@wildo-ai/framework-knowledge 1.1.8 → 1.1.9

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 (226) hide show
  1. package/dist/claude-marketplace/plugins/wildo/.claude-plugin/plugin.json +1 -1
  2. package/dist/claude-marketplace/plugins/wildo/references/composites-and-the-loop.md +9 -2
  3. package/dist/claude-marketplace/plugins/wildo/references/creation-state.md +21 -0
  4. package/dist/claude-marketplace/plugins/wildo/references/derived-artifacts.md +121 -0
  5. package/dist/claude-marketplace/plugins/wildo/references/runs-judges-and-questions.md +29 -5
  6. package/dist/claude-marketplace/plugins/wildo/skills/answer/SKILL.md +29 -2
  7. package/dist/claude-marketplace/plugins/wildo/skills/build/SKILL.md +69 -24
  8. package/dist/claude-marketplace/plugins/wildo/skills/doctor/SKILL.md +10 -0
  9. package/dist/claude-marketplace/plugins/wildo/skills/run/SKILL.md +21 -5
  10. package/dist/claude-marketplace/plugins/wildo/skills/start/SKILL.md +31 -5
  11. package/dist/claude-marketplace/plugins/wildo/skills/status/SKILL.md +2 -0
  12. package/dist/claude-marketplace/plugins/wildo/skills/why/SKILL.md +11 -1
  13. package/dist/esm/framework-knowledge.definitions.d.ts +1 -1
  14. package/dist/esm/framework-knowledge.definitions.d.ts.map +1 -1
  15. package/dist/esm/framework-knowledge.definitions.js +11 -1
  16. package/dist/esm/framework-knowledge.definitions.js.map +1 -1
  17. package/dist/esm/playbooks/bottom_up.assemble-verify/playbook.definition.d.ts +2 -2
  18. package/dist/esm/playbooks/bottom_up.assemble-verify/playbook.definition.js +2 -2
  19. package/dist/esm/playbooks/bottom_up.assemble-verify/playbook.definition.js.map +1 -1
  20. package/dist/esm/playbooks/bottom_up.build/playbook.definition.d.ts +11 -11
  21. package/dist/esm/playbooks/bottom_up.build/playbook.definition.d.ts.map +1 -1
  22. package/dist/esm/playbooks/bottom_up.build/playbook.definition.js +29 -22
  23. package/dist/esm/playbooks/bottom_up.build/playbook.definition.js.map +1 -1
  24. package/dist/esm/playbooks/bottom_up.build-decompose/playbook.definition.d.ts +3 -3
  25. package/dist/esm/playbooks/bottom_up.build-decompose/playbook.definition.d.ts.map +1 -1
  26. package/dist/esm/playbooks/bottom_up.build-decompose/playbook.definition.js +59 -10
  27. package/dist/esm/playbooks/bottom_up.build-decompose/playbook.definition.js.map +1 -1
  28. package/dist/esm/playbooks/bottom_up.build-dispatch/playbook.definition.d.ts +1 -1
  29. package/dist/esm/playbooks/bottom_up.build-dispatch/playbook.definition.js +3 -3
  30. package/dist/esm/playbooks/bottom_up.build-dispatch/playbook.definition.js.map +1 -1
  31. package/dist/esm/playbooks/bottom_up.build-scale/playbook.definition.d.ts +1 -1
  32. package/dist/esm/playbooks/bottom_up.build-scale/playbook.definition.js +1 -1
  33. package/dist/esm/playbooks/bottom_up.build-scale/playbook.definition.js.map +1 -1
  34. package/dist/esm/playbooks/bottom_up.conception/playbook.definition.js +1 -1
  35. package/dist/esm/playbooks/bottom_up.conception/playbook.definition.js.map +1 -1
  36. package/dist/esm/playbooks/bottom_up.domain-plan/playbook.definition.d.ts +2 -1
  37. package/dist/esm/playbooks/bottom_up.domain-plan/playbook.definition.d.ts.map +1 -1
  38. package/dist/esm/playbooks/bottom_up.domain-plan/playbook.definition.js +5 -5
  39. package/dist/esm/playbooks/bottom_up.domain-plan/playbook.definition.js.map +1 -1
  40. package/dist/esm/playbooks/bottom_up.foundation/playbook.definition.d.ts +8 -0
  41. package/dist/esm/playbooks/bottom_up.foundation/playbook.definition.d.ts.map +1 -1
  42. package/dist/esm/playbooks/bottom_up.foundation/playbook.definition.js +12 -4
  43. package/dist/esm/playbooks/bottom_up.foundation/playbook.definition.js.map +1 -1
  44. package/dist/esm/playbooks/top_down.brand-identity/playbook.definition.d.ts +25 -27
  45. package/dist/esm/playbooks/top_down.brand-identity/playbook.definition.d.ts.map +1 -1
  46. package/dist/esm/playbooks/top_down.brand-identity/playbook.definition.js +85 -43
  47. package/dist/esm/playbooks/top_down.brand-identity/playbook.definition.js.map +1 -1
  48. package/dist/esm/playbooks/top_down.brand-name-research/playbook.definition.d.ts +40 -0
  49. package/dist/esm/playbooks/top_down.brand-name-research/playbook.definition.d.ts.map +1 -0
  50. package/dist/esm/playbooks/top_down.brand-name-research/playbook.definition.js +41 -0
  51. package/dist/esm/playbooks/top_down.brand-name-research/playbook.definition.js.map +1 -0
  52. package/dist/esm/playbooks/top_down.brand-name-screening-produce/playbook.definition.d.ts +199 -0
  53. package/dist/esm/playbooks/top_down.brand-name-screening-produce/playbook.definition.d.ts.map +1 -0
  54. package/dist/esm/playbooks/top_down.brand-name-screening-produce/playbook.definition.js +201 -0
  55. package/dist/esm/playbooks/top_down.brand-name-screening-produce/playbook.definition.js.map +1 -0
  56. package/dist/esm/playbooks/top_down.brand-naming/playbook.definition.d.ts +67 -0
  57. package/dist/esm/playbooks/top_down.brand-naming/playbook.definition.d.ts.map +1 -0
  58. package/dist/esm/playbooks/top_down.brand-naming/playbook.definition.js +76 -0
  59. package/dist/esm/playbooks/top_down.brand-naming/playbook.definition.js.map +1 -0
  60. package/dist/esm/playbooks/top_down.business-model/playbook.definition.d.ts +3 -1
  61. package/dist/esm/playbooks/top_down.business-model/playbook.definition.d.ts.map +1 -1
  62. package/dist/esm/playbooks/top_down.business-model/playbook.definition.js +27 -18
  63. package/dist/esm/playbooks/top_down.business-model/playbook.definition.js.map +1 -1
  64. package/dist/esm/playbooks/top_down.compliance-analysis/playbook.definition.d.ts +11 -1
  65. package/dist/esm/playbooks/top_down.compliance-analysis/playbook.definition.d.ts.map +1 -1
  66. package/dist/esm/playbooks/top_down.compliance-analysis/playbook.definition.js +11 -11
  67. package/dist/esm/playbooks/top_down.compliance-analysis/playbook.definition.js.map +1 -1
  68. package/dist/esm/playbooks/top_down.compliance-conception/playbook.definition.d.ts +2 -0
  69. package/dist/esm/playbooks/top_down.compliance-conception/playbook.definition.d.ts.map +1 -1
  70. package/dist/esm/playbooks/top_down.compliance-conception/playbook.definition.js +3 -1
  71. package/dist/esm/playbooks/top_down.compliance-conception/playbook.definition.js.map +1 -1
  72. package/dist/esm/playbooks/top_down.compliance-posture/playbook.definition.d.ts +2 -0
  73. package/dist/esm/playbooks/top_down.compliance-posture/playbook.definition.d.ts.map +1 -1
  74. package/dist/esm/playbooks/top_down.compliance-posture/playbook.definition.js +3 -1
  75. package/dist/esm/playbooks/top_down.compliance-posture/playbook.definition.js.map +1 -1
  76. package/dist/esm/playbooks/top_down.definition/playbook.definition.d.ts +5 -2
  77. package/dist/esm/playbooks/top_down.definition/playbook.definition.d.ts.map +1 -1
  78. package/dist/esm/playbooks/top_down.definition/playbook.definition.js +16 -4
  79. package/dist/esm/playbooks/top_down.definition/playbook.definition.js.map +1 -1
  80. package/dist/esm/playbooks/top_down.phase-gate/playbook.definition.d.ts +7 -3
  81. package/dist/esm/playbooks/top_down.phase-gate/playbook.definition.d.ts.map +1 -1
  82. package/dist/esm/playbooks/top_down.phase-gate/playbook.definition.js +36 -2
  83. package/dist/esm/playbooks/top_down.phase-gate/playbook.definition.js.map +1 -1
  84. package/dist/esm/playbooks/top_down.risk-register/playbook.definition.d.ts +7 -1
  85. package/dist/esm/playbooks/top_down.risk-register/playbook.definition.d.ts.map +1 -1
  86. package/dist/esm/playbooks/top_down.risk-register/playbook.definition.js +4 -2
  87. package/dist/esm/playbooks/top_down.risk-register/playbook.definition.js.map +1 -1
  88. package/dist/esm/playbooks/top_down.scope-roadmap/playbook.definition.d.ts +9 -3
  89. package/dist/esm/playbooks/top_down.scope-roadmap/playbook.definition.d.ts.map +1 -1
  90. package/dist/esm/playbooks/top_down.scope-roadmap/playbook.definition.js +3 -3
  91. package/dist/esm/playbooks/top_down.scope-roadmap/playbook.definition.js.map +1 -1
  92. package/dist/esm/playbooks/top_down.vision-elicitation/playbook.definition.d.ts +35 -2
  93. package/dist/esm/playbooks/top_down.vision-elicitation/playbook.definition.d.ts.map +1 -1
  94. package/dist/esm/playbooks/top_down.vision-elicitation/playbook.definition.js +30 -17
  95. package/dist/esm/playbooks/top_down.vision-elicitation/playbook.definition.js.map +1 -1
  96. package/dist/knowledge/assurance-authority-bases/wildo-safe-baseline-v1/authority-basis.json +1 -1
  97. package/dist/knowledge/assurance-authority-bases/wildo-safe-baseline-v1/external-framework-mapping.json +1 -1
  98. package/dist/knowledge/manifest.json +8 -8
  99. package/dist/knowledge/playbooks/bottom_up.assemble-verify/method.md +22 -7
  100. package/dist/knowledge/playbooks/bottom_up.assemble-verify/playbook.json +3 -3
  101. package/dist/knowledge/playbooks/bottom_up.build/method.md +17 -15
  102. package/dist/knowledge/playbooks/bottom_up.build/playbook.json +4 -3
  103. package/dist/knowledge/playbooks/bottom_up.build-decompose/evaluation-protocol.json +1 -1
  104. package/dist/knowledge/playbooks/bottom_up.build-decompose/evaluation-treatment.json +1 -1
  105. package/dist/knowledge/playbooks/bottom_up.build-decompose/method.md +117 -28
  106. package/dist/knowledge/playbooks/bottom_up.build-decompose/playbook.json +4 -4
  107. package/dist/knowledge/playbooks/bottom_up.build-dispatch/method.md +133 -59
  108. package/dist/knowledge/playbooks/bottom_up.build-dispatch/playbook.json +2 -2
  109. package/dist/knowledge/playbooks/bottom_up.build-scale/method.md +169 -276
  110. package/dist/knowledge/playbooks/bottom_up.build-scale/playbook.json +2 -2
  111. package/dist/knowledge/playbooks/bottom_up.capability-composition/method.md +8 -1
  112. package/dist/knowledge/playbooks/bottom_up.capability-composition/playbook.json +1 -1
  113. package/dist/knowledge/playbooks/bottom_up.conception/method.md +3 -2
  114. package/dist/knowledge/playbooks/bottom_up.conception/playbook.json +2 -2
  115. package/dist/knowledge/playbooks/bottom_up.domain-plan/evaluation-protocol.json +1 -1
  116. package/dist/knowledge/playbooks/bottom_up.domain-plan/evaluation-treatment.json +1 -1
  117. package/dist/knowledge/playbooks/bottom_up.domain-plan/method.md +22 -6
  118. package/dist/knowledge/playbooks/bottom_up.domain-plan/playbook.json +4 -3
  119. package/dist/knowledge/playbooks/bottom_up.foundation/playbook.json +13 -5
  120. package/dist/knowledge/playbooks/refinement.compliance-document-candidate/method.md +5 -2
  121. package/dist/knowledge/playbooks/refinement.compliance-document-candidate/playbook.json +1 -1
  122. package/dist/knowledge/playbooks/top_down.brand-identity/evaluation-protocol.json +1 -1
  123. package/dist/knowledge/playbooks/top_down.brand-identity/evaluation-treatment.json +1 -1
  124. package/dist/knowledge/playbooks/top_down.brand-identity/method.md +71 -37
  125. package/dist/knowledge/playbooks/top_down.brand-identity/playbook.json +35 -38
  126. package/dist/knowledge/playbooks/top_down.brand-name-research/method.md +105 -0
  127. package/dist/knowledge/playbooks/top_down.brand-name-research/playbook.json +55 -0
  128. package/dist/knowledge/playbooks/top_down.brand-name-screening-produce/evaluation-protocol.json +1 -0
  129. package/dist/knowledge/playbooks/top_down.brand-name-screening-produce/evaluation-treatment.json +1 -0
  130. package/dist/knowledge/playbooks/top_down.brand-name-screening-produce/method.md +78 -0
  131. package/dist/knowledge/playbooks/top_down.brand-name-screening-produce/playbook.json +63 -0
  132. package/dist/knowledge/playbooks/top_down.brand-naming/method.md +56 -0
  133. package/dist/knowledge/playbooks/top_down.brand-naming/playbook.json +65 -0
  134. package/dist/knowledge/playbooks/top_down.business-model/evaluation-protocol.json +1 -1
  135. package/dist/knowledge/playbooks/top_down.business-model/evaluation-treatment.json +1 -1
  136. package/dist/knowledge/playbooks/top_down.business-model/method.md +60 -6
  137. package/dist/knowledge/playbooks/top_down.business-model/playbook.json +15 -10
  138. package/dist/knowledge/playbooks/top_down.compliance-analysis/playbook.json +30 -10
  139. package/dist/knowledge/playbooks/top_down.compliance-conception/playbook.json +3 -1
  140. package/dist/knowledge/playbooks/top_down.compliance-posture/playbook.json +4 -1
  141. package/dist/knowledge/playbooks/top_down.definition/method.md +8 -3
  142. package/dist/knowledge/playbooks/top_down.definition/playbook.json +6 -4
  143. package/dist/knowledge/playbooks/top_down.phase-gate/evaluation-protocol.json +1 -1
  144. package/dist/knowledge/playbooks/top_down.phase-gate/evaluation-treatment.json +1 -1
  145. package/dist/knowledge/playbooks/top_down.phase-gate/playbook.json +11 -1
  146. package/dist/knowledge/playbooks/top_down.risk-register/method.md +21 -0
  147. package/dist/knowledge/playbooks/top_down.risk-register/playbook.json +8 -2
  148. package/dist/knowledge/playbooks/top_down.scope-roadmap/playbook.json +4 -2
  149. package/dist/knowledge/playbooks/top_down.vision-elicitation/method.md +10 -0
  150. package/dist/knowledge/playbooks/top_down.vision-elicitation/playbook.json +40 -17
  151. package/dist/knowledge/references/brand.name-choice.config.md +127 -0
  152. package/dist/knowledge/references/brand.name-screening.config.md +253 -0
  153. package/dist/knowledge/references/build-plan.task-graph.config.md +64 -11
  154. package/dist/knowledge/references/business-model.pricing-plan.config.md +16 -0
  155. package/dist/knowledge/references/business-model.snapshot.config.md +64 -0
  156. package/dist/knowledge/references/business-model.snapshot.validation.md +68 -4
  157. package/dist/knowledge/references/capability-composition.config.md +67 -4
  158. package/dist/knowledge/references/example.data-seeding.minimal.md +29 -1
  159. package/dist/knowledge/references/market.snapshot.validation.md +1 -1
  160. package/dist/knowledge/references/saas.data-seeding.bootstrap-wiring.md +7 -1
  161. package/dist/knowledge/references/saas.data-seeding.definition-input.md +6 -0
  162. package/dist/knowledge/references/saas.models.queue.scheduled-recovery.md +81 -2
  163. package/dist/knowledge/references/saas.specifications.companion-exports.contract.md +10 -0
  164. package/dist/knowledge/references/saas.website.astro.label-loader.md +9 -0
  165. package/dist/knowledge/references/saas.website.factories.define-section.md +4 -0
  166. package/dist/knowledge/references/saas.website.page-manifest.schema.md +12 -3
  167. package/dist/knowledge/references/saas.website.page-sections.component.md +72 -0
  168. package/dist/knowledge/references/saas.website.sections.definition-schema.md +11 -0
  169. package/dist/knowledge/registries/assurance-authority-bases.json +1 -1
  170. package/dist/knowledge/registries/framework-topics.json +38 -1
  171. package/dist/knowledge/registries/playbooks.json +68 -20
  172. package/dist/knowledge/registries/source-tags.json +748 -695
  173. package/dist/knowledge/rules/live-context-door.md +5 -2
  174. package/dist/knowledge/skills/INDEX.md +2 -1
  175. package/dist/knowledge/skills/application-database-migrations/SKILL.md +1 -1
  176. package/dist/knowledge/skills/brand-name-screening/SKILL.md +154 -0
  177. package/dist/knowledge/skills/brand-name-screening/references/framework-brand-name-screening-api.md +394 -0
  178. package/dist/knowledge/skills/brand-name-screening/references/screening-procedure.md +65 -0
  179. package/dist/knowledge/skills/compliance-document-authoring/references/document-composition-privacy-notice.md +5 -2
  180. package/dist/knowledge/skills/config-saas/SKILL.md +31 -4
  181. package/dist/knowledge/skills/config-saas/references/compliance-declarations.md +14 -1
  182. package/dist/knowledge/skills/config-saas/references/framework-background-work-api.md +81 -2
  183. package/dist/knowledge/skills/email-add-template/SKILL.md +1 -1
  184. package/dist/knowledge/skills/i18n-rerun-pipeline/SKILL.md +1 -1
  185. package/dist/knowledge/skills/inbound-channels/SKILL.md +2 -2
  186. package/dist/knowledge/skills/local-environment-bootstrap/SKILL.md +11 -1
  187. package/dist/knowledge/skills/local-environment-bootstrap/references/derived-artifact-currency.md +122 -0
  188. package/dist/knowledge/skills/local-environment-bootstrap/references/framework-upgrade-and-dependency-alignment.md +11 -0
  189. package/dist/knowledge/skills/local-environment-bootstrap/references/wildo-cli-bootstrap-commands.md +9 -2
  190. package/dist/knowledge/skills/market-companion-prompt-enrichment/SKILL.md +11 -1
  191. package/dist/knowledge/skills/market-companion-prompt-enrichment/references/framework-companion-behavior-api.md +11 -1
  192. package/dist/knowledge/skills/market-message-templates-authoring/references/framework-message-template-api.md +1 -1
  193. package/dist/knowledge/skills/market-snapshot-validation/SKILL.md +1 -1
  194. package/dist/knowledge/skills/market-snapshot-validation/references/framework-snapshot-validation-api.md +1 -1
  195. package/dist/knowledge/skills/module-companion-exports/references/companion-exports-by-package.md +10 -0
  196. package/dist/knowledge/skills/module-relationships/SKILL.md +6 -4
  197. package/dist/knowledge/skills/seeding-authoring/SKILL.md +51 -3
  198. package/dist/knowledge/skills/seeding-authoring/references/framework-data-seeding-api.md +6 -0
  199. package/dist/knowledge/skills/seeding-operations/references/framework-data-seeding-operations-api.md +7 -1
  200. package/dist/knowledge/skills/spec-companion-exports/references/framework-companion-exports-contract-api.md +10 -0
  201. package/dist/knowledge/skills/website-factory-advanced/SKILL.md +4 -0
  202. package/dist/knowledge/skills/website-page-authoring/SKILL.md +84 -3
  203. package/dist/knowledge/skills/website-section-contact/SKILL.md +15 -0
  204. package/dist/knowledge/skills/website-section-contact/references/define-website-section-api.md +15 -0
  205. package/dist/knowledge/skills/website-section-cta/SKILL.md +15 -0
  206. package/dist/knowledge/skills/website-section-cta/references/define-website-section-api.md +15 -0
  207. package/dist/knowledge/skills/website-section-faq/SKILL.md +15 -0
  208. package/dist/knowledge/skills/website-section-faq/references/define-website-section-api.md +15 -0
  209. package/dist/knowledge/skills/website-section-feature-benefit-summary/SKILL.md +15 -0
  210. package/dist/knowledge/skills/website-section-feature-benefit-summary/references/define-website-section-api.md +15 -0
  211. package/dist/knowledge/skills/website-section-feature-detail/SKILL.md +15 -0
  212. package/dist/knowledge/skills/website-section-feature-detail/references/define-website-section-api.md +15 -0
  213. package/dist/knowledge/skills/website-section-footer-cta/SKILL.md +15 -0
  214. package/dist/knowledge/skills/website-section-footer-cta/references/define-website-section-api.md +15 -0
  215. package/dist/knowledge/skills/website-section-hero/SKILL.md +27 -9
  216. package/dist/knowledge/skills/website-section-hero/references/define-website-section-api.md +15 -0
  217. package/dist/knowledge/skills/website-section-pricing/SKILL.md +15 -0
  218. package/dist/knowledge/skills/website-section-pricing/references/define-website-section-api.md +15 -0
  219. package/dist/knowledge/skills/website-section-social-proof/SKILL.md +15 -0
  220. package/dist/knowledge/skills/website-section-social-proof/references/define-website-section-api.md +15 -0
  221. package/dist/knowledge/skills/website-seo-aeo-geo/SKILL.md +1 -1
  222. package/dist/knowledge/skills/website-seo-aeo-geo/references/aeo-api.md +11 -0
  223. package/dist/knowledge/topic-index.json +169 -43
  224. package/dist/tsconfig.build.tsbuildinfo +1 -1
  225. package/dist/tsconfig.tsbuildinfo +1 -1
  226. package/package.json +4 -4
@@ -1,4 +1,4 @@
1
- # Build Scale — Run the Build at Scale: Waves, Parallelism, Interleaved Generation (v1 scaffold)
1
+ # Build Scale — Run the Build at Scale: Waves, Parallelism, Interleaved Generation
2
2
 
3
3
  ### Target Application Framing
4
4
 
@@ -12,310 +12,203 @@ Read and apply these patterns from the application-developer point of view:
12
12
  - how the application should avoid mixing frontend, backend, and infrastructure concerns
13
13
 
14
14
 
15
- > **This is a v1 scaffold methodology, and a deliberately honest one.** It is a
16
- > working first cut of the *scale* orchestration — derive a wave of ready tasks,
17
- > run several coding agents in parallel under an isolation model, interleave the
18
- > deterministic companion-generation steps, and reconcile wave completion. But
19
- > the companion runner **hard-enforces single-flight today**: it refuses a run
20
- > plan with more than one coding-agent task, refuses a second dispatch while one
21
- > is active, holds one workspace-mutation lock per run, and has no run-outcome
22
- > reconciler. So the concurrency described below is the **declarative target** a
23
- > runner-serializer slice will realize. **On the current runner the waves
24
- > degrade to sequential single-flight dispatch** — read the "What the runner
25
- > actually does today" section before you assume anything ran in parallel. Do
26
- > not overclaim concurrency the runner cannot do.
15
+ > **The build already runs at scale, and this playbook describes how.** A run of
16
+ > a build-execution playbook starts a **wave**: up to N build-plan tasks at once
17
+ > in the application's one worktree, N being the application's **build
18
+ > parallelism** (`wildo workbench parallelism`, 5 unless the person set it). This
19
+ > playbook and the build-dispatch playbook declare the same executor and run
20
+ > through the same runtime branch, so they start the SAME wave. Prefer the
21
+ > build-dispatch playbook to start a build; read this one for how a wave is
22
+ > scheduled, how wide it is, and how it is reconciled. With a build parallelism
23
+ > of `1` the lane dispatches one task at a time (the single-flight mode).
27
24
 
28
25
  ## Where you are in the journey
29
26
 
30
- This is the **third and final build-execution playbook** — the pivot from
31
- **dispatching one task safely** (the build-dispatch playbook's single-flight
32
- contract) to **running the build at scale**. It runs in the **bottom-up phase**,
33
- after the build plan exists (`artifact_exists build-plan`) and the foundation
34
- step has authored the app's config and skeleton. You
35
- are not re-opening the decomposition here, and you are not re-specifying how one
36
- task dispatches — you are deciding **how many** tasks run **concurrently** and
37
- **safely**, in **what order**, with the companion-generation steps interleaved.
27
+ This is the **third build-execution playbook**. It runs in the **bottom-up
28
+ phase**, after the build plan exists (`artifact_exists build-plan`) and the
29
+ foundation step has authored the app's config and skeleton. You are not
30
+ re-opening the decomposition here, and you are not re-specifying how one task
31
+ dispatches — you are reading **how many** tasks run **concurrently**, in **what
32
+ order**, with the companion-generation steps interleaved, and how the wave
33
+ **ends**.
38
34
 
39
- Upstream: the **build-decompose** playbook produced the ordered task graph you
40
- schedule; the **build-dispatch** playbook defined how ONE task dispatches
41
- safely — you **invoke that contract per parallel slot**, you do not re-author
42
- it. **You author neither of those concerns** — see the boundary section.
35
+ Upstream: the **build-decompose** playbook produced the ordered task graph the
36
+ wave schedules; the **build-dispatch** playbook defines the per-task contract
37
+ every slot runs (context pack, one ACP prompt turn, change-set capture, verify
38
+ outside the agent, record + `onFailure`). **You author neither of those
39
+ concerns** — see the boundary section.
43
40
 
44
41
  ## What this playbook does
45
42
 
46
- You are the **scale orchestrator**. This playbook reads the approved build plan
47
- and the current journey state and answers, across the *whole* build: *"which set
48
- of tasks is ready to run now, how many run at once and on what isolation, when
49
- do the deterministic generation steps run, and when has a wave actually
50
- finished?"* You do **not** produce a new journey artifact — the observable
51
- result of the build is the per-task change-sets + task statuses the runner
52
- records on the task rows (the build-dispatch playbook's mechanism), not a
53
- schedule artifact authored here (see "What you emit"). You do **not** decompose
54
- the plan, re-specify the per-task dispatch mechanics, author any schema, or
55
- write any code.
56
-
57
- ## The scale contract — you sequence waves, the dispatch playbook runs each slot
58
-
59
- Most of the *per-task* mechanism is **not yours**: selecting one task,
60
- composing its context pack, running one prompt turn against the coding agent over
61
- the Agent Client Protocol (answering its permission requests within the declared
62
- ceiling), capturing the change-set, running the verify hook, holding the
63
- single-flight lock, applying the host's actual execution limits — all of that is the **build-dispatch** contract, which each
64
- parallel slot invokes. Your method here is the **declarative concurrency
65
- recipe** the runner (or a future runner-serializer) interprets: *which set* of
66
- tasks is ready, *how many* run at once, *how* they are isolated, *when* the
67
- deterministic generation steps interleave, and *when* a wave is done. Do not
68
- hand-drive git, do not compose a dispatch prompt, do not re-implement the
69
- single-flight capture — invoke the dispatch contract per slot and let it own
70
- those.
43
+ You are the **scale reader**. Across the *whole* build you answer: *"which tasks
44
+ are ready now, how many run at once, when do the deterministic generation steps
45
+ run, and when has a wave actually finished?"* You do **not** produce a journey
46
+ artifact — the observable result of the build is the per-task change-sets and
47
+ task statuses the runner records on the task rows (see "What you emit"). You do
48
+ **not** decompose the plan, re-specify the per-task dispatch mechanics, author
49
+ any schema, or write any code.
71
50
 
72
51
  ## Read state first — no artifact to write
73
52
 
74
- This playbook is **idempotent** and, on a re-run, *advances the schedule by one
75
- wave* (and, on today's single-flight runner, by one task within that wave).
76
- Before scheduling anything:
53
+ This playbook is **idempotent** and, on a re-run, *advances the build by one
54
+ wave*. Before reasoning about a wave:
77
55
 
78
- 1. Read the current journey state and scope from what this run is given — the loaded
79
- specifications and the workspace sections of your brief — and do not re-derive it. Confirm the **build plan** is
80
- present and approved and the **foundation** ran. For each build-plan task,
81
- read its dispatch state from its task row — **not yet dispatched**,
56
+ 1. Read the current journey state and scope from what this run is given — the
57
+ loaded specifications and the workspace sections of your brief — and do not
58
+ re-derive it. Confirm the **build plan** is present and approved and the
59
+ **foundation** ran. For each build-plan task, read its dispatch state from
60
+ its task row (`wildo build tasks`) — **not yet dispatched**,
82
61
  **dispatched-and-succeeded** (a captured change-set + a passed verify),
83
62
  **dispatched-and-failed** (a captured change-set but a failed/timed-out run
84
- or a failed verify), or **in-flight**. This full state — not just "has a
85
- change-set" — is what the wave derivation keys on (under `keep_and_flag` a
86
- *failed* task also has a captured change-set, so presence of a change-set
87
- alone never means "done").
63
+ or a failed verify), or **in-flight**. Under `keep_and_flag` a *failed* task
64
+ also has a captured change-set, so the presence of a change-set alone never
65
+ means "done".
88
66
  2. Read the approved **build plan** (`build-plan` — the ordered task graph, with
89
- each task's prose prerequisite references). It is your work list.
67
+ each task's declared prerequisite references). It is the work list.
90
68
 
91
69
  That there is an application to dispatch into is answered by the committed
92
- `./wildo.saas.config.ts` and skeleton, not by a journey artifact. (A
93
- `foundation-manifest` played that role until 2026-08-25 and was retired with the
94
- app-creation verification/digest aspect.) The honest limitation the foundation step
95
- carries still applies: on-disk skeleton materialization depends on greenfield
96
- bring-up tooling (deferred); the dogfood app already has a skeleton, so the scale
97
- contract runs against an existing workspace.
70
+ `./wildo.saas.config.ts` and skeleton, not by a journey artifact.
98
71
 
99
72
  // Operator status: `wildo context journey` (eligibility board + runs in flight); per-task landed
100
73
  // state is read from git by the runner.
101
74
 
102
- You **read** `build-plan`; you **write** no journey
103
- artifact — the runner records each dispatched task's change-set + status on its
104
- row.
105
-
106
- ## Step 1 — Derive the ready-set (the wave / frontier)
107
-
108
- Walk the build plan and compute the **frontier**: the set of tasks that are
109
- **not yet dispatched** (or previously-**failed** tasks cleared for retry) and
110
- whose prose prerequisites have **all SUCCEEDED**. That set is the next **wave**.
111
-
112
- - **A prerequisite must have SUCCEEDED, not merely been attempted.** A task
113
- enters the frontier only when every task it depends on is
114
- dispatched-and-succeeded (verify passed). A prerequisite that is
115
- dispatched-and-**failed** (its change-set flagged for triage under
116
- `keep_and_flag`), still **in-flight**, or un-run does **not** satisfy the
117
- dependency — its dependents stay **out of the frontier** until it succeeds.
118
- This is the U13b readiness discipline, lifted to a set: do not open a wave on
119
- a schema whose run has not verified.
120
- - **The wave respects the build-plan ordering.** The build-decompose playbook
121
- encoded the DEPENDS_ON edges (the per-resource stage order schema → config →
122
- ui-behavior → spec, the cross-topic edges, relationship-after-endpoints); the
123
- frontier is exactly the tasks with no unsatisfied edge remaining. You do not
124
- re-derive or re-order those edges — you read them and compute the ready-set.
125
- - **An empty frontier is a terminal state.** If no task is ready — every task
126
- has succeeded, or the only undispatched tasks are blocked behind an
127
- unresolved failure — the build is done (or blocked on a flagged failure):
128
- report it and stop. An empty build plan is the same.
129
-
130
- ## Step 2 — Decide the parallelism degree + the worktree-isolation model
131
-
132
- For the frontier, decide **how many** coding agents run at once and **on what
133
- isolation**. This is the core scale decision, and it is bounded by a hard
134
- safety constraint, not just throughput:
135
-
136
- - **Change-set attribution requires isolation.** The dispatch contract captures
137
- each run's change-set by diffing the worktree around the run and attributing
138
- every file as `RUN_EXCLUSIVE`, `APPEARED_DURING_RUN`, or
139
- `BASELINE_DIRTY_AMBIGUOUS`. **Two coding agents mutating the SAME worktree
140
- concurrently make every touched file `BASELINE_DIRTY_AMBIGUOUS`** — each run
141
- looks like it started against a tree the other was dirtying — which
142
- **destroys per-run attribution**. So real parallelism needs one of:
143
- - **git worktree-per-agent isolation** — each concurrent coding agent runs in
144
- its own git worktree (a disjoint checkout of the same repo), so each run's
145
- change-set is captured against a tree only it mutates; the worktrees are
146
- reconciled/merged after the wave. This is the general answer.
147
- - **module / resource partitioning** — schedule concurrent agents only onto
148
- tasks whose targets touch **disjoint file sets** (different modules /
149
- resources), so even on one shared worktree no two runs collide. This is the
150
- cheaper answer when the frontier naturally partitions by module.
151
- - **or keep coding single-flight** and only parallelise the deterministic
152
- companion-generation steps (Step 3), which are not coding-locked.
153
- - **Pick the degree deliberately, not maximally.** At scaffold depth prefer a
154
- small degree (or a module partition) over max fan-out — cost-optimal packing,
155
- a priority / work-stealing scheduler, and backoff are a later depth phase, not
156
- this scaffold. State the isolation model you chose; a wave with degree > 1 and
157
- **no** isolation answer is unsafe (the reviewer rubric flags it as a GAP).
158
-
159
- ## Step 3 — Interleave the deterministic companion-generation steps
75
+ You **read** `build-plan`; you **write** no journey artifact.
76
+
77
+ ## Step 1 — The ready-set (the frontier)
78
+
79
+ A task is **ready** when it is not yet dispatched (or a previously-**failed**
80
+ task cleared for retry), it is not already running, and every prerequisite it
81
+ declares has **SUCCEEDED** — not merely been attempted. A prerequisite that is
82
+ dispatched-and-failed, in-flight, or un-run does **not** satisfy the dependency.
83
+ The runner resolves this from the declared prerequisite refs and from git; it is
84
+ never inferred from prose, and you do not re-order the edges the build-decompose
85
+ playbook encoded.
86
+
87
+ The wave is a **rolling pool**, not a fixed batch: when a run settles, its slot
88
+ is refilled with whatever is ready THEN, so a task that just landed unlocks its
89
+ dependents within the same wave. An empty frontier with no run live ends the
90
+ wave; if every task has landed the build is done (and the wave reports whether
91
+ the application assembles), and if the only remaining tasks are stuck the wave
92
+ names them and why.
93
+
94
+ ## Step 2 — The width and the isolation model, as shipped
95
+
96
+ - **The width is the build parallelism.** `wildo workbench parallelism` shows it
97
+ and sets it, from `1` to a safety ceiling; the default is 5. The runner reads it
98
+ at the start of every wave, so a change needs no restart. Each slot is a live
99
+ coding agent spending on its own, so the width is the person's decision, not
100
+ the method's: do not propose a different degree as part of a run.
101
+ - **The runs share ONE worktree, by decision.** Concurrent agents editing the same
102
+ tree reconcile fine in practice, so the lane does not isolate them (no
103
+ worktree per agent, no partition by module). It prevents only a task being
104
+ built twice at once. Everything else is DETECTED and REPAIRED (Step 4).
105
+ - **Attribution is recorded, not guaranteed.** Each run still captures its
106
+ change-set and attributes every file (`RUN_EXCLUSIVE`, `APPEARED_DURING_RUN`,
107
+ `BASELINE_DIRTY_AMBIGUOUS`). In a wave, files a sibling also touched are
108
+ expected to come out ambiguous; the run's `concurrency` entry names the runs
109
+ that overlapped it and the paths both touched, which is what a reviewer reads
110
+ to settle authorship.
111
+ - **One wave is bounded.** A call takes at most four tasks per slot and then
112
+ stops taking new ones, waits for its live runs, and returns. The next run of a
113
+ build-execution playbook continues from where the build stands.
114
+ - **One wave at a time.** The companion refuses a second wave while one runs —
115
+ submitted, running, or in its end-of-wave commit and repair — because a second
116
+ wave would double the width and write while the first one repairs.
117
+
118
+ ## Step 3 — The deterministic companion-generation steps
160
119
 
161
120
  The build plan carries **companion-owned generation steps** — labels / i18n,
162
- email templates, seeds — as tasks. These are **deterministic companion
163
- endpoints**, not coding-agent dispatches: they do **not** contend for the
164
- coding single-flight lock. It is tempting to conclude they can freely run
165
- *alongside* a coding run — but there is a subtlety that makes that unsafe on
166
- today's runner, see the caveat below.
167
-
168
- - **Run each generation step at its input-ready point.** A labels/i18n step runs
169
- after the resources it labels exist; an email step after its templates'
170
- definition surfaces exist; a seed step after the resources it seeds exist.
171
- Schedule each at its input-ready point in the wave sequence — do not block a
172
- coding wave on them and do not run them before their inputs exist. (The
173
- *ordering* is what matters here; the *concurrency* is deferred — see the
174
- caveat.)
175
- - **Route them to their deterministic endpoint, not the coding lane.** Do not
176
- dispatch a coding agent for a labels/email/seed step — it has a deterministic
177
- companion generator. Treating a generation step as a coding dispatch is scope
178
- creep the reviewer flags.
179
- - **CAVEAT — no attribution-safe overlap with an active coding run today.** The
180
- coding run's change-set is captured as a **whole-worktree time-window diff**
181
- (every tracked-dirty + untracked file in the window). So a generation step
182
- writing label / email / seed files **while a coding dispatch is capturing** is
183
- misattributed to that coding run — its files show up as the coding run's own
184
- edits. This is the *same* whole-worktree pollution that forbids two concurrent
185
- coding agents (Step 2), and it applies identically to coding + generation. So
186
- on today's runner the generation steps must **also** be serialized to run
187
- **outside** an active coding-capture window (i.e. between coding runs). Genuine
188
- overlap — generation alongside coding, and coding alongside coding — both
189
- arrive only with git worktree-per-agent isolation (the runner task in Step 5's
190
- gap note). Until then, keep the input-ready *ordering* and serialize.
191
-
192
- // The trigger is `wildo creation run <playbookRef>` (or `/wildo:run` from a Claude Code session),
193
- // which reaches the companion's run route with the local token attached. Each coding slot invokes the
194
- // single-flight dispatch contract; each generation step runs its deterministic endpoint.
195
-
196
- ## Step 4 — Check the actual host authorization boundary
197
-
198
- A coding dispatch writes the application workspace. Read its selected task, readiness,
199
- mutation scope and the configured coding host's authorization policy before invoking it.
200
- The single-flight mutation lock still applies to the current shared workspace.
201
-
202
- The playbook's `approval` declaration and the journey autonomy presets express policy
203
- intent. The current playbook runner does not interpret them as pre-dispatch gates,
204
- automatic authorization, or a GUIDED/YOLO mapping. A declared `approve_before` does
205
- not create a pending Authority Request, and no authority workflow exists to create one —
206
- never invent one from an enum.
207
-
208
- Likewise, do not infer enforced per-playbook or per-journey run ceilings from declared
209
- budget fields. Respect the limits actually reported by the execution host. Inspect the
210
- recorded changes and verification result after each task; no preset commits or accepts them.
211
-
212
- ## Step 5 — Reconcile wave completion, then advance
213
-
214
- A wave is done only when **every task in the frontier reached a terminal
215
- SUCCEEDED outcome** (change-set captured + verify passed) or is explicitly held
216
- as a blocked / flagged failure. Only then does the next frontier open (Step 1).
217
-
218
- - **Do not open the next wave on an unfinished one.** Advancing while a wave's
219
- tasks are still in-flight, or with an unreconciled failure, risks a downstream
220
- task building on an unfinished prerequisite. Reconcile first.
221
- - **Failure disposition across siblings + downstream** — this playbook's
222
- declared `onFailure` describes **`keep_and_flag`**; the current host retains changes for review. A failed/timed-out task keeps its captured
223
- change-set flagged for human triage (never silently reverted, never an
224
- un-recorded half-applied state). Its already-dispatched **siblings** in the
225
- same wave are left to complete; its **dependents** in later waves stay
226
- **blocked** (Step 1) until it is triaged and re-dispatched successfully.
227
- Triage is a human decision: keep + fix forward, unwind the flagged change-set
228
- by hand, or re-dispatch once the blocker is resolved. This playbook does not
229
- auto-retry a failed task (automated retry / backoff is a deferred build-loop
230
- resilience concern).
231
-
232
- > **Why keep_and_flag, not revert_changeset (a v1 boundary).** Reverting a
233
- > failed run's edits is a *mutating* git op, deliberately **outside** the
234
- > read-only change-set carve-out the dispatch lane observes. So `keep_and_flag`
235
- > is both the correct F-5 workspace-safety posture and the behavior the runner
236
- > enforces today; a `revert_changeset` path is a deliberate later boundary.
237
-
238
- Then **loop**: on the next run, the build plan's task records reflect the just-completed
239
- wave's outcomes (`wildo build tasks`), and you derive the next frontier (Step 1). Close a
240
- significant stretch of the build with a progress summary — which waves ran, which
241
- change-sets landed, what verify said, which generation steps interleaved, and
242
- any failed task left flagged for triage.
243
-
244
- ## What the runner actually does today (do not overclaim)
245
-
246
- Be honest with the creator about the gap between this method and the runner:
247
-
248
- - **The runner enforces single-flight.** It refuses a run plan carrying more
249
- than one coding-agent task, refuses a second dispatch while one is active, and
250
- holds one workspace-mutation lock per background run. So a wave with
251
- parallelism degree > 1 is **not** realizable on today's runner — the wave
252
- **degrades to sequential single-flight dispatch** (repeated invocation of the
253
- build-dispatch contract, one task at a time, in frontier order). The frontier
254
- math, the readiness discipline, and the reconciliation are all still correct
255
- and useful sequentially — you just do not get concurrency yet.
256
- - **No overlap is attribution-safe today — not even a companion-generation
257
- step.** The generation endpoints are not coding-locked, but the coding
258
- change-set capture is a whole-worktree time-window diff, so a generation step
259
- writing files while a coding dispatch is capturing is misattributed to that
260
- run (Step 3's caveat). So the generation steps **also** serialize outside an
261
- active coding-capture window: their input-ready *ordering* holds, only the
262
- *concurrency* is deferred. Safe overlap of generation-with-coding, and
263
- coding-with-coding, both require git worktree-per-agent isolation.
264
- - **The concurrency this method describes is the declarative target** for a
265
- runner-serializer slice (multi-coding-agent run-completion serialization,
266
- git worktree-per-agent isolation, a run-outcome reconciler, and a cursor lock)
267
- — filed as runner tasks, **not** built by this scaffold playbook. Until they
268
- land, run the waves sequentially and do not claim a parallel run (or a
269
- generation-alongside-coding overlap) happened when it did not.
270
-
271
- ## Scaffold-depth boundary (owned by a sibling, or deferred to a later phase)
272
-
273
- Keep this playbook at **scale-orchestration depth**. Explicitly **out of scope
274
- here**:
121
+ email templates, seeds — as tasks. They run through their **deterministic
122
+ companion endpoint**, never a coding agent.
123
+
124
+ - **They run at their input-ready point**, like any task: a labels step after the
125
+ resources it labels exist, a seed step after the resources it seeds exist. Their
126
+ readiness is declared in the plan the same way.
127
+ - **In a wave they run in their turn while the coding runs already started keep
128
+ going.** The runner runs a ready generation step inline as it fills the wave;
129
+ it does not wait for the live coding runs to settle first. A generation step's
130
+ files that land inside a coding run's capture window are recorded as overlap on
131
+ that run, and the end-of-wave repair commits them with the rest.
132
+ - **Each generation step runs at most once per wave.** If its landing is still unmet
133
+ afterwards, the wave stops taking tasks rather than re-select it forever; the
134
+ next wave tries again, and a step that never lands is a stuck task to report.
135
+
136
+ // The trigger is `wildo build run` from a console (or `/wildo:build` in a Claude Code session),
137
+ // which starts a wave through the build-dispatch playbook; `wildo creation run <playbookRef>`
138
+ // starts one through this playbook or that one alike.
139
+
140
+ ## Step 4 — How a wave ends: reconcile, repair, advance
141
+
142
+ A wave stops taking tasks for one of four reasons — nothing ready, its per-call
143
+ cap, a generation step it already ran, or a refused admission — and it always
144
+ drains the runs already live before it ends.
145
+
146
+ - **The repair, once nothing is writing.** The runner commits what the runs' own
147
+ accepts left in the tree as ONE change naming every run whose writes it may
148
+ carry, failed runs included (a failed run's files sit in the same tree). Then
149
+ each run that COMPLETED but whose declared product a sibling overwrote is
150
+ re-armed and re-run ALONE, so nothing can overwrite the second attempt; the
151
+ number of such re-runs per task is bounded, after which the task waits for a
152
+ person. Each repair is on the wave's result and on the task's record
153
+ (`interference`).
154
+ - **Failure disposition across siblings + downstream** — `keep_and_flag`. A
155
+ failed or timed-out run keeps its captured change-set flagged for human triage
156
+ (never silently reverted). Its siblings in the same wave run to completion. Its
157
+ **dependents** stay out of every frontier until it is triaged and re-dispatched
158
+ successfully. This playbook does not auto-retry a failure that was the task's
159
+ own; only a suspected interference is re-run.
160
+ - **Advance.** On the next run, the task records reflect the wave's outcomes
161
+ (`wildo build tasks`), and the next wave selects from what is ready now. Close a
162
+ significant stretch of the build with a progress summary — which waves ran,
163
+ which change-sets landed, what verify said, which runs overlapped or were
164
+ repaired, and any task left flagged for triage.
165
+
166
+ > **Why keep_and_flag, not revert_changeset.** Reverting a failed run's edits is
167
+ > a *mutating* git op, deliberately **outside** the read-only change-set
168
+ > carve-out the dispatch lane observes; with a shared worktree it would also
169
+ > reach a sibling's files. A `revert_changeset` path is a deliberate later
170
+ > boundary.
171
+
172
+ ## Step 5 — Check the actual host authorization boundary
173
+
174
+ A coding dispatch writes the application workspace. Each run is confined to its
175
+ own task's mutation scope by the configured coding host's authorization policy;
176
+ nothing confines it from its siblings, which is what Steps 2 and 4 are about.
177
+
178
+ The playbook's `approval` declaration and the journey autonomy presets express
179
+ policy intent. The current playbook runner does not interpret them as
180
+ pre-dispatch gates, automatic authorization, or a GUIDED/YOLO mapping. A declared
181
+ `approve_before` does not create a pending Authority Request — never invent one
182
+ from an enum. Do not infer enforced run ceilings from declared budget fields;
183
+ respect the limits actually reported by the execution host.
184
+
185
+ ## Boundary (owned by a sibling, or deferred to a later phase)
186
+
187
+ Explicitly **out of scope here**:
275
188
 
276
189
  - **Re-decomposition.** You do NOT split modules/resources into tasks, re-order
277
- the graph, or add tasks — the **build-decompose** playbook owns the task
278
- graph. You read it and schedule from it.
279
- - **The per-task dispatch contract.** You do NOT re-specify single-flight
280
- locking, context-pack composition, change-set capture, or the verify hook —
281
- the **build-dispatch** playbook owns "how ONE task dispatches safely", and
282
- each parallel slot **invokes** that contract. If you catch yourself describing
283
- how a single task captures its change-set, stop — that is the dispatch
284
- playbook.
285
- - **The runner-serializer mechanism.** The multi-coding-agent serializer, git
286
- worktree-per-agent provisioning, the run-outcome reconciler, and the cursor
287
- lock are **companion-runner mechanism**, filed as runner tasks — this
288
- declarative method does not build them.
289
- - **A priority / work-stealing scheduler + cost-optimal packing.** Deciding an
290
- optimal task ordering within a band, work-stealing across idle slots, backoff,
291
- and cost-optimal packing are a later depth phase — the v1 scaffold picks a
292
- small parallelism degree (or a module partition) and a simple frontier order.
293
- - **A `revert_changeset` failure path.** Deferred; `keep_and_flag` is the v1
190
+ the graph, or add tasks — the **build-decompose** playbook owns the task graph.
191
+ - **The per-task dispatch contract.** You do NOT re-specify context-pack
192
+ composition, change-set capture, or the verify hook — the **build-dispatch**
193
+ playbook owns "how one task dispatches safely", and every slot runs it.
194
+ - **Choosing the width or a different isolation model.** The width is the
195
+ application's build parallelism, set by the person; the shared worktree with
196
+ detection and repair is the lane's decision. A priority / work-stealing
197
+ scheduler and cost-optimal packing are a later depth phase.
198
+ - **A `revert_changeset` failure path.** Deferred; `keep_and_flag` is the
294
199
  disposition.
295
- - **Auto-executed reviewer checklist.** The scale-soundness checklist ships
296
- **declarative** — the deterministic typecheck gate runs automatically; nothing
297
- executes the checklist prose (rubric-only judge gates were retired at the
298
- flexible-intelligence hard cut). Do not assume it auto-ran.
299
200
  - **Schema / field authoring + code generation.** The **coding agents** author
300
201
  fields, schemas, ui-behavior, and code at dispatch time — never the
301
202
  orchestrator.
302
203
 
303
- If you find yourself designing how ONE task dispatches safely rather than how
304
- MANY tasks are scheduled + reconciled, stop — that is the build-dispatch
305
- playbook.
306
-
307
204
  ## What you emit
308
205
 
309
- **No journey artifact.** This is a pure-orchestration playbook: the observable
310
- result of the build — each task's **change-set** (files touched, attributed,
311
- summarized) and **status** — is materialized by the runner on the dispatched
312
- task rows (the build-dispatch playbook's mechanism), not authored here. A
313
- separate schedule / wave-log artifact family is deliberately not added: the task
314
- rows already *are* the dispatch record and a second home for the same record
315
- would drift (a typed wave-schedule / run-ledger contract is a filed later-phase
316
- item). Use the configured host's actual authorization boundary and review its
317
- recorded changes and verification outcomes. The approval declaration alone does
318
- not authorize or gate dispatch.
206
+ **No journey artifact.** The observable result of the build — each task's
207
+ **change-set** and **status**, plus each run's overlap and repair record — is
208
+ materialized by the runner on the task rows and on the wave's result, not
209
+ authored here. A separate wave-log artifact family is deliberately not added:
210
+ the task rows already *are* the dispatch record, and a second home for it would
211
+ drift.
319
212
 
320
213
  ## References
321
214
 
@@ -324,6 +217,6 @@ not authorize or gate dispatch.
324
217
 
325
218
  ## Reviewer checklist — `scale-schedule-sound`
326
219
 
327
- Formerly authored as an `llm_judge` quality gate; the generic rubric-only gate kind was retired at the flexible-intelligence hard cut (judgment now runs only through a compiled Evaluation Treatment). The checklist itself is creator-facing review guidance and is kept here, verbatim, minus the retired runner's closing sentence:
220
+ Creator-facing review guidance (the generic rubric-only gate kind was retired at the flexible-intelligence hard cut; nothing executes this checklist):
328
221
 
329
- Judge whether the SCALE orchestration ran the build soundly — the trust-but-verify reviewer pass over how MANY tasks were scheduled, parallelised, and reconciled, WITHOUT straying into how ONE task dispatches (that is the build-dispatch playbook's rubric) or how the plan was decomposed (the build-decompose playbook's). Assess six invariants and mark each PASS / CONCERN / GAP with a one-line evidence note. (1) WAVE READINESS — every task dispatched in a wave had ALL its prose prerequisites already SUCCEEDED (dispatched-and-verified), not merely attempted; a task dispatched while a prerequisite was un-run, in-flight, or dispatched-and-FAILED is a GAP (the build stage ran against a missing prerequisite). A prerequisite whose run FAILED (its change-set flagged under keep_and_flag) does NOT satisfy the dependency — its dependents must stay blocked, and dispatching them anyway is a GAP. (2) PARALLELISM SAFETY / ATTRIBUTION — the concurrent coding runs in a wave were isolated so each run's change-set is cleanly attributable (files RUN_EXCLUSIVE / APPEARED_DURING_RUN per run), via git worktree-per-agent isolation OR a module/resource partition that puts every concurrent agent on a DISJOINT file set. Two concurrent coding agents mutating the SAME worktree — which makes every touched file BASELINE_DIRTY_AMBIGUOUS and destroys per-run attribution — is a GAP (the single-flight floor was bypassed without an isolation answer). Note the honest floor: the runner enforces single-flight today, so on the current runner a 'wave' degrades to SEQUENTIAL single-flight dispatch and this invariant reads whether the METHOD declared a sound isolation model, not whether concurrency actually ran. (3) COMPANION-GENERATION INTERLEAVE — the deterministic companion-owned generation steps (labels / i18n, email templates, seeds) ran through their deterministic companion endpoints AFTER the resources / modules whose output they consume existed, at the right point in the schedule order; a generation step ordered before its inputs exist, or a missing generation step for authored surfaces, is a CONCERN; treating a deterministic generation step as a coding-agent dispatch (dispatching an agent for it) is a CONCERN. Note the honest floor: the generation steps do NOT take the coding single-flight lock, but the coding change-set capture is a whole-worktree time-window diff, so a generation step writing files WHILE a coding dispatch is capturing is misattributed to that coding run — so on today's runner the generation steps also serialize outside an active coding-capture window, and this invariant reads whether they ran at the right point in the ORDER, not whether they truly overlapped a coding run (safe overlap needs worktree isolation). (4) WAVE-COMPLETION RECONCILIATION — the next wave opened only after every task in the current frontier reached a terminal SUCCEEDED outcome (or was explicitly held as a blocked / flagged failure); advancing the schedule while a wave's tasks were still in-flight or with an unreconciled failure is a GAP (a downstream task could build on an unfinished prerequisite). (5) FAILURE DISPOSITION ACROSS SIBLINGS + DOWNSTREAM — a failed / timed-out task in a wave kept its captured change-set flagged for human triage (keep_and_flag), its already-dispatched siblings in the same wave were left to complete, and its DEPENDENTS in later waves stayed blocked until it was triaged and re-dispatched successfully; a failed task whose dependents were dispatched anyway, or a failure silently reverted / left un-recorded, is a GAP (F-5 workspace-safety violation). (6) BOUNDARY DISCIPLINE — the orchestration stays at SCALE depth: it derives ready-sets, schedules waves, sets the parallelism + isolation model, interleaves the companion-generation steps, and reconciles completion. It does NOT re-decompose the plan (the build-decompose playbook owns the task graph), does NOT re-specify the per-task dispatch mechanics (single-flight lock / context-pack / change-set capture / verify — the build-dispatch playbook owns those; the scale orchestrator INVOKES that contract per slot), and does NOT itself author schemas / fields / code (the coding agents do that at dispatch time). Evidence of re-decomposition, re-authoring the single-task dispatch contract, or the orchestrator authoring code itself is a CONCERN (scope creep out of the scale lane). (The checklist is declarative: nothing executes it; the current host's deterministic checks apply; the approval enum alone adds no gate.)
222
+ Judge whether a WAVE ran the build soundly — how MANY tasks were scheduled, run concurrently, and reconciled, WITHOUT straying into how ONE task dispatches (the build-dispatch playbook's checklist) or how the plan was decomposed (the build-decompose playbook's). Assess six invariants and mark each PASS / CONCERN / GAP with a one-line evidence note. (1) WAVE READINESS — every task the wave dispatched had ALL its declared prerequisites already SUCCEEDED (dispatched-and-verified), not merely attempted; a task dispatched while a prerequisite was un-run, in-flight, or dispatched-and-FAILED is a GAP. (2) WIDTH — the wave ran no more coding runs at once than the application's build parallelism; more is a GAP, and a width chosen by the method rather than by the person's setting is a CONCERN. (3) OVERLAP ACCOUNTED FOR — the runs shared one worktree by design, so overlap is expected; each run that overlapped another carries a `concurrency` record naming the runs and paths, and a change-set dominated by BASELINE_DIRTY_AMBIGUOUS files is resolved against that record. An overlap the reviewer cannot account for from the records is a CONCERN; a run accepted as clean while its record says `interferenceSuspected` is a GAP. (4) REPAIR — after the runs settled, the wave's end-of-run commit names the runs whose writes it carries, and every completed run whose product a sibling overwrote was re-run alone (or is shown waiting for a person once its re-runs were spent); a lost product with no repair and no record is a GAP. (5) FAILURE DISPOSITION ACROSS SIBLINGS + DOWNSTREAM — a failed / timed-out task kept its captured change-set flagged for human triage (keep_and_flag), its siblings in the wave were left to complete, and its DEPENDENTS stayed out of later frontiers until it was triaged and re-dispatched successfully; a failed task whose dependents were dispatched anyway, or a failure silently reverted / left un-recorded, is a GAP. (6) GENERATION STEPS + BOUNDARY — the deterministic companion-generation steps ran through their companion endpoints at their input-ready point, at most once per wave, and never as a coding-agent dispatch (a CONCERN otherwise); the orchestration did not re-decompose the plan, re-specify the per-task contract, or author code itself (a CONCERN otherwise).