cap near with tasks neither `DONE` nor `BLOCKED` → stop after the current task's log line; the workflow re-runs the group once with a fresh brief, not as a failed attempt. The group's `TEST FIRST` tests are already on disk, written by `bkit-build-plan-tests` and recorded red by the engine before this step: makes them pass and never edits, renames, moves or deletes a planned test or a `TEST EXCEPTION` field (`steps finish` refuses the step when a planned test file differs from its red blob, exit 11); a test recorded `passed-accepted` → `DONE — TASK {N}: already landed` when `DONE WHEN` already holds with no task code written; never undoes code to force a red (`.bkit/workflows.md` § Test-first). Honours the brief's `Reuse Inventory` rows and `Architecture` entries at the only point the decision can still be quietly discarded — `EXTEND` row modifies named artifact, never a parallel file; rows or entries the brief left out (8 KB work cap) → reads the plan section it names; reusable artifact found mid-build folds in, never forks; reads each task's `DONE WHEN` field as the completion criterion to satisfy; modify source; write stub artifact first then incremental `Task Log` entries — one `DONE — TASK {N}:` or `BLOCKED — TASK {N}:` line per task at line start, which the next brief reads to skip finished tasks. Reads `build-plan-current.md` when present: findings are fact, plan is intent — `Reword` tasks build at the paths the report names as current, `Broken` tasks whose work already landed log `DONE — TASK {N}: already landed`, the rest build the intent against current code. Never edits the plan file; tasks absent from the findings follow plan wording verbatim. Does not tick plan checkboxes (verify owns that). Does not write `docs/deferrals.jsonl` — a blocked task is a `BLOCKED` log line for verify to catch, not a deferral. |
337
+ | `bkit-build-plan-verify` | Verify implementation vs plan; writes Plan Verification log. `Group` mode judges one group's briefs against the group diff (changes since the group's start commit); `Holistic` mode runs once after the last group, across every group, plus integration. Ticks are the engine's: the header's `Tasks` line names the task numbers the codebase satisfies and `bkit tasks tick --green=` writes the `[x]`; the prompt edits no plan and never ticks a box. `bkit tests green` prints `verdict` and a `fixable` per gap, so the prompt holds no kind→`Fixable` table and no verdict mapping. Its reads decide only a `null` gap: `delete-still-referenced` → each hit read, a different thing with the same name → no gap, else `Fixable: true`; each `Updated tests` entry → the test's diff since its `base` read: only added expectations → no gap, a removed or loosened expectation → a `Fixable: false` gap; `Unproven` tests → `Notes`, no gap, because the owner answers them once at the build's end. Each gap is appended as the report's `logLine` with the decided `Fixable` in place of `null`; every other gap keeps its printed `fixable`; header `Failures` = unresolved gap count and `Fixable` = `true` when any is (build → fixable-only remediate, then Decision Pause); a test a task's `TEST EXCEPTION` field names is no gap; `Already landed` tests → `Notes` — a `BLOCKED` log line or a deferral row never discharges a task, so unfinished work routes to `@bkit-build-plan-gaps-remediate` in-loop instead of surfacing at ship. Missing `DONE WHEN` on any task → FAIL (template structure broken; remediation cannot recover). |
338
+ | `bkit-build-plan-gaps-remediate` | Fix specific gaps from the group's latest verify log + group gate log failures (the tests of the groups so far); a Context `Group gate` command is re-run after each fix until exit 0 — each newly exposed failure is a gap of the same round, other tests built on the same failing assumption are swept first — and logged `DONE — Group gate: exit 0`, which `PASS` needs; reads the group's briefs for its task blocks; Standard / Pivot / Final escalation. Context `Scope: fixable only` → `— Fixable: false` gaps out of scope. Header `Fixed` = `DONE — Gap` lines this attempt (0 → build Decision Pause). A supplied decision record → the owner's `Decision:` words steer the round. No strategy records a red by undoing or swapping out code, or rewrites a planned test case (`steps finish` exit 11 names the case); other tests in the same file are its to change; never touches a `TEST EXCEPTION` field (`.bkit/workflows.md` § Test-first). |
339
+ | `bkit-build-challenge-1` | Adversarial review of `git diff origin/dev...HEAD` — bug-hunting lens; effort slot 1 (`high`). Parallel with `-2`. Writes each finding as one task block — `Severity`, `Class` (`mechanical\|architectural`), `Fixable`, `Location`, `DONE WHEN`, `TEST FIRST` within the per-task `bkit.workflows.tasks` caps the workflow passes in Context — that `bkit challenge merge` appends to the plan as a new last group. Reads each `TEST FIRST` test's logic against the code the diff leaves: it must fail there, or `tests red` refuses it as `passed` and the fix task blocks; a finding an existing plan task or test already covers gets no block. Never references sibling output (circular reasoning guard). |
285
340
  | `bkit-build-challenge-2` | Same as `-1` — security lens; effort slot 2 (`high` on Claude / opus — depth-sensitive; Grok on Cursor — first-party depth, same native pool as `-1`). |
286
341
 
287
- > **Cross-phase reuse:** Build Steps 12-13 + Ship Steps 11-12 invoke `bkit-debug-failure-diagnose` / `bkit-debug-failure-remediate` (in-build test failures, in-ship merge failures) — see Debug Phase entries.
342
+ > **Cross-phase reuse:** Build Steps 12-13 and Ship Steps 11-12 invoke `bkit-debug-failure-diagnose` / `bkit-debug-failure-remediate` (in-build test failures, in-ship merge failures). See the Debug Phase entries.
288
343
 
289
344
  ## Debug Phase
290
345
 
@@ -292,7 +347,7 @@ Canonical per-agent metadata — **single source of truth for model per agent**.
292
347
  | Agent | Scope |
293
348
  | --- | --- |
294
349
  | `bkit-debug-failure-reproduce` | Bug report, no failing test → co-located tests failing on the reported symptoms, lowest tier reaching it; no implementation changes. Max 3 rewrites / 3 paths; not reproduced → test changes removed, `Reproduced: NO`. `Mode: SOAK` → tests repeat the scenario until they fail reliably. Wired into `/bkit-debug` § Failure Reproduce on `[repro]` + first-attempt PASS; `YES` → Run Tests records the red, or passes with no fix → Summary (`CONCERNS`: `Reproduced: YES` with no remediate / soak log); `NO` → Summary. |
295
- | `bkit-debug-failure-diagnose` | Investigate test / CI failure → diagnosis log; no fixes. Standard / Pivot / Final by attempt. Folds `docs/knowledge.jsonl` by `key`, highest `ts` wins → reads only winning rows where `status` is `ACTIVE` or absent → cites matching entries in `## Prior Knowledge` → uses them to seed hypotheses. Output uses append-only ReAct `Hypothesis {N}` blocks; refuted hypotheses must remain. A supplied decision record → the owner's `Decision:` words steer the diagnosis. |
350
+ | `bkit-debug-failure-diagnose` | Investigate test / CI failure → diagnosis log; no fixes. Standard / Pivot / Final by attempt. Folds `docs/knowledge.jsonl` by `key`, highest `ts` wins → reads only winning rows where `status` is `ACTIVE` or absent → cites matching entries in `Prior Knowledge` → uses them to seed hypotheses. Output uses append-only ReAct `Hypothesis {N}` blocks; refuted hypotheses must remain. A supplied decision record → the owner's `Decision:` words steer the diagnosis. |
296
351
  | `bkit-debug-failure-remediate` | Apply diagnosis fix; lint / typecheck / test gate; may append `knowledge.jsonl`/`decisions.jsonl`. |
297
352
  | `bkit-debug-failure-soak` | One accumulating context across iterations (reads its prior `debug-failure-soak-*-log.md`); applies the single next instrument-or-fix step against the failing boundary; refuses to theorise; adds `[DEBUG-OBSERVE]` scaffolding and strips every marker before declaring a fix (no-leak). Wired into `/bkit-debug` § Failure Remediate Soak (`DEBUG_MODE` = SOAK); the orchestrator re-runs the tests and loops it within `SOAK_ATTEMPT` (`debug.soak.maxAttempts`, counted across every debug run of the feature). Flat append-only `DONE`/`BLOCKED` log. |
298
353
  | `bkit-debug-abort-handover` | Read all prior diagnose / remediate / verify logs → handover doc + `failures.jsonl` / `deferrals.jsonl` append. Called from build / debug / ship Failure Paths. |
@@ -302,16 +357,16 @@ Canonical per-agent metadata — **single source of truth for model per agent**.
302
357
  <!-- prettier-ignore -->
303
358
  | Agent | Scope |
304
359
  | --- | --- |
305
- | `bkit-ship-conflict-resolve` | Mid-rebase mechanical conflict resolver. Invoked from `/bkit-ship` Step 4 B2 when the PR conflicts textually with `origin/dev`: rebases onto `origin/dev`, runs `bkit branches rebase` first (unions ledgers, regenerates the lockfile), then hand-resolves only what that leaves conflicted. Allowed: non-overlapping additive edits, formatting drift, dep-version supersedes, additive JSON entries. Refused: same-function edits, semantic overlaps, ambiguous intent — FAIL aborts the rebase cleanly and routes to `bkit-debug-abort-handover`. Verifies via `bkit:gate:install-static-unit-build` before PASS. |
306
- | `bkit-ship-deferred-sweep` | Pre-merge completeness gate. Builds the intended-delivery set from plan `Acceptance Scenario` sections + `DONE WHEN` fields and asserts each against the codebase. No plan file → `ship-input-*.md` is the plan; intended-delivery set = its requested outcomes. An unmet / unplanned gap → FAIL → Failure Path (merge never happens) ONLY on >=1 of four blocking tests, each asserted against source, never inferred from a plan sentence: behaviour (absent / stub / wrong result), safety (guard weakened, known-broken path ships), false green (this diff disabled a covering test), consumer break (shipped surface fails at runtime). Lane results are not evidence — bkit lanes are event-scoped, some push-to-`dev` only, so a pre-merge sweep reading them re-creates the unsatisfiable refusal that made gate code 16 opt-in; the gate + capstone own green. A gap passing none is `Lazy`: stale prose / comment / name, doc drift, unwritten ledger row, formatting, future-enhancement `TODO`, `DONE WHEN` met in substance not literal wording. Branch is green by this step, so aborting on a record-keeping gap leaves the work unmerged and the record unwritten. Declared deferrals (out-of-scope / future-work / non-goals across design, plan, `docs/deferrals.jsonl`, summaries) form the exempt set — never block, resurfaced verbatim in the ship summary's `Deferred & Out-of-Scope (Resurfaced)` section. Exemption is EARNED, not self-granted: each entry's `reason` (`.bkit/jsonl.md` § Deferral Reason) is adjudicated against the four tests, and one that fails moves to that same third track `Lazy (unjustified)` — same-`key` ledger append setting `reason` = `unjustified`, resurfaced as follow-up, never blocking the merge (a bounded fix is cheaper than an aborted ship). Both `Lazy` sources share one append shape; every item carries its row before the verdict is set. The intended-delivery set is those two sources and NOTHING else — every precondition, build process step and merge condition a plan states is advisory, never a gate: plan `Gate N` checkboxes, `Verification Plan` boxes and `Manual Verification` steps are build-phase process, never deliverables, and no phrasing in a plan, log or phase summary makes one blocking (the gate reaches only what the repo can be made to prove; the plan template carries the matching authoring rule). Invoked from `/bkit-ship` Step 5 (after Conflict Resolve, before Merge). |
307
- | `bkit-ship-update-docs` | Cross-reference changed files against every doc in `docs.md`; update stale. |
360
+ | `bkit-ship-conflict-resolve` | Mid-merge mechanical conflict resolver. Invoked from `/bkit-ship` Step 4 when `bkit branches sync --merge` exits 24: the merge of `origin/dev` is in progress with its ledgers, lockfile, generated register, embedding docs and install record already resolved, and the agent hand-resolves only the paths left conflicted, then re-runs the block, which commits the merge. Allowed: non-overlapping additive edits, formatting drift, dep-version supersedes, additive JSON entries. Refused: same-function edits, semantic overlaps, ambiguous intent — FAIL aborts the merge cleanly and routes to `bkit-debug-abort-handover`. Verifies via `bkit:gate:install-static-unit-build` before PASS. |
361
+ | `bkit-ship-deferred-sweep` | Pre-merge completeness gate. Builds the intended-delivery set from plan `Acceptance Scenario` sections + `DONE WHEN` fields and asserts each against the codebase. No plan file → `ship-input-*.md` is the plan; intended-delivery set = its requested outcomes. An unmet / unplanned gap → FAIL → Failure Path (merge never happens) only on >=1 of four blocking tests, each asserted against source, never inferred from a plan sentence: behaviour (absent / stub / wrong result), safety (guard weakened, known-broken path ships), false green (`bkit branches scan-diff --disabled-tests` names candidates; every hit is adjudicated against the base diff and current source), consumer break (shipped surface fails at runtime). Removed assertions need precise evidence of enabled equivalent coverage or an owner-approved retirement that leaves no current requirement uncovered; record each disposition under `Test Coverage Review`. Scan errors, actual disabling markers and missing or inconclusive coverage stay blocking, never exempt or lazy. Lane results are not evidence — bkit lanes are event-scoped, some push-to-`dev` only, so a pre-merge sweep reading them re-creates the unsatisfiable refusal that made gate code 16 opt-in; the gate + capstone own green. A gap passing none is `Lazy`: stale prose / comment / name, doc drift, unwritten ledger row, formatting, future-enhancement `TODO`, `DONE WHEN` met in substance not literal wording. Branch is green by this step, so aborting on a record-keeping gap leaves the work unmerged and the record unwritten. Declared deferrals (out-of-scope / future-work / non-goals across design, plan, `docs/deferrals.jsonl`, summaries) form the exempt set — never block, resurfaced verbatim in the ship summary's `Deferred & Out-of-Scope (Resurfaced)` section. Exemption is earned, not self-granted: the prompt holds no reason test — `bkit ledgers triage --fid= --plan=` applies the deferral tests and writes every `unjustified` / `TRIAGED` row itself, printing each plan-related row's `key`, `bucket` and test result, and one that fails moves to that third track `Lazy (unjustified)`, resurfaced as follow-up, never blocking the merge (a bounded fix is cheaper than an aborted ship). Both `Lazy` sources share one append shape; every item carries its row before the verdict is set. The intended-delivery set is those two sources and nothing else — every precondition, build process step and merge condition a plan states is advisory, never a gate: plan `Gate N` checkboxes, `Verification Plan` boxes and `Manual Verification` steps are build-phase process, never deliverables, and no phrasing in a plan, log or phase summary makes one blocking (the gate reaches only what the repo can be made to prove; the plan template carries the matching authoring rule). Invoked from `/bkit-ship` Step 5 (after Conflict Resolve, before Merge). |
362
+ | `bkit-ship-update-docs` | Cross-reference changed files against every doc in `docs.md`; update stale. `Pending Items` comes from the `ship-pending-items.md` deferral scan the workflow writes with `bkit context ledger-scan --terms={FID} --source=docs/deferrals.jsonl`; the agent writes no ledger row and reads no `docs/deferrals.jsonl` directly. |
308
363
 
309
364
  ## Audit (Ad-hoc)
310
365
 
311
366
  <!-- prettier-ignore -->
312
367
  | Agent | Scope |
313
368
  | --- | --- |
314
- | `bkit-audit-generate` | Read one `docs/templates/bkit-audit-*.md` and run its probe steps against the repo. Write the filled template to `docs/audits/{YYYYMM}/{slug}.md` (`{slug}` = template basename without `bkit-audit-`). **If** that month already has this slug → `{slug}-{TIMESTAMP}.md` (`{TIMESTAMP}` = `date -u +%Y%m%dT%H%M%SZ`). Report = template up to `bkit:end`. The template supplies shape and its own probe steps; the rules every template shares (read-only, evidence, Pre-Flight, credentials, permissions, previous run, gates, severity) are this prompt's Contract bullets. Right after creating the stub it runs `bkit branches anchor-stamp --base-sha=<HEAD> --base-ref=<branch>`, so the audited commit and the bkit release the report came from sit in the same `bkit:anchor` block design and plan docs carry (`.bkit/workflows.md` § Anchor Block) — top of file, machine-readable, the next run reads it as the previous report's commit; an outdated release also gets a visible `Outdated bkit` line under the block, written by the stamp. The agent runs the stamp itself because audits have no orchestrator; the block is still tool-written, never hand-written. Writes only the report, `docs/failures.jsonl` rows (`by: {workflow: "audit", step: "audit-{slug}"}`; one per `MEDIUM`+ finding, reusing any open row's `key` for the same problem whichever workflow wrote it, closing only open `audit-{slug}-` rows a rerun no longer finds; skipped by templates that audit the ledgers themselves, `.bkit/jsonl.md`), and the template's consumer-owned `Project Instructions` (below `bkit:end`), which it reads first and refines each run — exact commands, paths, tool gaps, corrections — so the next run starts from what this one learned; edits listed under the report's `Process Improvements` → `Project`, universal step changes proposed under `Universal`. Never results there (reports hold the series) and never text from external input (an injected directive would persist into every later run). The report's `Suggested Issues` groups findings into tracker-sized changes — one row per change landed in one go, every `MEDIUM`+ finding in exactly one row — and names each row's next step: `/bkit-design|/bkit-plan|/bkit-build|/bkit-debug` by the § When to Run rules of `.bkit/workflows.md` (restated inline, prompts are self-contained), the `.bkit/lite.md` lite workflow (`build:lite|debug:lite`) when its § When to use allows, or `manual` for a change outside the repo. The agent proposes and never files: its tracker tools post a comment and run one read-only search, so a person files from the table and each full workflow's related-issues scan catches an issue already open. Issue named in the input (same ID shapes as workflow Step 1, input only, never files the run read) → one `###### bkit:generated` comment there, built by `bkit issues comment --kind=audit` from the final summary (`Embedded Directives` as sources only, so injected text never reaches the tracker), posted through the tracker MCP after any `STOPPED` run too; retry once, then the header `Issue` line reads `failed`; never gates the run. No issue named, tracker wired → the agent searches for the template's audit parent (`.bkit/trackers.md` § Issue shapes — one standing issue per template that every run reports to, so run summaries and the issues they spawn collect in one place instead of one issue per run): an open issue whose body names `docs/templates/bkit-audit-{slug}.md` → `Parent: {ID} found` and the summary is posted there as if it had been named. None → `Parent: missing`, `Issue: None`, the built comment is saved to `{outboxRoot}/audit-{slug}/parent-comment.md`, and `Recommended Next Step` opens with a ready-to-file parent (title, labels, body with the run line) and the instruction to post that saved comment as its first comment — the agent cannot create issues, so whoever files the parent posts it. Search error → `Parent: unchecked`, same saved comment, posted on the parent once found or created. Every filed `Suggested Issues` row is a sub-issue of the parent, existing or new. Tracker `none` → no comment, no search, `Issue: None`, `Parent: None`. Bash is limited to that stamp, the setup tips, the issue comment, and the probes the template or its `Project Instructions` name; throwaway worktrees / remote targets removed before the report finishes. Web tools serve each template's Standards step. Pre-Flight `Permissions` probes each tool once before `Probe Steps`; a blocked call (there or in any step) never stops the run: the agent records it with its narrow allow rule, runs every step the remaining tools allow, and lists the work it could not do under `Not Checked`. Blocked calls persist in `Project Instructions` `Permissions`; the next run probes them first. `XL` tier: the longest templates run ~20 probe steps of several queries each against the repo and live systems. A missing surface is an explicit skip, not a clean finding. The final message is a fixed block shown verbatim, so the caller never paraphrases it: a machine-readable header (`Status` `COMPLETE` or `STOPPED`, report path, per-severity counts) above `<!-- verdict:end -->`, then the `Suggested Issues` table copied verbatim — rewording a row loses the grouping a person files from — plus `Not Checked`, ledger rows, `Process Improvements` and embedded directives. Every early STOP still returns that header with a `Stopped` reason. `Recommended Next Step` names which rows to file as issues (every `MEDIUM`+ row; the batched `LOW` row optional) and the highest-severity row's next step; each body follows `.bkit/trackers.md` § Issue shapes and carries an `**Audit reports:**` line with the row's finding IDs, so `/bkit-plan` reads the report; filed as sub-issues of the template's audit parent when a tracker is wired; shown whether or not a tracker is wired. The final message ends with the fixed related-utilities line from the `docs/roadmap.md` header, word for word. |
369
+ | `bkit-audit-generate` | Read one `docs/templates/bkit-audit-*.md` and run its probe steps against the repo. Write the filled template to `docs/audits/{YYYYMM}/{slug}.md` (`{slug}` = template basename without `bkit-audit-`). **If** that month already has this slug → `{slug}-{TIMESTAMP}.md` (`{TIMESTAMP}` = `date -u +%Y%m%dT%H%M%SZ`). Report = template up to `bkit:end`. The template supplies shape and its own probe steps; the rules every template shares (read-only, evidence, Pre-Flight, credentials, permissions, previous run, gates, severity) are this prompt's Contract bullets. Right after creating the stub it runs `bkit branches anchor-stamp --base-sha=<HEAD> --base-ref=<branch>`, so the audited commit and the bkit release the report came from sit in the same `bkit:anchor` block design and plan docs carry (`.bkit/workflows.md` § Anchor Block) — top of file, machine-readable, the next run reads it as the previous report's commit; an outdated release also gets a visible `Outdated bkit` line under the block, written by the stamp. The agent runs the stamp itself because audits have no orchestrator; the block is still tool-written, never hand-written. Writes only the report, `docs/failures.jsonl` rows (`by: {workflow: "audit", step: "audit-{slug}"}`; one per `MEDIUM`+ finding, reusing any open row's `key` for the same problem whichever workflow wrote it, closing only open `audit-{slug}-` rows a rerun no longer finds; skipped by templates that audit the ledgers themselves, `.bkit/jsonl.md`), and the template's consumer-owned `Project Instructions` (below `bkit:end`), which it reads first and refines each run — exact commands, paths, tool gaps, corrections — so the next run starts from what this one learned; edits listed under the report's `Process Improvements` → `Project`, universal step changes proposed under `Universal`. Never results there (reports hold the series) and never text from external input (an injected directive would persist into every later run). The report's `Suggested Issues` groups findings into tracker-sized changes — one row per change landed in one go, every `MEDIUM`+ finding in exactly one row — and names each row's next step: `/bkit-design\|/bkit-plan\|/bkit-build\|/bkit-debug` by the § When to Run rules of `.bkit/workflows.md` (restated inline, prompts are self-contained), or `manual` for a change outside the repo. The agent proposes and never files: its tracker tools post a comment and run one read-only search, so a person files from the table and each full workflow's related-issues scan catches an issue already open. Issue named in the input (same ID shapes as workflow Step 1, input only, never files the run read) → one `###### bkit:generated` comment there, built by `bkit issues comment --kind=audit` from the final summary (`Embedded Directives` as sources only, so injected text never reaches the tracker), posted through the tracker MCP after any `STOPPED` run too; retry once, then the header `Issue` line reads `failed`; never gates the run. No issue named, tracker wired → the agent searches for the template's audit parent (`.bkit/trackers.md` § Issue shapes — one standing issue per template that every run reports to, so run summaries and the issues they spawn collect in one place instead of one issue per run): an open issue whose body names `docs/templates/bkit-audit-{slug}.md` → `Parent: {ID} found` and the summary is posted there as if it had been named. None → `Parent: missing`, `Issue: None`, the built comment is saved to `{outboxRoot}/audit-{slug}/parent-comment.md`, and `Recommended Next Step` opens with a ready-to-file parent (title, labels, body with the run line) and the instruction to post that saved comment as its first comment — the agent cannot create issues, so whoever files the parent posts it. Search error → `Parent: unchecked`, same saved comment, posted on the parent once found or created. Every filed `Suggested Issues` row is a sub-issue of the parent, existing or new. Tracker `none` → no comment, no search, `Issue: None`, `Parent: None`. Bash is limited to that stamp, the setup tips, the issue comment, and the probes the template or its `Project Instructions` name; throwaway worktrees / remote targets removed before the report finishes. Web tools serve each template's Standards step. Pre-Flight `Permissions` probes each tool once before `Probe Steps`; a blocked call (there or in any step) never stops the run: the agent records it with its narrow allow rule, runs every step the remaining tools allow, and lists the work it could not do under `Not Checked`. Blocked calls persist in `Project Instructions` `Permissions`; the next run probes them first. `XL` tier: the longest templates run ~20 probe steps of several queries each against the repo and live systems. A missing surface is an explicit skip, not a clean finding. The final message is a fixed block shown verbatim, so the caller never paraphrases it: a machine-readable header (`Status` `COMPLETE` or `STOPPED`, report path, per-severity counts) above `<!-- verdict:end -->`, then the `Suggested Issues` table copied verbatim — rewording a row loses the grouping a person files from — plus `Not Checked`, ledger rows, `Process Improvements` and embedded directives. Every early STOP still returns that header with a `Stopped` reason. `Recommended Next Step` names which rows to file as issues (every `MEDIUM`+ row; the batched `LOW` row optional) and the highest-severity row's next step; each body follows `.bkit/trackers.md` § Issue shapes and carries an `**Audit reports:**` line with the row's finding IDs, so `/bkit-plan` reads the report; filed as sub-issues of the template's audit parent when a tracker is wired; shown whether or not a tracker is wired. The final message ends with the fixed related-utilities line from the `docs/roadmap.md` header, word for word. |
315
370
 
316
371
  Every template under `docs/templates/bkit-audit-*.md`:
317
372
 
@@ -359,77 +414,129 @@ Every template under `docs/templates/bkit-audit-*.md`:
359
414
  <!-- prettier-ignore -->
360
415
  | Agent | Scope |
361
416
  | --- | --- |
362
- | `bkit-utility-update-deferrals` | Fold `docs/deferrals.jsonl`, check every winning row not `CLOSED` against the codebase, append `CLOSED` rows with evidence. An entry an open tracker issue already covers gets a same-`key` `TRIAGED` row carrying that issue id in `references` — disposition known, not fixed (`.bkit/jsonl.md` § Updating a Row), so it still counts as open work. Only codebase evidence closes a row: the issue is a second home, never a second source of truth, and the tracker is optional per repo. Never creates a tracker issue. An unambiguous stale locator is folded into the same append; `status` unchanged. One append per `key` per run. Unresolved locators are reported, not guessed. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
363
- | `bkit-utility-update-failures` | Same over `docs/failures.jsonl`, judged on whether the bug still reproduces from the code. A flaky row closes only on a landed determinism fix, never a green run. Same `TRIAGED` promotion on an open tracker issue; a `failure` carries no `reason` field, so its `TRIAGED` rows only ever mean issue-promotion, unlike `deferral` where `bkit-ship-deferred-sweep` also writes `TRIAGED` alongside `reason` = `unjustified`. Same locator append. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
364
- | `bkit-utility-update-decisions` | Fold `docs/decisions.jsonl` by `key`, highest `ts` wins → read only winning rows where `status` is `ACTIVE` or absent. Classify `HOLDS` / `CONTRADICTED` / `OBSOLETE` / `AMBIGUOUS` against `docs/` and the codebase. `CONTRADICTED` with one unambiguous replacement appends the corrected decision `ACTIVE`; `OBSOLETE` (nothing replaced it) appends `SUPERSEDED`. An unambiguous stale locator is folded into the same append and does not reclassify. One append per `key` per run. `AMBIGUOUS` and unresolved locators are left. Doc sentences that restate a corrected row are edited to match; implementation code is not. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
365
- | `bkit-utility-update-knowledge` | Fold `docs/knowledge.jsonl` by `key`, highest `ts` wins → read only winning rows where `status` is `ACTIVE` or absent. Classify `HOLDS` / `CONTRADICTED` / `OBSOLETE` / `AMBIGUOUS`. An entry about upstream behaviour is never retracted from absence of local evidence. `OBSOLETE` appends same-`key` `SUPERSEDED`; `CONTRADICTED` appends the corrected lesson `ACTIVE`. Same locator append. Same doc-sentence edit; implementation code is not. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
366
- | `bkit-utility-update-roadmap` | Rank next best steps from the board, both `## Notes` blocks, `.bkit/issues.jsonl` and open deferrals / failures. When a tracker is wired, always names the next milestone and the next distinct umbrella; a child is included only when open, unblocked, and able to start now. Those containers stay visible and enter the list only by priority, one row each, action = those children. List length is `package.json` `bkit.prompts.roadmapNext` (positive integer; absent or invalid → 25). Due audits compete by priority (risk covered first, then how overdue) for at most 3 rows. The length hint sits in the generated `roadmap-status` header. The full workflow is the default recommendation; names the `.bkit/lite.md` lite workflow (`<phase>:lite`) only where it is clearly enough. An item with no tracker issue (ledger row, due audit) gets a file-an-issue-first recommendation, shaped per `.bkit/trackers.md` § Issue shapes, when a tracker is wired. Rewrites ONLY the `roadmap-analysis` block; bugs and deferrals it finds go under `Found Bugs` in its output for the caller to record. |
417
+ | `bkit-utility-update-deferrals` | Fold `docs/deferrals.jsonl`, check every winning row not `CLOSED` against the codebase, append `CLOSED` rows with evidence. An entry an open tracker issue already covers gets a same-`key` `TRIAGED` row carrying that issue id in `references` — disposition known, not fixed (`.bkit/jsonl.md` § Updating a Row), so it still counts as open work. Only codebase evidence closes a row: the issue is a second home, never a second source of truth, and the tracker is optional per repo. Never creates a tracker issue. An unambiguous stale locator is folded into the same append; `status` unchanged. One append per `key` per run. Unresolved locators are reported, not guessed. Each appended `key` is looked up in the other three ledgers: a live row pointing at it (`references`, a `hard-blocked` blocker named in `description`, or the replaced locator) is reported under `Cross-Ledger Impact` as `blocker-cleared` / `evidence-superseded` / `shared-locator` for the caller to hand to that ledger's updater. Report only — `KIND_WRITERS` refuses a cross-ledger append, and a same-`key` correction leaves the pointer valid. Ends with a best-effort `bkit ledgers compact --ledger=deferrals`: superseded rows dropped, each winner stamped `firstTs` / `updates` (`.bkit/jsonl.md` § Compaction); a non-zero exit is ignored, and lines are never removed by hand. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
418
+ | `bkit-utility-update-failures` | Same over `docs/failures.jsonl`, judged on whether the bug still reproduces from the code. A flaky row closes only on a landed determinism fix, never a green run. Same `TRIAGED` promotion on an open tracker issue; a `failure` carries no `reason` field, so its `TRIAGED` rows only ever mean issue-promotion, unlike `deferral` where `bkit-ship-deferred-sweep` also writes `TRIAGED` alongside `reason` = `unjustified`. Same locator append. Same cross-ledger impact report. Same closing compact, on `failures`; a flaky row's flake count is its winner's `updates`, which survives the compaction. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
419
+ | `bkit-utility-update-decisions` | Fold `docs/decisions.jsonl` by `key`, highest `ts` wins → read only winning rows where `status` is `ACTIVE` or absent. Classify `HOLDS` / `CONTRADICTED` / `OBSOLETE` / `AMBIGUOUS` against `docs/` and the codebase. `CONTRADICTED` with one unambiguous replacement appends the corrected decision `ACTIVE`; `OBSOLETE` (nothing replaced it) appends `SUPERSEDED`. An unambiguous stale locator is folded into the same append and does not reclassify. One append per `key` per run. `AMBIGUOUS` and unresolved locators are left. Doc sentences that restate a corrected row are edited to match; implementation code is not. Same cross-ledger impact report and closing compact (on `decisions`) as `bkit-utility-update-deferrals`. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
420
+ | `bkit-utility-update-knowledge` | Fold `docs/knowledge.jsonl` by `key`, highest `ts` wins → read only winning rows where `status` is `ACTIVE` or absent. Classify `HOLDS` / `CONTRADICTED` / `OBSOLETE` / `AMBIGUOUS`. An entry about upstream behaviour is never retracted from absence of local evidence. `OBSOLETE` appends same-`key` `SUPERSEDED`; `CONTRADICTED` appends the corrected lesson `ACTIVE`. Same locator append. Same doc-sentence edit; implementation code is not. Same cross-ledger impact report. Same closing compact, on `knowledge`. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. Final message is the output format verbatim. |
421
+ | `bkit-utility-update-roadmap` | Rank next best steps from the board, both `Notes` blocks, `.bkit/issues.jsonl` and open deferrals / failures. When a tracker is wired, always names the next milestone and the next distinct umbrella; a child is included only when open, unblocked, and able to start now. Those containers stay visible and enter the list only by priority, one row each, action = those children. List length is `package.json` `bkit.prompts.roadmapNext` (positive integer; absent or invalid → 25). Due audits compete by priority (risk covered first, then how overdue) for at most 3 rows. The length hint sits in the generated `roadmap-status` header. The full workflow is the default recommendation. An item with no tracker issue (ledger row, due audit) gets a file-an-issue-first recommendation, shaped per `.bkit/trackers.md` § Issue shapes, when a tracker is wired. Rewrites only the `roadmap-analysis` block; bugs and deferrals it finds go under `Found Bugs` in its output for the caller to record. Appends no ledger rows; its one ledger write is a closing best-effort `bkit ledgers compact --ledger=issues` (`.bkit/jsonl.md` § Compaction), which removes superseded `feature` rows and adds none — `.bkit/issues.jsonl` has no updater of its own, and this is the utility that already folds it. |
367
422
  | `bkit-utility-update-tracker` | Bring every open issue (or the issues named in input) in line with `.agents/rules.md` tracker rules: title <=70 characters, type label plus exactly one `next:*` when no workflow label, board status from the workflow label, priority and size, milestone, parent, blocked-by. Sources in order: newest issues audit report `Proposal` / `Triage` / `Links`, then `docs/templates/bkit-audit-issues.md` § `Ranking Rules` and § `Size Scale`, then `docs/roadmap.md` `roadmap-analysis`. A report value overwrites a set field; its own judgement only fills empty ones and lists the rest under `Proposed`, so a human's manual setting survives unless an audit disagreed. Re-reads each issue right before writing and writes the full label set from that read, so concurrent workflow label changes survive; GitHub blocked-by goes in the body `###### bkit:blocked-by` block build reads, every other body byte unchanged. Never creates, closes, reopens or comments; never touches `bkit:*` labels or a done state; never removes a link; new milestones, parents and roadmap items with no issue are suggestions only. `Follow-ups` names the issues audit when due and the roadmap when priorities or links changed. No file writes. Bugs it finds go under `Found Bugs` in its output for the caller to record. |
368
423
  | `bkit-utility-compress-markdown` | Token-optimize a single markdown file; preserve 100 % load-bearing tokens. **Destructive** (overwrites target). |
369
424
  | `bkit-utility-update-documentation` | Sync every doc in `.agents/docs.md` with system state — drift cleanup across multiple plans. Ensure agent memories (`MEMORY.md` / `memory/*.md`) do not duplicate content and only contain pointers (path, section, one-line hook) to relevant `docs/*.md` sections / `docs/*.jsonl` entries. **If** memory contains duplicate / stale content → update memory. |
370
425
 
371
426
  ## Hook-injected text
372
427
 
373
- Text hooks add to an agent's context mid-run — every agent and subagent, no frontmatter opt-out. Canonical wording lives in the engine file; per-host delivery → `.bkit/providers.md` § Guardrails, § GateGuard, § ReadGuard.
428
+ Hooks add text to an agent's context mid-run. The text reaches every agent and subagent, and no frontmatter opts out of it. The canonical wording lives in the engine file. Per-host delivery → `.bkit/providers.md` § Guardrails, § GateGuard, § ReadGuard.
374
429
 
375
430
  <!-- prettier-ignore -->
376
431
  | Hook | Fires on | Injects | Hosts | Canonical text |
377
432
  | --- | --- | --- | --- | --- |
378
433
  | GateGuard | First edit per (session, agent, file) | First touch → investigation checklist (file read, callers, symbols, contracts, conventions, instructions quoted); new file → reuse-over-duplicate checklist; gate-config path → fix-code-not-gate advisory; `docs/**.md` (not `docs/plans/`) → density advisory appended | All five | `.bkit/core/hooks/gateguard-engine.js` |
379
434
  | ReadGuard | Repeat read of an unchanged line range, same agent | Use-earlier-output note; re-read only if the output left context | All five | `.bkit/core/hooks/readguard-engine.js` |
380
- | Guardrails | Shell edit (`sed -i`, redirect into a source file); unlisted `curl` / `wget` host on Claude, Codex; Codex ask tier | Advisory naming the bypassed gate; command still runs | Claude, Cursor, Codex, Antigravity | `.bkit/core/hooks/guardrails-shell.js`, `.bkit/core/hooks/guardrails-tiers.js` |
435
+ | Guardrails | Shell edit (`sed -i`, redirect into a source file); unlisted `curl` / `wget` host on Claude, Codex; Codex ask tier | Advisory naming the bypassed gate; command still runs | Claude, Codex, Antigravity, Cursor | `.bkit/core/hooks/guardrails-shell.js`, `.bkit/core/hooks/guardrails-tiers.js` |
381
436
 
382
- - **Never a must-do:** hook text is advisory and fail-open; it never fires on an untrusted Codex `.codex/` layer and arrives after the edit on opencode. A prompt instruction that must hold stays in the prompt.
437
+ - **Never a must-do:** hook text is advisory and fails open. It never fires on an untrusted Codex `.codex/` layer, and it arrives after the edit on OpenCode. A prompt instruction that must hold stays in the prompt.
383
438
 
384
439
  ## Special Behaviors (frontmatter-adjacent contracts)
385
440
 
386
441
  <!-- prettier-ignore -->
387
442
  | Behavior | Agents | Notes |
388
443
  | --- | --- | --- |
389
- | **Parallel-fan-out circular-reasoning guard** | `bkit-design-brainstorm-ideate-*` (triplet), `bkit-plan-review-*` (pair) + `bkit-plan-scope-check`, `bkit-build-challenge-*` (pair) | Each agent's prompt explicitly forbids reading any sibling output. The triplet carries per-slot lenses (pragmatic / optimal-tradeoff / risk-premortem for ideate); pairs carry per-slot lenses (plan text / code for plan-review, beside the scope check; bug-hunting / security for challenge). Effort-tier and model diversity provide reasoning-depth + cross-family spread. |
390
- | **Loop context accumulation** | `bkit-build-plan-verify`/`bkit-build-plan-gaps-remediate`, `bkit-debug-failure-diagnose` | Per-attempt log files (`*-{N}-*.md`) — agent globs its own prior logs and adjusts strategy (Standard / Pivot / Final). Each loop stops at its ask point's limit (`bkit.workflows.<workflow>.<point>`) → Failure Path; with `ask` on, the orchestrator asks before each further round (`.bkit/workflows.md` § Ask points). `bkit-debug-failure-remediate` is strategy-blind — applies whatever the strategy-aware diagnosis generates. See `workflows.md` § Loop Context Accumulation. |
391
- | **Tracker-bound output (`Output cap: 500 token prose`)** | `bkit-debug-abort-handover` | Output is pasted verbatim into the tracker comment thread (workflow `comment-post-step` block). Cap applies to prose only — sections that list items verbatim are exempt, since a reworded acceptance item or option loses the trade-off the human decides on. Self-contained — don't add references to other files the human reviewer can't open. **Cap belongs only here** — critique / review / diagnosis agents (`bkit-*-review-*`, `bkit-*-scope-check-*`, `bkit-build-plan-verify`, `bkit-build-challenge-*`, `bkit-debug-failure-diagnose`, `bkit-ship-deferred-sweep`) are uncapped because their output drives downstream apply steps (`*-update`, `*-remediate`) and detail compression there pushes interpretation cost onto the apply agent. |
392
- | **Stubbed output (verdict + section placeholders)** | All multi-section verdict-bearing agents — critique / review (`bkit-design-review-*`, `bkit-design-spike`, `bkit-plan-review-*`, `bkit-plan-scope-check`, `bkit-build-plan-current`, `bkit-build-plan-verify`, `bkit-build-challenge-1`, `bkit-build-challenge-2`, `bkit-debug-failure-diagnose`, `bkit-ship-deferred-sweep`), top-of-pipeline synthesizers (`bkit-design-generate`, `bkit-plan-generate`, `bkit-audit-generate`), the tracker-bound handover (`bkit-debug-abort-handover`), and source modifiers (`bkit-build-plan-execute`, `bkit-build-plan-gaps-remediate`, `bkit-debug-failure-remediate`, `bkit-ship-conflict-resolve`) — see "Flat append-only log for source modifiers" row below. | Process shape: (1) Initialise **Create** output stub with PENDING placeholders (matches Output format); (2-3) re-read inputs + load docs; (4) Update **Create** output — replace each PENDING via `Edit` as content becomes available (standing directive for the analysis below); (5+) analysis steps fill PENDING incrementally as findings emerge. Output format shows each section ending with `; initially set to PENDING` annotation. Truncation recovery is **section-granular** — orchestrator greps for PENDING markers and re-issues specific gaps. Per-step gate `No PENDING placeholders remain` paired with `Verdict valid` enforces completeness. Plan-update / design-update / design-organise follow the same shape but bootstrap from the prior v(N) with `bkit steps stub --from= --to=` (Bash; `--strip-header` drops the prior header) instead of fresh stub (Write), so steps 2-4 above are steps 1-3 there. |
393
- | **Untrusted-input prompt-defense** | every `.bkit/core/prompts/*.md` and every `.bkit/core/workflows/*.md` — no exceptions, no selection step | A single dense bullet in the `**Rules:**` block, identical in every file — the `external-input` clause in `.bkit/instructions.md`, a copy of the `.agents/rules.md` rule. `.bkit/core/engine/tools/instructions.test.ts` enumerates both directories at run time and fails on a file missing the clause, carrying a near-copy, or retaining the retired pre-universal variant, and asserts the clause keeps all three legs (data-not-instructions / report-directives / token-scope), so a newly added prompt is covered-or-failing by construction and no list is maintained anywhere. **Coverage is universal by decision, reversing an earlier scoped rule** that carried the bullet on 10 ingest-and-reason agents only and rejected blanket coverage on per-run token cost — classification is itself the defect (the judgement is re-made on every new prompt, silently, by whoever adds it), and per-file edit cost is unavoidable either way. Threat model, the write-time fence and the spawn-line notice: `docs/security.md` § Untrusted Content & Prompt Injection. |
394
- | **Central agent rules (`**Rules file:**` bullet, `**Rules:**` block)** | every `.bkit/core/prompts/*.md` — no exceptions | `.agents/rules.md` is a 2x2: `bkit Rules` and `Project Rules` each split into `Universal Rules` (binds every agent) and `Orchestrator Rules` (workflows / commands only — git, CI, merge, tracker, and human-gate rules a leaf must not act on). **bkit rules are copied:** the `**Rules:**` block after `**Contract:**` holds, word for word and in `rules.md` order, every bkit `Universal Rules` bullet whose `- carry:` line in `.bkit/instructions.md` names the prompt (the agent acts on it: its carriers plus its tool set — `Bash` → `Destructive git PROHIBITED`, `Other parallel agents`; `Edit` / `Write` → `File edits`, `Edit tools ONLY`). A rule never appears in `**Protocol:**` or `**Contract:**` as well. A leaf sees only the bkit rules its carry lines name. **Project rules are read:** a prompt with `Grep` or `Bash` says ``Find the line of `<!-- bkit:end -->` in `.agents/rules.md` (`grep -n`), then read the file from that line on → follow its `Universal Rules`; skip its `Orchestrator Rules`.``, so its agent never loads the bkit block it already carries; a prompt with neither says ``Read `.agents/rules.md` below `<!-- bkit:end -->` → …``. Either way a consumer's universal rules and `## Project Settings` reach leaves while their orchestrator-only rules do not — the split is the consumer's declaration of which is which. Subsection names are shared across the two groups (`Agent Management`, `Git, CI & Merges`); a subsection is omitted only where it has no entry. **Position is contract:** `**Rules file:**` is always the first bullet of the `**Protocol:**` block, in every prompt and every workflow. Workflows and the repo-only `utility-*` agents (they push and merge) take the unscoped `**Rules file:**` line, and the entry files (`CLAUDE.md`, `AGENTS.md`, `.cursor/rules/bkit-rules.mdc`, un-quoted `- **Rules:**`) the same text under `**Rules:**`: ``Read `.agents/rules.md` → follow all rules.``. `.bkit/core/engine/tools/instructions.test.ts` asserts the three header blocks, the `**Rules file:**` anchor, the block and order of every placed header clause, `rules-leaf` forbidden on workflows and `utility-*` agents, both groups under `bkit Rules`, and a universal counterpart for every orchestrator subsection; `.bkit/core/engine/tools/instructions-coverage.test.ts` asserts every prompt holds each rule its carry line names, no prompt holds a bkit rule no carry line names, and every prompt carries at least one. **Assertions stop at `bkit:end`:** `Project Rules` is consumer-owned; `bkit install` never writes it and `bkit check` ignores it. Header clauses (`**Order:**`, `**No improv:**`, `**No commits:**` (prompts with `Bash`)) stay inline in every prompt they apply to. `**Auto-apply:**` is `Orchestrator Rules` only and inlined in no prompt. |
395
- | **`.agents/docs.md` read exemption** | `bkit-utility-compress-markdown` | Single-file mechanical utility — compress prose in one supplied target. No domain reasoning, so the doc catalog earns nothing against its token cost. Every other prompt loads it. |
444
+ | **Parallel-fan-out circular-reasoning guard** | `bkit-design-brainstorm-ideate-*` (triplet), `bkit-plan-review-*` (pair) + `bkit-plan-scope-check`, `bkit-build-challenge-*` (pair) | Each agent's prompt explicitly forbids reading any sibling output. The triplet carries per-slot lenses (pragmatic / optimal trade-off / risk-premortem for ideate); pairs carry per-slot lenses (plan text / code for plan-review, beside the scope check; bug-hunting / security for challenge). Effort-tier and model diversity provide reasoning-depth + cross-family spread. |
445
+ | **Loop context accumulation** | `bkit-build-plan-verify`/`bkit-build-plan-gaps-remediate`, `bkit-debug-failure-diagnose` | Per-attempt log files (`*-{N}-*.md`): the agent globs its own prior logs and adjusts its strategy (Standard / Pivot / Final). Each loop stops at its ask point's limit (`bkit.workflows.<workflow>.<point>`) → Failure Path. With `ask` on, the orchestrator asks before each further round (`.bkit/workflows.md` § Ask points). `bkit-debug-failure-remediate` is strategy-blind: it applies whatever the strategy-aware diagnosis generates. See `.bkit/workflows.md` § Loop context accumulation. |
446
+ | **Tracker-bound output (`Output cap: 500 token prose`)** | `bkit-debug-abort-handover` | Output is pasted verbatim into the tracker comment thread (workflow `comment-post-step` block). The cap applies to prose only. Sections that list items verbatim are exempt, because a reworded acceptance item or option loses the trade-off the human decides on. The output is self-contained: don't add references to other files the human reviewer can't open. **Cap belongs only here:** critique / review / diagnosis agents (`bkit-*-review-*`, `bkit-*-scope-check-*`, `bkit-build-plan-verify`, `bkit-build-challenge-*`, `bkit-debug-failure-diagnose`, `bkit-ship-deferred-sweep`) are uncapped because their output drives downstream apply steps (`*-update`, `*-remediate`) and detail compression there pushes interpretation cost onto the apply agent. |
447
+ | **Stubbed output (verdict + section placeholders)** | All multi-section verdict-bearing agents — critique / review (`bkit-design-review-*`, `bkit-design-spike`, `bkit-plan-review-*`, `bkit-plan-scope-check`, `bkit-build-plan-current`, `bkit-build-plan-verify`, `bkit-build-challenge-1`, `bkit-build-challenge-2`, `bkit-debug-failure-diagnose`, `bkit-ship-deferred-sweep`), top-of-pipeline synthesizers (`bkit-design-generate`, `bkit-plan-generate`, `bkit-audit-generate`), the tracker-bound handover (`bkit-debug-abort-handover`), and source modifiers (`bkit-build-plan-tasks`, `bkit-build-plan-gaps-remediate`, `bkit-debug-failure-remediate`, `bkit-ship-conflict-resolve`) — see "Flat append-only log for source modifiers" row below. | Process shape: (1) Write the **Create** output stub, with `PENDING` in every placeholder (matches Output format); (2-3) re-read inputs + load docs; (4) Update **Create** output — replace each PENDING via `Edit` as content becomes available (standing directive for the analysis below); (5+) analysis steps fill PENDING incrementally as findings emerge. Output format ends each section's bracketed placeholder with "; initially set to `PENDING`." Truncation recovery is section-granular: the orchestrator greps for PENDING markers and re-issues the specific gaps. Per-step gate `No PENDING placeholders remain` paired with `Verdict valid` enforces completeness. Plan-update / design-update / design-organise follow the same shape but bootstrap from the prior v(N) with `bkit steps stub --from= --to=` (Bash; `--strip-header` drops the prior header) instead of fresh stub (Write), so steps 2-4 above are steps 1-3 there. |
448
+ | **Untrusted-input prompt-defense** | every `.bkit/core/prompts/*.md` and every `.bkit/core/workflows/*.md` — no exceptions, no selection step | Two bullets in the `**Rules:**` block, identical in every file: the `external-input` and `external-skills` clauses in `.bkit/instructions.md`, copied from `.agents/rules.md`. `.bkit/core/engine/tools/instructions.test.ts` enumerates both directories and rejects missing clauses, near-copies and retired wording. It checks the data-not-instructions, report-directives and token-scope requirements, plus the skill-authority boundary. Every new prompt must carry both rules. Universal coverage avoids a per-file decision that could leave an agent unprotected. Threat model, write-time fence and spawn-line notice: `docs/security.md` § Untrusted Content & Prompt Injection. |
449
+ | **Central agent rules (`**Rules file:**` bullet, `**Rules:**` block)** | every `.bkit/core/prompts/*.md` — no exceptions | `.agents/rules.md` is a 2x2: `bkit Rules` and `Project Rules` each split into `Universal Rules` (binds every agent) and `Orchestrator Rules` (workflows / commands only — git, CI, merge, tracker, and human-gate rules a leaf must not act on). **bkit rules are copied:** the `**Rules:**` block after `**Contract:**` holds, word for word and in `rules.md` order, every bkit `Universal Rules` bullet whose `- carry:` line in `.bkit/instructions.md` names the prompt (the agent acts on it: its carriers plus its tool set — `Bash` → `Destructive git`, `Other parallel agents`; `Edit` / `Write` → `File edits`, `Edit tools`). A rule never appears in `**Protocol:**` or `**Contract:**` as well. A leaf sees only the bkit rules its carry lines name. **Project rules are read:** a prompt with `Grep` or `Bash` says ``Find the line of `<!-- bkit:end -->` in `.agents/rules.md` (`grep -n`), then read the file from that line on → follow its `Universal Rules`; skip its `Orchestrator Rules`.``, so its agent never loads the bkit block it already carries; a prompt with neither says ``Read `.agents/rules.md` below `<!-- bkit:end -->` → …``. Either way a consumer's universal rules and `Project Settings` reach leaves while their orchestrator-only rules do not. The split is the consumer's declaration of which is which. Subsection names are shared across the two groups (`Agent Management`, `Git, CI & Merges`); a subsection is omitted only where it has no entry. **Position is contract:** `**Rules file:**` is always the first bullet of the `**Protocol:**` block, in every prompt and every workflow. Workflows and the repo-only `utility-*` agents (they push and merge) take the unscoped `**Rules file:**` line, and the entry files (`CLAUDE.md`, `AGENTS.md`, `.cursor/rules/bkit-rules.mdc`, un-quoted `- **Rules:**`) the same text under `**Rules:**`: ``Read `.agents/rules.md` → follow all rules.``. `.bkit/core/engine/tools/instructions.test.ts` asserts the three header blocks, the `**Rules file:**` anchor, the block and order of every placed header clause, `rules-leaf` forbidden on workflows and `utility-*` agents, both groups under `bkit Rules`, and a universal counterpart for every orchestrator subsection; `.bkit/core/engine/tools/instructions-coverage.test.ts` asserts every prompt holds each rule its carry line names, no prompt holds a bkit rule no carry line names, and every prompt carries at least one. **Assertions stop at `bkit:end`:** `Project Rules` is consumer-owned; `bkit install` never writes it and `bkit check` ignores it. Header clauses (`**Order:**`, `**No improv:**`, `**No commits:**` (prompts with `Bash`)) stay inline in every prompt they apply to. `**Auto-apply:**` is `Orchestrator Rules` only and inlined in no prompt. |
450
+ | **`.agents/docs.md` read exemption** | `bkit-utility-compress-markdown` | A single-file mechanical utility that compresses prose in one supplied target. It does no domain reasoning, so the doc catalog earns nothing against its token cost. Every other prompt loads it. |
396
451
  | **`acceptEdits` hard requirement** | All in-pipeline agents | `Bash(*)`/`Edit(*)`/`Write(*)` blanket-allow lives in `.claude/settings.json`; `acceptEdits` skips per-call prompts. Stricter modes break end-to-end runs. |
397
- | **Plan task ticking owned by verify** | `bkit-build-plan-execute` (ticks only applicable Pre-Flight and gate boxes), `bkit-build-plan-verify` (ticks task boxes `[ ]`→`[x]` only on verified codebase artifact) | Task checkboxes are verify's derived view of the codebase. The executor's durable progress record is its flat append-only log of `DONE —` lines, which keeps resume safe and keeps per-task `Edit` ticks out of its turn budget. Execute ticks each applicable Pre-Flight and gate box as that gate passes and continues without asking; it never ticks a task box or a human-confirmation box. |
398
- | **Flat append-only log for source modifiers** | `bkit-build-plan-execute`, `bkit-build-plan-gaps-remediate`, `bkit-debug-failure-remediate` | Output format collapses to a single H1 + flat log body (no separate `Brief` / `Strategy` body sections — `Strategy` lives in machine-parseable header alongside `Verdict`; reasoning is implicit in `DONE —` / `BLOCKED —` lines). Stub state: `Verdict: PENDING` header + H1 + empty body. After every discrete unit of work (task / gap / finding / fix step): `Bash echo "DONE — {Unit} {N}: {name} — {outcome} ({path or commit ref})" >> {artifact}`. On failure: `BLOCKED — {Unit} {N}: {name} — {reason}`. Step 4 standing directive: `Update **Create** output — append DONE/BLOCKED log entries at EOF as content becomes available.` (vs canonical `replace each PENDING via Edit` for multi-section verdict-bearing agents). `Verdict` flips PENDING → PASS / CONCERNS / FAIL at the end via `Edit` on the header line. **Two wins:** (1) timeout resilience — partial progress is on disk after every appended line, so a budget-exhausted agent leaves a usable record (vs the prior antipattern: ~30 ticks clustered at end → all lost on timeout); (2) `Bash echo >>` is cheaper than `Edit` (no file-read for `old_string` match). **Skip-done logic** (read prior attempts via Read glob, skip units already in `DONE —`) added only where workflow already attempt-numbers and globs the artifact: `bkit-build-plan-gaps-remediate` (via `VERIFY_ATTEMPT`, glob already in its Read) and `bkit-build-plan-execute` (one `build-plan-execute-{GROUP}-log.md` per group; every attempt — a turn-cap re-run or a Decision Pause `Fix now` re-run — appends to it, and `bkit context brief --next` leaves out tasks it logs `DONE — TASK {N}:`). On PENDING (subagent timeout) the verify→remediate loop bounded by `build.verify.maxAttempts` absorbs recovery via `bkit-build-plan-gaps-remediate`. `bkit-debug-failure-remediate` is diagnosis-driven (each attempt fresh-strategy via `bkit-debug-failure-diagnose`) — skip-done would be wrong shape; gets stub+log only, but flat log still feeds `bkit-debug-failure-diagnose`'s loop-context glob with finer evidence. |
452
+ | **Plan task ticking owned by verify** | `bkit-build-plan-tasks` (ticks only applicable Pre-Flight and gate boxes), `bkit-build-plan-verify` (ticks task boxes `[ ]`→`[x]` only on verified codebase artifact) | Task checkboxes are verify's derived view of the codebase. The executor's durable progress record is its flat append-only log of `DONE —` lines, which keeps resume safe and keeps per-task `Edit` ticks out of its turn budget. Execute ticks each applicable Pre-Flight and gate box as that gate passes and continues without asking; it never ticks a task box or a human-confirmation box. |
453
+ | **Flat append-only log for source modifiers** | `bkit-build-plan-tests`, `bkit-build-plan-tasks`, `bkit-build-plan-gaps-remediate`, `bkit-debug-failure-remediate` | Output format collapses to a single H1 + flat log body (no separate `Brief` / `Strategy` body sections: `Strategy` lives in the machine-parseable header alongside `Verdict`, and the reasoning is implicit in the `DONE —` / `BLOCKED —` lines). Stub state: `Verdict: PENDING` header + H1 + empty body. After every discrete unit of work (task / gap / finding / fix step): `Bash echo "DONE — {Unit} {N}: {name} — {outcome} ({path or commit ref})" >> {artifact}`. On failure: `BLOCKED — {Unit} {N}: {name} — {reason}`. Step 4 standing directive: `Update **Create** output — append DONE/BLOCKED log entries at EOF as content becomes available.` (vs canonical `replace each PENDING via Edit` for multi-section verdict-bearing agents). `Verdict` flips PENDING → PASS / CONCERNS / FAIL at the end via `Edit` on the header line. **Two wins:** (1) timeout resilience: partial progress is on disk after every appended line, so a budget-exhausted agent leaves a usable record, where ~30 ticks clustered at the end would all be lost on a timeout; (2) `Bash echo >>` is cheaper than `Edit`, because it needs no file read for an `old_string` match. **Skip-done logic** (read prior attempts via Read glob, skip units already in `DONE —`) is added only where the workflow already attempt-numbers and globs the artifact: `bkit-build-plan-gaps-remediate` (via `VERIFY_ATTEMPT`, glob already in its Read) and `bkit-build-plan-tasks` (one `build-plan-tasks-{GROUP}-log.md` per group; every attempt — a turn-cap re-run or a Decision Pause `Fix now` re-run — appends to it, and `bkit context brief --next` leaves out tasks it logs `DONE — TASK {N}:`). On PENDING (subagent timeout) the verify→remediate loop bounded by `build.verify.maxAttempts` absorbs recovery via `bkit-build-plan-gaps-remediate`. `bkit-debug-failure-remediate` is diagnosis-driven (each attempt takes a fresh strategy via `bkit-debug-failure-diagnose`), so skip-done would be the wrong shape. It gets stub+log only, but the flat log still feeds `bkit-debug-failure-diagnose`'s loop-context glob with finer evidence. |
399
454
 
400
- <!-- bkit:end -->
455
+ ## Model Presets
401
456
 
402
- ## Model Overrides
457
+ Two settings for OpenRouter, the overflow column, which covers the days after the week's OpenAI, Anthropic and Google credits run out. Costs are per **2-day gap**, the binding case against the weekly cap (`.bkit/providers.md` § Budget band), at the measured OpenCode profile with each agent carrying an equal 1/40 share. Judges and key gates: § Models Judges note.
403
458
 
404
- > [!CAUTION]
405
- > Add rows below to override bkit default model settings above.
406
- > Providers: `cursor`, `openrouter`, `ollama`, …
407
- > Example (non-Cursor): `| bkit-build-plan-verify | openrouter | openrouter/deepseek/deepseek-v4-pro |`
408
- > Install merges overrides over defaults (override wins). Re-run `bkit install --cursor` after editing. Claude Code reads frontmatter — not overridable via this table yet.
409
- > A row naming an agent bkit no longer ships is not applied; `bkit install` / `check` / `upgrade` list it as the `model-override-unknown-agent` advisory.
459
+ <!-- prettier-ignore -->
460
+ | Preset | Judges (12) | Builders (27) | 2-day gap | Cap |
461
+ | --- | --- | --- | --- | --- |
462
+ | **Option 1 — default** | `deepseek-v4.1-flash-1017` 40.9 on eight; `gpt-6.1-sol` 68.7 · $0.10 on the four key gates | `deepseek-v4.1-flash-1017` 40.9 · $0.003 | **A$74 (US$52)** | A$70/week |
463
+ | **Option 2 — spend more** | `gpt-6.1-sol` 68.7 on nine; `deepseek-v4.1-flash-1017` on three | `deepseek-v4.1-flash-1017` | **A$137 (US$96)** | A$140/week |
464
+
465
+ **Option 1 spends the cap on the four key gates.** `gpt-6.1-sol` is the only model over the sonnet bar that the cap carries. It costs about A$13 per agent per gap against DeepSeek's A$0.62, so the cap holds about four. They go where a miss passes downstream unchecked, and the other eight judges and all builders run DeepSeek. A$74 sits A$4 over the A$70 cap, which is accepted. Option 2 adds five judges on sol for A$137. `design-review-1`, `plan-review-1` and `build-challenge-1` stay on DeepSeek in both, and all twelve on sol would land at A$175. Sol beats `gemini-4-argon` at the same token price: +7.5 Long Agentic and $116.78 per LiveBench task against $249.15, because Gemini burns more tokens to finish the same work.
466
+
467
+ **Both totals are floors.** These costs are not priced in:
468
+
469
+ - Sol bills cache writes at $2.50/Mtok against $2 for plain input (A$78 and A$146 if every fresh input token is written).
470
+ - Past a 272k-token prompt, every sol rate doubles, cache reads included.
471
+ - `deepseek-v4.1-flash-1017` doubles on weekdays 01:00-04:00 and 06:00-10:00 UTC.
472
+
473
+ Measure one gap on your repo under Option 2 to replace these estimates.
474
+
475
+ OpenRouter IDs: `gpt-6.1-sol` = `openai/gpt-6.1-sol`, `deepseek-v4.1-flash-1017` = `deepseek/deepseek-v4.1-flash-1017`. **Bold** marks a cell that differs from the default.
476
+
477
+ <!-- prettier-ignore -->
478
+ | Agent | Shipped today | Option 1 — default | Option 2 — spend more |
479
+ | --- | --- | --- | --- |
480
+ | `bkit-design-brainstorm-ideate-1` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
481
+ | `bkit-design-brainstorm-ideate-2` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
482
+ | `bkit-design-brainstorm-ideate-3` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | **gpt-6.1-sol** |
483
+ | `bkit-design-generate` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
484
+ | `bkit-design-review-1` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
485
+ | `bkit-design-review-2` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
486
+ | `bkit-design-spike` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
487
+ | `bkit-design-update` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
488
+ | `bkit-design-organise` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
489
+ | `bkit-plan-generate` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
490
+ | `bkit-plan-review-1` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
491
+ | `bkit-plan-review-2` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | **gpt-6.1-sol** |
492
+ | `bkit-plan-scope-check` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | **gpt-6.1-sol** |
493
+ | `bkit-plan-update` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
494
+ | `bkit-build-plan-convert` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
495
+ | `bkit-build-plan-current` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | **gpt-6.1-sol** |
496
+ | `bkit-build-install-skills` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
497
+ | `bkit-build-plan-tests` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
498
+ | `bkit-build-plan-tasks` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
499
+ | `bkit-build-plan-verify` | gpt-6.1-sol | gpt-6.1-sol | gpt-6.1-sol |
500
+ | `bkit-build-plan-gaps-remediate` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
501
+ | `bkit-build-challenge-1` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
502
+ | `bkit-build-challenge-2` | gpt-6.1-sol | gpt-6.1-sol | gpt-6.1-sol |
503
+ | `bkit-debug-failure-reproduce` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
504
+ | `bkit-debug-failure-diagnose` | gpt-6.1-sol | gpt-6.1-sol | gpt-6.1-sol |
505
+ | `bkit-debug-failure-remediate` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
506
+ | `bkit-debug-failure-soak` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | **gpt-6.1-sol** |
507
+ | `bkit-debug-abort-handover` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
508
+ | `bkit-ship-conflict-resolve` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
509
+ | `bkit-ship-deferred-sweep` | gpt-6.1-sol | gpt-6.1-sol | gpt-6.1-sol |
510
+ | `bkit-ship-update-docs` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
511
+ | `bkit-audit-generate` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
512
+ | `bkit-utility-update-deferrals` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
513
+ | `bkit-utility-update-failures` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
514
+ | `bkit-utility-update-decisions` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
515
+ | `bkit-utility-update-knowledge` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
516
+ | `bkit-utility-update-roadmap` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
517
+ | `bkit-utility-update-tracker` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
518
+ | `bkit-utility-compress-markdown` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
519
+ | `bkit-utility-update-documentation` | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 | deepseek-v4.1-flash-1017 |
520
+
521
+ **Option 2 — spend more.** Five more judges move to `openai/gpt-6.1-sol`, so nine of twelve run 68.7, which is 27.8 above the DeepSeek workhorse. Copy into § Model Overrides:
410
522
 
411
523
  <!-- prettier-ignore -->
412
524
  | Prompt | Provider | Model |
413
525
  | --- | --- | --- |
526
+ | bkit-design-brainstorm-ideate-3 | openrouter | openai/gpt-6.1-sol |
527
+ | bkit-plan-review-2 | openrouter | openai/gpt-6.1-sol |
528
+ | bkit-plan-scope-check | openrouter | openai/gpt-6.1-sol |
529
+ | bkit-build-plan-current | openrouter | openai/gpt-6.1-sol |
530
+ | bkit-debug-failure-soak | openrouter | openai/gpt-6.1-sol |
414
531
 
415
- > [!NOTE]
416
- > **Cursor — API credits / third-party models.** Shipped Cursor defaults are **first-party only** (`cursor-grok-4.5-high`, `composer-2.5`) so workflows keep running when API credits are exhausted. **If** you have third-party / additional API credits and want Claude Opus (or other API-pool models) on Cursor workflows, add rows to the table above — do **not** edit the managed `## Models` table (upgrade overwrites it).
417
- >
418
- > **Mapping (align Cursor to Claude-tier settings):**
532
+ **Cursor — first-party pins only.** `cursor-grok-4.5-high` carries depth, synthesizers and the parallel slots; `composer-2.5` carries volume, the build and ship pipelines. `/v0/models` has no Opus 5.5 ID: `claude-opus-5-thinking-high` is the API-pool equivalent for `claude-opus-5-5` rows and `claude-sonnet-5-thinking-high` for `claude-sonnet-5-5`, both opt-in through § Model Overrides. Orchestrators take the session model, not a table row.
419
533
 
420
- <!-- prettier-ignore -->
421
- | Claude column | Cursor override | Notes |
422
- | --- | --- | --- |
423
- | `claude-opus-5-5` | `claude-opus-5-thinking-high` | `/v0/models` has no Opus 5.5 ID. Depth / synthesizers — paste block below |
424
- | `claude-sonnet-5-5` | `claude-sonnet-5-thinking-high` | `/v0/models` has Sonnet 5, not 5.5. Shipped default stays `composer-2.5` / `cursor-grok-4.5-high` |
425
- | `claude-haiku-5-5` | leave `composer-2.5` | Haiku tier — no change |
534
+ <!-- bkit:end -->
426
535
 
427
- > **Orchestrators** (`/design` `/plan` `/debug`) are session model, not the override table — set the Cursor session / Cloud Agent model to `claude-opus-5-thinking-high` when invoking. `/build` `/ship` stay `composer-2.5` unless you override the session.
428
- >
429
- > **Ready-to-paste — restore Opus on every Claude-opus-tier Cursor agent** (copy rows into the empty table above):
536
+ ## Model Overrides
537
+
538
+ > Add rows below to override bkit's default models, one per agent — `| bkit-plan-review-2 | openrouter | openai/gpt-6.1-sol |`. Override wins; re-run `bkit install --opencode=<provider>` after editing. Preset blocks to copy live in § Model Presets above. Claude pins are prompt frontmatter and are not overridable here.
430
539
 
431
540
  <!-- prettier-ignore -->
432
541
  | Prompt | Provider | Model |
433
542
  | --- | --- | --- |
434
-
435
- > Optional API cross-family (sonnet-tier diversity slots): `| bkit-design-brainstorm-ideate-2 | cursor | claude-sonnet-5-thinking-high |` (+ same for `bkit-design-review-2`). `gpt-5.6-sol-high` is the non-Claude alternative. Probe live `/v0/models` before pinning — catalog churns. See `.bkit/providers.md` § Cursor model catalog.