mixdog 1.4.3 → 1.4.5

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 (281) hide show
  1. package/package.json +1 -2
  2. package/src/cli.mjs +5 -1
  3. package/src/defaults/shipped-history.json +30 -2
  4. package/src/defaults/skills/browser-use/SKILL.md +148 -273
  5. package/src/defaults/skills/browser-use/references/devtools.md +14 -0
  6. package/src/defaults/skills/browser-use/references/page-flows.md +66 -0
  7. package/src/defaults/skills/browser-use/references/sign-in.md +70 -0
  8. package/src/defaults/skills/code-tidy/SKILL.md +60 -84
  9. package/src/defaults/skills/code-tidy/references/agent-cleanup.md +27 -2
  10. package/src/defaults/skills/computer-use/SKILL.md +145 -229
  11. package/src/defaults/skills/computer-use/references/flows.md +35 -0
  12. package/src/defaults/skills/computer-use/references/operations.md +22 -0
  13. package/src/defaults/skills/computer-use/references/platforms.md +12 -0
  14. package/src/defaults/skills/docx/SKILL.md +23 -31
  15. package/src/defaults/skills/docx/references/native-authoring.md +26 -3
  16. package/src/defaults/skills/docx/references/redlining.md +15 -0
  17. package/src/defaults/skills/image/SKILL.md +37 -77
  18. package/src/defaults/skills/image/references/prompt-forms.md +51 -0
  19. package/src/defaults/skills/image/references/text.md +6 -0
  20. package/src/defaults/skills/local-provider/SKILL.md +75 -105
  21. package/src/defaults/skills/local-provider/references/lifecycle.md +14 -0
  22. package/src/defaults/skills/local-provider/references/maintenance.md +19 -0
  23. package/src/defaults/skills/local-provider/references/recovery.md +30 -0
  24. package/src/defaults/skills/memory-management/SKILL.md +22 -21
  25. package/src/defaults/skills/pdf/SKILL.md +21 -23
  26. package/src/defaults/skills/pdf/references/blocks.md +20 -0
  27. package/src/defaults/skills/pdf/references/forms.md +12 -0
  28. package/src/defaults/skills/pdf/references/html.md +2 -2
  29. package/src/defaults/skills/pdf/references/ocr.md +7 -0
  30. package/src/defaults/skills/pdf/references/security.md +8 -0
  31. package/src/defaults/skills/pptx/SKILL.md +30 -109
  32. package/src/defaults/skills/pptx/references/audit.md +45 -0
  33. package/src/defaults/skills/pptx/references/brief-critique.md +47 -0
  34. package/src/defaults/skills/pptx/references/editing.md +24 -0
  35. package/src/defaults/skills/pptx/references/html.md +1 -1
  36. package/src/defaults/skills/pptx/references/script.md +33 -0
  37. package/src/defaults/skills/setup/SKILL.md +34 -102
  38. package/src/defaults/skills/setup/references/actions.md +73 -14
  39. package/src/defaults/skills/setup/references/surfaces.md +23 -8
  40. package/src/defaults/skills/setup-guide/SKILL.md +28 -115
  41. package/src/defaults/skills/setup-guide/references/features.md +36 -0
  42. package/src/defaults/skills/setup-guide/references/glossary.md +38 -0
  43. package/src/defaults/skills/setup-guide/references/workflow-models.md +62 -0
  44. package/src/defaults/skills/test-suite/SKILL.md +15 -127
  45. package/src/defaults/skills/test-suite/references/case-management.md +17 -9
  46. package/src/defaults/skills/test-suite/references/diagnose.md +23 -0
  47. package/src/defaults/skills/test-suite/references/prune.md +27 -0
  48. package/src/defaults/skills/test-suite/references/write.md +44 -0
  49. package/src/defaults/skills/video/SKILL.md +28 -42
  50. package/src/defaults/skills/video/references/prompting.md +31 -0
  51. package/src/defaults/skills/visualize/SKILL.md +73 -107
  52. package/src/defaults/skills/visualize/references/craft.md +1 -2
  53. package/src/defaults/skills/visualize/references/feedback.md +36 -0
  54. package/src/defaults/skills/visualize/references/showing.md +13 -0
  55. package/src/defaults/skills/xlsx/SKILL.md +23 -50
  56. package/src/defaults/skills/xlsx/references/editing.md +15 -0
  57. package/src/defaults/skills/xlsx/references/model-conventions.md +1 -1
  58. package/src/defaults/skills/xlsx/references/recalculation.md +37 -0
  59. package/src/defaults/skills/xlsx/references/report-design.md +18 -2
  60. package/src/lib/keychain-cjs.cjs +22 -1
  61. package/src/lib/rules-builder.cjs +1 -1
  62. package/src/rules/agent/AGENT.md +3 -0
  63. package/src/rules/agent-mode/MIXY.md +5 -3
  64. package/src/rules/lead/LEAD.md +3 -1
  65. package/src/runtime/agent/orchestrator/agent-runtime/cache-strategy.mjs +8 -30
  66. package/src/runtime/agent/orchestrator/agent-runtime/title-completion.mjs +5 -0
  67. package/src/runtime/agent/orchestrator/mcp/oauth-request.mjs +21 -2
  68. package/src/runtime/agent/orchestrator/providers/account-pool.mjs +3 -0
  69. package/src/runtime/agent/orchestrator/providers/admission-scheduler.mjs +4 -3
  70. package/src/runtime/agent/orchestrator/providers/anthropic-api-request.mjs +11 -1
  71. package/src/runtime/agent/orchestrator/providers/anthropic-api-transport.mjs +2 -0
  72. package/src/runtime/agent/orchestrator/providers/anthropic-message-threads.mjs +583 -0
  73. package/src/runtime/agent/orchestrator/providers/anthropic-oauth-client-version.mjs +1 -1
  74. package/src/runtime/agent/orchestrator/providers/anthropic-oauth.mjs +81 -19
  75. package/src/runtime/agent/orchestrator/providers/anthropic-sse-turn/turn-events/content-block-start.mjs +4 -0
  76. package/src/runtime/agent/orchestrator/providers/anthropic-sse-turn/turn-outcome.mjs +2 -0
  77. package/src/runtime/agent/orchestrator/providers/anthropic-sse.mjs +180 -13
  78. package/src/runtime/agent/orchestrator/providers/anthropic-turn-result.mjs +2 -0
  79. package/src/runtime/agent/orchestrator/providers/anthropic.mjs +58 -6
  80. package/src/runtime/agent/orchestrator/providers/cursor-wire.mjs +4 -2
  81. package/src/runtime/agent/orchestrator/providers/custom-provider-config.mjs +121 -9
  82. package/src/runtime/agent/orchestrator/providers/custom-provider.mjs +4 -1
  83. package/src/runtime/agent/orchestrator/providers/gemini-error-contract.mjs +6 -6
  84. package/src/runtime/agent/orchestrator/providers/gemini-model-redirect.mjs +44 -0
  85. package/src/runtime/agent/orchestrator/providers/gemini-reasoning-relay.mjs +5 -0
  86. package/src/runtime/agent/orchestrator/providers/gemini-request-body.mjs +34 -5
  87. package/src/runtime/agent/orchestrator/providers/gemini-response.mjs +10 -0
  88. package/src/runtime/agent/orchestrator/providers/gemini-schema.mjs +11 -6
  89. package/src/runtime/agent/orchestrator/providers/gemini-sdk-request.mjs +4 -100
  90. package/src/runtime/agent/orchestrator/providers/gemini-stream.mjs +3 -227
  91. package/src/runtime/agent/orchestrator/providers/gemini.mjs +226 -118
  92. package/src/runtime/agent/orchestrator/providers/lib/anthropic-request-utils.mjs +11 -1
  93. package/src/runtime/agent/orchestrator/providers/lib/claude-bridge-schema.mjs +15 -3
  94. package/src/runtime/agent/orchestrator/providers/lib/gemini-model-catalog.mjs +2 -1
  95. package/src/runtime/agent/orchestrator/providers/lib/generation-window.mjs +9 -0
  96. package/src/runtime/agent/orchestrator/providers/openai-compat-chat-send.mjs +9 -1
  97. package/src/runtime/agent/orchestrator/providers/openai-compat-chat-stream/stream-result.mjs +2 -0
  98. package/src/runtime/agent/orchestrator/providers/openai-compat-chat-stream/stream-state.mjs +5 -0
  99. package/src/runtime/agent/orchestrator/providers/openai-compat-options.mjs +3 -0
  100. package/src/runtime/agent/orchestrator/providers/openai-compat-response-normalization.mjs +2 -0
  101. package/src/runtime/agent/orchestrator/providers/openai-compat-responses.mjs +10 -1
  102. package/src/runtime/agent/orchestrator/providers/openai-direct-request.mjs +5 -28
  103. package/src/runtime/agent/orchestrator/providers/openai-http-sse-response-state/response-events.mjs +3 -0
  104. package/src/runtime/agent/orchestrator/providers/openai-http-sse-response-state/response-outcome.mjs +2 -0
  105. package/src/runtime/agent/orchestrator/providers/openai-ws-pool.mjs +50 -28
  106. package/src/runtime/agent/orchestrator/providers/openai-ws-response-state.mjs +9 -2
  107. package/src/runtime/agent/orchestrator/providers/openai-ws-stream/socket-handlers.mjs +14 -10
  108. package/src/runtime/agent/orchestrator/providers/openai-ws.mjs +1 -3
  109. package/src/runtime/agent/orchestrator/providers/openrouter-provider-routing.mjs +28 -0
  110. package/src/runtime/agent/orchestrator/providers/openrouter.mjs +4 -1
  111. package/src/runtime/agent/orchestrator/providers/registry.mjs +30 -25
  112. package/src/runtime/agent/orchestrator/providers/stream-json-pool/worker-slots/slot-spawn.mjs +8 -3
  113. package/src/runtime/agent/orchestrator/providers/vertex-ai-maas.mjs +2 -2
  114. package/src/runtime/agent/orchestrator/runtime-core/ask-user-result.mjs +66 -0
  115. package/src/runtime/agent/orchestrator/runtime-core/config-helpers.mjs +4 -5
  116. package/src/runtime/agent/orchestrator/runtime-core/goal-tool-defs.mjs +2 -2
  117. package/src/runtime/agent/orchestrator/runtime-core/provider-request-tools.mjs +15 -8
  118. package/src/runtime/agent/orchestrator/runtime-core/session-title.mjs +1 -0
  119. package/src/runtime/agent/orchestrator/runtime-core/todo-list.mjs +9 -32
  120. package/src/runtime/agent/orchestrator/runtime-core/todo-tool-defs.mjs +4 -5
  121. package/src/runtime/agent/orchestrator/session/compact/summary-route.mjs +1 -3
  122. package/src/runtime/agent/orchestrator/session/loop/compact-policy.mjs +5 -2
  123. package/src/runtime/agent/orchestrator/session/loop/no-tool-turn/todo-check.mjs +2 -17
  124. package/src/runtime/agent/orchestrator/session/loop/usage.mjs +7 -3
  125. package/src/runtime/agent/orchestrator/session/manager/ask-turn-settle.mjs +9 -2
  126. package/src/runtime/agent/orchestrator/session/manager/runtime-liveness.mjs +8 -1
  127. package/src/runtime/agent/orchestrator/session/manager/session-close.mjs +26 -13
  128. package/src/runtime/agent/orchestrator/session/manager/session-lifecycle.mjs +4 -5
  129. package/src/runtime/agent/orchestrator/session/manager/usage-metrics.mjs +16 -5
  130. package/src/runtime/agent/orchestrator/session/store/lifecycle-barriers.mjs +22 -1
  131. package/src/runtime/agent/orchestrator/session/store/live-state.mjs +4 -0
  132. package/src/runtime/agent/orchestrator/session/store.mjs +12 -6
  133. package/src/runtime/agent/orchestrator/tools/builtin/cache-layers.mjs +11 -9
  134. package/src/runtime/agent/orchestrator/tools/builtin/grep-formatting.mjs +2 -8
  135. package/src/runtime/agent/orchestrator/tools/builtin/lib/search-input-helpers.mjs +1 -9
  136. package/src/runtime/agent/orchestrator/tools/builtin/lib/shell-output-silence.mjs +64 -0
  137. package/src/runtime/agent/orchestrator/tools/builtin/lib/shell-task-output.mjs +20 -6
  138. package/src/runtime/agent/orchestrator/tools/builtin/local-search-telemetry.mjs +2 -2
  139. package/src/runtime/agent/orchestrator/tools/builtin/native-search-client.mjs +1 -1
  140. package/src/runtime/agent/orchestrator/tools/builtin/runtime-capabilities.mjs +12 -3
  141. package/src/runtime/agent/orchestrator/tools/builtin/task-tool.mjs +12 -1
  142. package/src/runtime/agent/orchestrator/tools/code-graph/ast-calls.mjs +9 -2
  143. package/src/runtime/agent/orchestrator/tools/code-graph/build.mjs +137 -7
  144. package/src/runtime/agent/orchestrator/tools/code-graph/calls-cache.mjs +43 -14
  145. package/src/runtime/agent/orchestrator/tools/code-graph/calls-sidecar-worker.mjs +33 -0
  146. package/src/runtime/agent/orchestrator/tools/code-graph/disk-cache.mjs +146 -11
  147. package/src/runtime/agent/orchestrator/tools/code-graph/dispatch.mjs +31 -24
  148. package/src/runtime/agent/orchestrator/tools/code-graph/graph-binary/spawn-jsonl.mjs +172 -25
  149. package/src/runtime/agent/orchestrator/tools/code-graph/graph-model.mjs +100 -2
  150. package/src/runtime/agent/orchestrator/tools/code-graph/main-thread-slice.mjs +18 -0
  151. package/src/runtime/agent/orchestrator/tools/code-graph/mask-pool.mjs +219 -0
  152. package/src/runtime/agent/orchestrator/tools/code-graph/mask-worker.mjs +12 -0
  153. package/src/runtime/agent/orchestrator/tools/code-graph/memory-cache.mjs +6 -1
  154. package/src/runtime/agent/orchestrator/tools/code-graph/modes/calls.mjs +9 -2
  155. package/src/runtime/agent/orchestrator/tools/code-graph/modes/dependents.mjs +5 -1
  156. package/src/runtime/agent/orchestrator/tools/code-graph/modes/structure.mjs +8 -2
  157. package/src/runtime/agent/orchestrator/tools/code-graph/modes/symbols.mjs +14 -11
  158. package/src/runtime/agent/orchestrator/tools/code-graph/search-references.mjs +41 -18
  159. package/src/runtime/agent/orchestrator/tools/code-graph/search.mjs +391 -83
  160. package/src/runtime/agent/orchestrator/tools/code-graph/symbol-index.mjs +109 -9
  161. package/src/runtime/agent/orchestrator/tools/code-graph/text-mask.mjs +73 -29
  162. package/src/runtime/agent/orchestrator/tools/code-graph-prewarm-worker.mjs +6 -5
  163. package/src/runtime/agent/orchestrator/tools/lib/native-spawn-client.mjs +7 -3
  164. package/src/runtime/channels/lib/crash-log.mjs +21 -1
  165. package/src/runtime/channels/lib/routines-db.mjs +35 -0
  166. package/src/runtime/channels/lib/schedules-db.mjs +4 -0
  167. package/src/runtime/channels/lib/webhooks-db.mjs +4 -0
  168. package/src/runtime/channels/lib/worker-main.mjs +8 -25
  169. package/src/runtime/computer-bridge/action-schema.mjs +2 -2
  170. package/src/runtime/computer-bridge/core-actions.mjs +16 -2
  171. package/src/runtime/computer-bridge/result-canonical.mjs +6 -0
  172. package/src/runtime/media/jobs.mjs +5 -0
  173. package/src/runtime/media/store.mjs +146 -3
  174. package/src/runtime/memory/tool-defs.mjs +1 -1
  175. package/src/runtime/office/core/office-write-ledger.mjs +2 -2
  176. package/src/runtime/shared/agent-reap-state.mjs +19 -0
  177. package/src/runtime/shared/desktop-notice-host.d.mts +13 -0
  178. package/src/runtime/shared/desktop-notice-host.mjs +87 -0
  179. package/src/runtime/shared/file-lock.mjs +2 -2
  180. package/src/runtime/shared/heap-cap.mjs +9 -11
  181. package/src/runtime/shared/llm/http-agent.mjs +11 -1
  182. package/src/runtime/shared/llm/system-proxy.mjs +100 -25
  183. package/src/runtime/shared/llm/usage-accounting.mjs +121 -19
  184. package/src/runtime/shared/llm/usage-context.mjs +3 -3
  185. package/src/runtime/shared/llm/usage-ledger-quota.mjs +2 -0
  186. package/src/runtime/shared/llm/usage-ledger-requests.mjs +190 -0
  187. package/src/runtime/shared/llm/usage-ledger-rollup.mjs +7 -3
  188. package/src/runtime/shared/llm/usage-ledger-worker.mjs +2 -0
  189. package/src/runtime/shared/llm/usage-ledger.mjs +69 -1
  190. package/src/runtime/shared/llm/usage-request-status.mjs +48 -0
  191. package/src/runtime/shared/provider-accounts.mjs +22 -0
  192. package/src/runtime/shared/provider-auth-binding.mjs +2 -2
  193. package/src/runtime/shared/resource-admission.mjs +2 -2
  194. package/src/runtime/shared/root-async-context.mjs +51 -0
  195. package/src/runtime/shared/run-child.mjs +18 -1
  196. package/src/runtime/shared/slow-span.mjs +2 -2
  197. package/src/runtime/shared/tool-execution-owner.mjs +2 -2
  198. package/src/runtime/shared/tool-result-summary.mjs +1 -1
  199. package/src/runtime/shared/tool-work-units.mjs +2 -0
  200. package/src/runtime/shared/turn-snapshot-store.mjs +75 -18
  201. package/src/runtime/shared/turn-worktree-snapshot.mjs +38 -8
  202. package/src/runtime/shared/user-cwd.mjs +2 -2
  203. package/src/runtime/shared/worker-requests.mjs +3 -1
  204. package/src/runtime/shared/worktrees/lock.mjs +2 -2
  205. package/src/runtime/visualize-bridge/client.mjs +26 -4
  206. package/src/runtime/visualize-bridge/tool-defs.mjs +46 -5
  207. package/src/session-runtime/agent-home-delivery.mjs +76 -8
  208. package/src/session-runtime/agent-task-sessions.mjs +63 -22
  209. package/src/session-runtime/boot/apis.mjs +16 -0
  210. package/src/session-runtime/boot/begin.mjs +2 -0
  211. package/src/session-runtime/boot/tools.mjs +8 -2
  212. package/src/session-runtime/internal-tool-executor/ask-user-tool.mjs +3 -58
  213. package/src/session-runtime/media-api.mjs +15 -0
  214. package/src/session-runtime/model-route/route-persist.mjs +12 -4
  215. package/src/session-runtime/provider-models/row-hydration.mjs +2 -0
  216. package/src/session-runtime/provider-readiness.mjs +10 -2
  217. package/src/session-runtime/routine-owner.mjs +9 -3
  218. package/src/session-runtime/runtime-core.mjs +35 -8
  219. package/src/session-runtime/runtime-review-api.mjs +3 -1
  220. package/src/session-runtime/services/agent-task-status.mjs +29 -1
  221. package/src/session-runtime/services/agent-tool/close-flow.mjs +6 -1
  222. package/src/session-runtime/services/agent-tool/execute-resume.mjs +94 -0
  223. package/src/session-runtime/services/agent-tool/execute-send.mjs +77 -27
  224. package/src/session-runtime/services/agent-tool/execute-spawn.mjs +28 -7
  225. package/src/session-runtime/services/agent-tool/execute.mjs +6 -1
  226. package/src/session-runtime/services/agent-tool/helpers.mjs +11 -0
  227. package/src/session-runtime/services/agent-tool/job-views/job-render.mjs +13 -10
  228. package/src/session-runtime/services/agent-tool/job-views/session-progress.mjs +3 -0
  229. package/src/session-runtime/services/agent-tool/render.mjs +8 -1
  230. package/src/session-runtime/services/agent-tool/send-flow.mjs +1 -0
  231. package/src/session-runtime/services/agent-tool/spawn-flow/spawn-prep.mjs +6 -0
  232. package/src/session-runtime/services/agent-tool/spawn-flow/spawn-run.mjs +2 -1
  233. package/src/session-runtime/services/agent-tool/tag-registry/reaped-sessions.mjs +82 -0
  234. package/src/session-runtime/services/agent-tool/tag-registry/record-facts-worker.mjs +36 -0
  235. package/src/session-runtime/services/agent-tool/tag-registry.mjs +4 -0
  236. package/src/session-runtime/services/agent-tool/terminal-reaper/row-helpers.mjs +29 -0
  237. package/src/session-runtime/services/agent-tool/terminal-reaper/scan-admission.mjs +11 -3
  238. package/src/session-runtime/services/agent-tool/terminal-reaper/stale-rows.mjs +19 -8
  239. package/src/session-runtime/services/agent-tool/terminal-reaper.mjs +5 -2
  240. package/src/session-runtime/services/agent-tool/worker-index.mjs +1 -0
  241. package/src/session-runtime/services/channel-admin.mjs +44 -3
  242. package/src/session-runtime/services/daemon-crash-capture/daemon-exit-record.mjs +2 -2
  243. package/src/session-runtime/services/plugin-admin.mjs +46 -7
  244. package/src/session-runtime/services/provider-admin.mjs +5 -0
  245. package/src/session-runtime/session-create.mjs +5 -0
  246. package/src/session-runtime/setup-tool/agent-settings-actions.mjs +1 -1
  247. package/src/session-runtime/setup-tool/executor.mjs +1 -1
  248. package/src/session-runtime/setup-tool/extended-actions.mjs +1 -1
  249. package/src/session-runtime/setup-tool/settings-contract.mjs +4 -2
  250. package/src/session-runtime/usage-stats-api.mjs +120 -1
  251. package/src/session-runtime/workflow-agents-api/agent-connectors.mjs +36 -29
  252. package/src/session-runtime/workflow-agents-api/agent-editor.mjs +183 -23
  253. package/src/session-runtime/workspace/admission-gate.mjs +25 -7
  254. package/src/session-runtime/workspace/transitions.mjs +4 -0
  255. package/src/standalone/daemon-log.mjs +28 -4
  256. package/src/standalone/daemon-owner.mjs +41 -7
  257. package/src/standalone/daemon-work-gate.mjs +139 -0
  258. package/src/standalone/daemon.mjs +80 -11
  259. package/src/standalone/desktop-service-registry.mjs +11 -4
  260. package/src/standalone/session-client.mjs +57 -5
  261. package/src/standalone/session-protocol.mjs +6 -0
  262. package/src/standalone/session-service/agent-tree/agent-cancel.mjs +17 -1
  263. package/src/standalone/session-service/agent-tree.mjs +7 -0
  264. package/src/standalone/session-service/retention/busy.mjs +3 -2
  265. package/src/standalone/session-service/session-calls/session-turns.mjs +12 -5
  266. package/src/standalone/session-transport/call-route.mjs +10 -4
  267. package/src/standalone/session-transport/routes.mjs +21 -0
  268. package/src/standalone/session-transport/stack.mjs +3 -1
  269. package/src/standalone/session-transport.mjs +5 -0
  270. package/src/tui/app/model-picker/model-footer.mjs +1 -1
  271. package/src/tui/app/model-picker/provider-models-list.mjs +7 -3
  272. package/src/tui/app/model-picker/route-selection/fast-parameters.mjs +23 -3
  273. package/src/tui/app/model-picker/route-selection.mjs +1 -0
  274. package/src/tui/dist/index.mjs +536 -484
  275. package/src/tui/session/session-api/integrations.mjs +3 -0
  276. package/src/tui/session/session-api/media.mjs +5 -0
  277. package/src/tui/session/tool-approval-settle.mjs +24 -4
  278. package/src/tui/session/tool-approval.mjs +35 -10
  279. package/src/tui/session/turn.mjs +13 -0
  280. package/src/runtime/agent/orchestrator/providers/gemini-sdk-stream/cancellation.mjs +0 -71
  281. package/src/runtime/agent/orchestrator/providers/gemini-sdk-stream/reader.mjs +0 -108
@@ -0,0 +1,38 @@
1
+ # Glossary
2
+
3
+ Read this to answer a concept question. Answer in a sentence or two from the
4
+ matching entry, then stop.
5
+
6
+ - Bot (봇): a contact you chat with in bot mode (봇 모드). Mixy is the representative
7
+ bot; the user can have it make others, such as a GitHub Watcher.
8
+ - Agent (에이전트): a workflow sub-agent such as Worker or Reviewer that
9
+ does part of a task. Users add their own on the Workflows page with the
10
+ "에이전트 +" button. Bots are people-facing; agents work behind a task.
11
+ - Default agents (기본 에이전트): built-in helpers listed on Workflows, such as
12
+ Web Search (runs the web search tool), the Maintainer (memory upkeep and
13
+ session titles) and Explore (code search for the Lead). They need a model
14
+ route to work.
15
+ - Agent mode (에이전트 모드): a coding work session run by Lead and agents.
16
+ A session a bot starts for a job is an agent task session (에이전트 작업
17
+ 세션); others are just sessions (세션).
18
+ - Sessions and projects: an agent-mode session runs in a Project, a folder whose
19
+ instructions and memories it loads.
20
+ - Workflows and work rules: a workflow sets how a request gets done; work
21
+ rules are the instructions it applies.
22
+ - Connectors and extensions: connectors are MCP services such as Notion or
23
+ Linear (`connect_connector`, the user presses the card); extensions are
24
+ plugins that add tools and skills that teach how to work.
25
+ - Group rooms (그룹 대화방): one chat with several bots. `@name` or "봇 부르기"
26
+ addresses one bot; otherwise the Maintainer decides who answers next.
27
+ - Schedules and webhooks: a timed or incoming-event trigger (`save_automation`).
28
+ Each run is one turn in its bot's own chat, with that bot's persona, rules,
29
+ permissions, memory and model. Without a bot they cannot be used. A routine
30
+ can also name an executor sub-agent (`executorAgentId`, e.g. Worker): that
31
+ sub-agent does each run on its own model and the bot only relays its report.
32
+ In the Schedules/Webhooks editor this is the "Executor" field (Korean UI:
33
+ "실행 에이전트"); its default "The bot itself" ("직접 처리") means the bot
34
+ does the run itself.
35
+ - Approval cards: when a tool needs permission it waits in the chat until the
36
+ user approves or declines.
37
+ - Phone remote access: pair a phone under Settings → Connection to follow and
38
+ steer tasks away from the PC.
@@ -0,0 +1,62 @@
1
+ # Workflow step: models, delegation and Auto reasoning
2
+
3
+ Read this at the workflow step, before recommending any model or roster.
4
+
5
+ ## Model picks
6
+
7
+ Recommend from `${MIXDOG_SKILL_DIR}/model-recommendations.json`, role by role.
8
+
9
+ - Use only providers the user is signed in to (`status providers`) and only a
10
+ model the provider's model list actually offers.
11
+ - Picks are an ordered list of any length: recommend the first one that
12
+ qualifies, otherwise take the next.
13
+ - A subscription sign-in comes first: when the user has one, take the first
14
+ qualifying pick it offers before any pick that would bill an API key; fall
15
+ back to API-key providers only when no subscription offers one.
16
+ - A pick may carry `plans`: when the Lead provider's plan (`plan` in
17
+ `status providers`) is listed there, recommend that model instead. A null or
18
+ unlisted plan keeps the pick as is; never fail on an unknown plan.
19
+ - Roles with `preferOtherProviderThan` (Reviewer, Advisor) work best with the
20
+ opposite tendency from the model doing the work: pick the first qualifying
21
+ model whose provider differs from the chosen Lead's and Heavy Worker's
22
+ providers, and fall back to the plain first pick only when the user has a
23
+ single signed-in provider. Give the reason in one line: a different model
24
+ family catches what the main one misses.
25
+ - If `asOf` is older than `staleAfterDays`, say the list may be out of date.
26
+ - Never print ids or file names.
27
+ - Suggest turning on the Maintainer with a cheap model at the lowest effort; it
28
+ also decides turns in group rooms.
29
+ - Offer Explore the same way, from its own picks: a cheap model that searches
30
+ code for the Lead in separate parallel sessions.
31
+
32
+ ## Delegation and roster
33
+
34
+ Delegation and roster come from `delegation` in the data file.
35
+
36
+ 1. Ask once how the user likes to work: one agent doing everything, or
37
+ splitting work across agents. Say plainly that Solo is a good choice at any
38
+ plan: it is the most efficient, and with a strong model its quality is often
39
+ not much lower; delegation trades more usage for speed and parallel work.
40
+ 2. Suggest the mode whose `plans` lists the Lead provider's plan (the default
41
+ suggestion); the user's preference overrides it. Ask once, set it with
42
+ `set_orchestration_mode`, and say it can be changed later in settings.
43
+ 3. The roster follows the chosen mode: turn on only `baseRoster` plus that
44
+ mode's `agents`, using `set_agent_route` with `route.disabled` false for
45
+ what the user accepts and true for the rest; do not enable others.
46
+ 4. Offer `leanReview` for Solo or focused as a cost saver: the Lead reviews its
47
+ own work, Reviewer stays off, Advisor is on but used rarely (only when
48
+ stuck). Advisor in balanced or swarm is a light, occasional second opinion
49
+ from a different model family at key decision points, not on every task.
50
+ 5. Say in one line that more agents can be turned on later on the Workflows
51
+ page.
52
+
53
+ ## Auto reasoning
54
+
55
+ Right after the models are picked, explain it: Auto adjusts reasoning depth per
56
+ message around a medium baseline, so hard work may come out slightly lower in
57
+ quality but cost is much lower. The Lead runs the main coding work, where
58
+ stability matters most, so also recommend fixing the Lead's effort at High or
59
+ above in the model picker (choosing an effort there turns Auto off for that
60
+ model). Ask whether to turn Auto on (`install_builtin` `autoEffort`). If
61
+ declined, say in one line that it can be turned on later by choosing Auto at the
62
+ top of the model picker's reasoning effort list.
@@ -20,13 +20,15 @@ device by hand is outside it.
20
20
 
21
21
  ## Pick the mode
22
22
 
23
- | Request | Mode |
24
- |---|---|
25
- | Add a test, fix a bug test-first | Write |
26
- | Check a change while editing | Run |
27
- | A test fails sometimes, only in a full run, only in CI, only on this machine | Diagnose |
28
- | Duplicate, dead, obsolete or worthless tests | Prune |
29
- | Review the suite's health or a whole area | Audit |
23
+ Read the mode's procedure before acting.
24
+
25
+ | Request | Mode | Procedure |
26
+ |---|---|---|
27
+ | Add a test, fix a bug test-first | Write | `references/write.md`; adding a new test area also `references/case-management.md` |
28
+ | Check a change while editing | Run | below |
29
+ | A test fails sometimes, only in a full run, only in CI, only on this machine | Diagnose | `references/diagnose.md` |
30
+ | Duplicate, dead, obsolete or worthless tests; any change that edits an existing test file | Prune | `references/prune.md` |
31
+ | Review the suite's health or a whole area | Audit | `references/case-management.md` |
30
32
 
31
33
  ## Prerequisites: find the project's own way
32
34
 
@@ -50,44 +52,12 @@ Tests run on the user's working machine. These hold in every mode.
50
52
  reruns, fix attempts, heavy runs and stopping what you started; follow it.
51
53
  - Never loop a test to "reproduce" flakiness here, and never shard in
52
54
  parallel locally; repetition belongs to CI history.
53
- - Confirm by PID or command line that detached processes you started are gone.
54
-
55
- ## Write
56
-
57
- What a change gets:
58
-
59
- | Change | New tests |
60
- |---|---|
61
- | Bug fix | One regression test for the reported behaviour |
62
- | New behaviour | One per observable behaviour the requirement names (its success and its stated rejections) |
63
- | Refactor, rename, move | None; the existing tests must still pass unchanged |
64
- | Removal | None; delete the removed behaviour's tests, never assert the absence |
65
- | Docs, copy, style, config values | None |
66
-
67
- Only a test owning a lasting contract joins the suite and its CI lane.
68
- Reproductions, probes and one-off checks are scratch: they live in a temp
69
- directory outside the repository, are never committed, and are deleted when
70
- the task ends. A scratch reproduction that proves a bug becomes the
71
- regression test above or is deleted, never both kept.
72
-
73
- 1. Name the behaviour and its owner: the one contract this test owns, beside
74
- the code it exercises. If a test already owns it, extend that test
75
- (`references/suite-design.md`). Result: one file and test name chosen.
76
- 2. Name the production change that would make the test fail; if you cannot,
77
- redesign it around an observable behaviour (`references/suite-design.md`).
78
- Write the test through the user's interface (caller, CLI,
79
- HTTP, rendered output) with the expected value from the requirement, not
80
- from running the code. For a new regression test, show it failing first:
81
- the failure names the missing behaviour. An existing failing test already
82
- counts as that evidence.
83
- 3. Apply `references/reliability.md` for any time, process, port, file,
84
- environment, locale or platform dependency. Result: no fixed sleep, no
85
- fixed port, no unrestored global, no host-locale assumption.
86
- 4. Make it pass with the smallest change, then rerun the same file. Result:
87
- a new regression test's fail-before and pass-after pair goes into the
88
- report. Add no further
89
- tests unless the user asks.
90
- 5. Put it in the lane its duration and needs call for (`suite-design.md`).
55
+ - Never skip or disable a failing test: fix it, or report it and leave the
56
+ decision to the user. Only a capability skip (a platform or tool the test
57
+ needs is absent) is allowed, and its reason names that capability.
58
+ - Confirm by PID or command line that detached processes you started are gone:
59
+ a load generator or server left detached outlives the agent and can freeze
60
+ the machine. Nothing detached survives a report.
91
61
 
92
62
  ## Run
93
63
 
@@ -101,86 +71,6 @@ regression test above or is deleted, never both kept.
101
71
  checks.
102
72
  4. After two evidence-backed fix attempts, report what was tried and the log.
103
73
 
104
- ## Diagnose
105
-
106
- An intermittent failure is a defect in the test or the product, never noise
107
- (`references/flaky-policy.md`).
108
-
109
- 1. Collect evidence without reruns: the failing assertion, the full log, what
110
- else was running, the OS and locale. Result: the failing line and the
111
- condition it waited on.
112
- 2. Classify the cause with the table in `references/reliability.md`. Result:
113
- one named cause with the line that assumes it.
114
- 3. Propose the fix: in the test, or in the product when real users can hit
115
- the same race. A longer timeout, a retry or a skip is not a fix.
116
-
117
- Diagnose ends at a report: cause, evidence, proposed fix. Only when the user
118
- asked for a fix, apply it and prove it with the focused file plus a
119
- deterministic trigger of the cause (a held promise, a fake clock, a localized
120
- fixture), not by looping. It never becomes an open-ended fix loop. Never
121
- skip or disable a failing test; when the fix cannot land now, report it and
122
- leave that decision to the user. Only a capability skip (a platform or tool
123
- the test needs is absent) is allowed, and its reason names that capability.
124
-
125
- ## Prune
126
-
127
- First tell redundant coverage from removing a supported behaviour, and keep
128
- distinct observable success, rejection, error and compatibility cases.
129
- Removing a behaviour with its tests needs evidence independent of those tests
130
- (current requirements, real callers, a specification); a test existing only
131
- for a branch, or a fallback and test justifying each other, is no evidence,
132
- and unresolved intent blocks deletion.
133
-
134
- - Delete a test only when it is dominated (same branch and result as a
135
- stronger test), tautological (cannot fail), a permutation crossing no new
136
- branch, or verification theater (verifier and producer share one failure
137
- domain). Name the test that still owns the contract.
138
- - Deleting a contract test needs the user's approval.
139
- - Same contract owned twice → merge the unique cases into the owner at the
140
- user's interface.
141
- - Asserts that something was removed, restates a constant, copies production
142
- logic or source text into the test, or pins internals no user observes →
143
- delete it, or rewrite it at the interface when a real contract is behind it.
144
- - Passes only alone → Diagnose; never mark it serial to hide it.
145
-
146
- Prune as you go: a change that touches a test file applies these rules to that
147
- file only. A whole-suite sweep is an Audit the user asked for.
148
-
149
- Result: each removal names the owning test that still covers the contract, or
150
- states that the contract is gone.
151
-
152
- ## Audit
153
-
154
- 1. Run `node ${MIXDOG_SKILL_DIR}/scripts/inventory.mjs <project-root>`
155
- (`--json` for data). Result: per-area counts, lanes, languages, risk
156
- markers, and defect skips for JS/TS, Python, Go and Rust.
157
- 2. Read `references/case-management.md` and rank findings by impact: defect
158
- skips, risky markers in the fast lane, duplicate owners.
159
- 3. Report one table (area, finding, count, proposed action) and wait for the
160
- user's choice of batches. Audit ends at this report; fix only the batches
161
- the user picks, one or two at a time, each verified by its covering files.
162
-
163
- ## Pitfalls
164
-
165
- - Parsing localized tool output (a console in another language, a busy host
166
- dropping rows) passed on the author's machine and failed elsewhere. Query a
167
- structured source.
168
- - Fixed sleeps passed alone and failed in bulk when other work ran.
169
- - A load generator or server started by an agent was detached, outlived the
170
- agent and froze the machine. Nothing detached survives a report.
171
- - Render tests passed while the real screen hid a row: assert the rendered
172
- output, not the inputs.
173
- - Full lanes that take many minutes fail under load; selection, not more
174
- retries, fixes that.
175
-
176
- ## Verification
177
-
178
- - Write/fix: the focused file passes; a new regression test also failed
179
- before, and the report quotes both.
180
- - Diagnose: the named cause and its evidence. Only when the user asked for a
181
- fix: the deterministic trigger and its passing run.
182
- - Every mode: no process you started is still running.
183
-
184
74
  ## Resources
185
75
 
186
76
  - `references/reliability.md` — read when a test touches time, processes,
@@ -188,7 +78,5 @@ states that the contract is gone.
188
78
  - `references/flaky-policy.md` — read when a failure is intermittent.
189
79
  - `references/suite-design.md` — read when writing a test, choosing its lane
190
80
  or file, or reviewing how a suite is organized.
191
- - `references/case-management.md` — read for Audit, or when adding a new
192
- area.
193
81
  - `scripts/inventory.mjs` — run for Audit; read-only, bounded, Markdown or
194
82
  `--json`.
@@ -1,8 +1,11 @@
1
1
  # Test case management
2
2
 
3
- Every test has an owner, a single contract and a lane. None of this belongs in a hand-written registry: hand lists
4
- drift from the code. It is derived from where the file lives, its name and its
5
- text, and `scripts/inventory.mjs` reports it.
3
+ Read for Audit, or when adding a new area.
4
+
5
+ Every test has an owner, a single contract and a lane. None of this belongs in
6
+ a hand-written registry: hand lists drift from the code. It is derived from
7
+ where the file lives, its name and its text, and `scripts/inventory.mjs`
8
+ reports it.
6
9
 
7
10
  ## Where a test lives
8
11
 
@@ -43,9 +46,14 @@ needs a fix; an issue number does not make it acceptable.
43
46
  A fake-clock library in the file clears `timer` and `clock`. Markers are
44
47
  leads, not verdicts: a timer in a test about timers is correct.
45
48
 
46
- ## Audit report
47
-
48
- One table per project: area, files, fast/slow counts, marker counts, defect
49
- skips. Rank batches by: defect skips, fast-lane
50
- files with `timer`+`spawn`, then the rest by area. The audit ends at the
51
- report; fix only the batches the user picks.
49
+ ## Audit procedure
50
+
51
+ 1. Run `node ${MIXDOG_SKILL_DIR}/scripts/inventory.mjs <project-root>`
52
+ (`--json` for data). Result: per-area counts, lanes, languages, risk
53
+ markers and defect skips for JS/TS, Python, Go and Rust.
54
+ 2. Rank findings by impact: defect skips, then risky markers in the fast lane
55
+ (files with `timer`+`spawn`), duplicate owners, then the rest by area.
56
+ 3. Report one table per project (area, files, fast/slow counts, marker counts,
57
+ defect skips, proposed action) and wait for the user's choice of batches.
58
+ The audit ends at this report; fix only the batches the user picks, one or
59
+ two at a time, each verified by its covering files.
@@ -0,0 +1,23 @@
1
+ # Diagnose an intermittent failure
2
+
3
+ Read when a test fails sometimes, only in a full run, only in CI or only on
4
+ this machine. An intermittent failure is a defect in the test or the product,
5
+ never noise (`references/flaky-policy.md`).
6
+
7
+ 1. Collect evidence without reruns: the failing assertion, the full log, what
8
+ else was running, the OS and locale. Result: the failing line and the
9
+ condition it waited on.
10
+ 2. Classify the cause with the table in `references/reliability.md`. Result:
11
+ one named cause with the line that assumes it.
12
+ 3. Propose the fix: in the test, or in the product when real users can hit
13
+ the same race. A longer timeout, a retry or a skip is not a fix.
14
+
15
+ Diagnose ends at a report: cause, evidence, proposed fix. Only when the user
16
+ asked for a fix, apply it and prove it with the focused file plus a
17
+ deterministic trigger of the cause (a held promise, a fake clock, a localized
18
+ fixture), not by looping; the deterministic trigger and its passing run go in
19
+ the report. It never becomes an open-ended fix loop. When the fix cannot land
20
+ now, report it and leave the skip decision to the user. Only a capability skip
21
+ (a platform or tool the test needs is absent) is allowed, and its reason
22
+ names that capability. Never mark a test serial to hide an
23
+ order or contention failure.
@@ -0,0 +1,27 @@
1
+ # Prune tests
2
+
3
+ Read to remove duplicate, dead, obsolete or worthless tests, and whenever a
4
+ change edits an existing test file (apply these rules to that file only; a
5
+ whole-suite sweep is an Audit the user asked for).
6
+
7
+ First tell redundant coverage from removing a supported behaviour, and keep
8
+ distinct observable success, rejection, error and compatibility cases.
9
+ Removing a behaviour with its tests needs evidence independent of those tests
10
+ (current requirements, real callers, a specification); a test existing only
11
+ for a branch, or a fallback and test justifying each other, is no evidence,
12
+ and unresolved intent blocks deletion.
13
+
14
+ - Delete a test only when it is dominated (same branch and result as a
15
+ stronger test), tautological (cannot fail), a permutation crossing no new
16
+ branch, or verification theater (verifier and producer share one failure
17
+ domain). Name the test that still owns the contract.
18
+ - Deleting a contract test needs the user's approval.
19
+ - Same contract owned twice → merge the unique cases into the owner at the
20
+ user's interface.
21
+ - Asserts that something was removed, restates a constant, copies production
22
+ logic or source text into the test, or pins internals no user observes →
23
+ delete it, or rewrite it at the interface when a real contract is behind it.
24
+ - Passes only alone → Diagnose; never mark it serial to hide it.
25
+
26
+ Result: each removal names the owning test that still covers the contract, or
27
+ states that the contract is gone.
@@ -0,0 +1,44 @@
1
+ # Write a test
2
+
3
+ Read before adding a test or fixing a bug test-first.
4
+
5
+ ## What a change gets
6
+
7
+ | Change | New tests |
8
+ |---|---|
9
+ | Bug fix | One regression test for the reported behaviour |
10
+ | New behaviour | One per observable behaviour the requirement names (its success and its stated rejections) |
11
+ | Refactor, rename, move | None; the existing tests must still pass unchanged |
12
+ | Removal | None; delete the removed behaviour's tests, never assert the absence |
13
+ | Docs, copy, style, config values | None |
14
+
15
+ Only a test owning a lasting contract joins the suite and its CI lane.
16
+ Reproductions, probes and one-off checks are scratch: they live in a temp
17
+ directory outside the repository, are never committed, and are deleted when
18
+ the task ends. A scratch reproduction that proves a bug becomes the
19
+ regression test above or is deleted, never both kept.
20
+
21
+ ## Steps
22
+
23
+ 1. Name the behaviour and its owner: the one contract this test owns, beside
24
+ the code it exercises. If a test already owns it, extend that test
25
+ (`references/suite-design.md`). Result: one file and test name chosen.
26
+ 2. Name the production change that would make the test fail; if you cannot,
27
+ redesign it around an observable behaviour. Write the test through the
28
+ user's interface (caller, CLI, HTTP, rendered output) with the expected
29
+ value from the requirement, not from running the code. Assert the rendered
30
+ output, not the inputs: render tests passed while the real screen hid a
31
+ row. For a new regression test, show it failing first: the failure names
32
+ the missing behaviour. An existing failing test already counts as that
33
+ evidence.
34
+ 3. Apply `references/reliability.md` for any time, process, port, file,
35
+ environment, locale or platform dependency. Result: no fixed sleep, no
36
+ fixed port, no unrestored global, no host-locale assumption.
37
+ 4. Make it pass with the smallest change, then rerun the same file. Add no
38
+ further tests unless the user asks.
39
+ 5. Put it in the lane its duration and needs call for (`suite-design.md`).
40
+
41
+ ## Done when
42
+
43
+ The focused file passes. A new regression test also failed before, and the
44
+ report quotes both runs.
@@ -12,58 +12,44 @@ dependencies:
12
12
 
13
13
  # Video generation (media tool · Media Studio)
14
14
 
15
- Use `media` to generate short video clips. Read this guide before the first
16
- video-generation call.
17
-
18
- This file owns the judgement around one tool for moving pictures. It is not tied to any deliverable. The lane catalog (providers, models, controls, sign-in state) lives in the runtime and is read with `list`; nothing here names a lane or a model. A still that the clip needs first is the `image` skill's job — load it for that step.
15
+ Use `media` to generate short video clips; read this guide before the first call. The `media` tool schema and `list` own the controls, lane catalog and sign-in state; nothing here names a lane or a model. A still the clip needs first is the `image` skill's job.
19
16
 
20
17
  ## 1. Which route
21
18
  **Hard rule — video only when motion was asked for**: a deliverable never gets a clip by default; a still that carries the meaning is not turned into a video.
19
+
22
20
  | Situation | Route |
23
21
  |---|---|
24
22
  | a scene whose identity, exact composition, or continuity must survive the motion | **image first**: load the `image` skill and generate a still at the clip's aspect with the protected subjects and environment composed in one picture; inspect it; then generate the video with that file as the reference |
25
23
  | several separately supplied subjects that must stay recognizable | references-to-video with each identity as its own reference (index and role, identity first), duration at the short end of the model's range |
26
- | continuing an earlier clip | a new job with the previous clip's last frame as the reference; only the pose carries over — motion, trajectory, and camera do not |
24
+ | continuing an earlier clip | a new job with the previous clip's last frame as the reference (read `references/prompting.md` first); only the pose carries over |
27
25
  | a pure text-to-video request with no subject to protect | generate directly from the prompt |
28
26
 
29
27
  **Hard rule — no signed-in lane, no clip**: when `list` (or a failed `generate`) reports no video lane, say so and stop; never substitute a slideshow, a GIF from stock, or a placeholder.
30
28
  **Default — one generation per job, one corrected retry**: a miss is retried once with a single changed variable and every invariant restated; a second miss is reported, not attempted a third time. Several candidates only when the user asks.
31
29
 
32
30
  ## 2. Call order
33
- 1. `media action:'list' kind:'video'` once per session: signed-in lanes, models, and `remembered` — the lane/model a generate runs on when none is passed (the user's current Studio selection, else the last clip's). Add `model:` for that model's controls (`aspectRatio`, `resolution`, `durations` or `durationRange`, `maxReferences`).
34
- 2. `media action:'generate' kind:'video', prompt, path, aspect, duration, resolution, references, wait:true` — **lane and model are omitted by default** so the clip comes from what the user chose; controls only from the remembered model's list. Pass `lane`/`model` only when the job needs a capability that model lacks (reference count, a duration outside its range), chosen from `list`. Use `wait:false` only when background execution is requested or useful independent work remains.
35
- 3. For a background or resumed job, use `media action:'status' job:<id> path:<file>` at spaced intervals; the call that sees `done` writes the file. A running job is pending, not completion. Do not regenerate it or tight-loop status calls.
36
- 4. **Hard rule — look before using**: frame-sample or play the file; check the opening frame, that the one change happens, the ending frame, and that no text, logo, or extra subject appeared.
37
- 5. The result is also a Studio asset (`assetId`); the file at `path` is the copy the work uses.
38
-
39
- ## 3. Prompt — an active shot, not a list of elements
40
- **Hard rule — the prompt names, in this order**:
41
- 1. *Shot design*: the opening frame, one motivated reveal or change, and a settling final frame.
42
- 2. *Camera intent*: the move that serves the scene — macro push-in for a product, orbit for space, handheld for documentary, crane for scale, static when nothing needs to move. Never a default "slow dolly in".
43
- 3. *Production*: concrete material and texture, a motivated light source, depth layers (foreground / mid / background), lens framing.
44
- 4. *Sound*: music style, no music, room tone, or effects only.
45
- 5. *Dialogue*: the exact line in its original language, or "no dialogue".
46
- 6. *Ending frame*: final pose, camera state, last sound cue — clear enough to be the first frame of a next clip.
47
- Each item is one focused sentence or two, labeled; never keywords, never JSON.
48
- **Hard rule — pacing follows duration**: 1-4 s one action; 5-7 s setup, turn, hold; 8-10 s two connected beats; 11-15 s a three-beat arc. Pick the duration from the model's list to fit the beats, not the other way round.
49
- **Hard rule — no slop words**: "cinematic", "volumetric lighting", "neon glow", "AAA trailer", "shot on RED", and unmotivated dark or moody defaults are rejected; write what the camera actually sees.
50
- **Default — specificity policy**: a detailed user prompt is normalized into the six items and nothing is added; a generic one gets shot design, camera intent, and a concrete environment — never extra characters, props, brand names, or story beats.
51
- **Default — exclusions**: `no text, no logos, no watermarks`; `no faces` only when the clip is a field for copy or the user asked for none.
52
-
53
- ## 4. Controls
54
- - `aspect` from the destination: a feed or story clip `9:16`, an embed or presentation `16:9`, a tile `1:1`; the still generated first uses the same aspect. The value must be one the model lists.
55
- - `duration` from the pacing rule; within the model's list or range.
56
- - `resolution`: the lowest listed for a draft that checks motion and pacing, the highest listed for delivery.
57
- - References beyond one shorten the allowed duration on most models — read the model's controls before promising a length.
58
-
59
- ## 5. Reference images
60
- `references:` takes file paths; the model's `maxReferences` caps them. A reference is the user's own file, a still from the `image` skill, or an earlier clip's frame — never a downloaded third-party image.
61
- **Hard rule — each reference gets an index and a role in the prompt**: `Image 1: the scene to animate. Image 2: subject identity.` followed by the relationship. Identity references (a face, a product, a logo) go first.
62
- ## 6. Saving, reporting, provenance
63
- **Hard rule — never overwrite**: an existing file gets a sibling (`clip-v2.mp4`) unless the user asked for replacement; output goes beside the work that uses it or where the user named.
64
- **Hard rule — the report names what was made**: final path, lane/model (the one that actually ran, when the runtime reports a fallback), duration, resolution, aspect, and the prompt.
65
- **Hard rule — generated media is declared where it lands**: a caption, a note, or the message says the clip is generated, with lane/model and prompt; it is never presented as footage of a real event, place, or person.
66
- ## 7. Failure and cost
67
- - A `generate` error carries `lanes` (what is signed in for video) or `available` (models); choose from it and call again, or stop.
68
- - Video costs far more credit than a still: one call per job, drafts at low resolution, no speculative variants.
69
- - A job still running past the tool's wait is not a failure. Follow the pending-job flow above; `cancel` only when the user moves on.
31
+ 1. `media action:'list' kind:'video'` once per session: signed-in lanes, models, and `remembered` — the lane/model a generate runs on when none is passed. Add `model:` for that model's controls; pass a control only when that model lists it.
32
+ 2. **Hard rule — pick the source in this order**: a source the user names is passed as `lane` (and `model` if named), resolved from `list`; if it is not signed in, say so rather than swap. Otherwise omit `lane`/`model`; the runtime uses the configured default (pinned), else the current Studio selection or last clip's lane/model (remembered), else the signed-in ChatGPT lane if it makes video, else the first signed-in lane, so the clip comes from what the user chose. Choose a lane/model by capability only when the job needs one the selected model lacks (reference count, a duration outside its range).
33
+ 3. Before writing the prompt, read `references/prompting.md`.
34
+ 4. `media action:'generate' kind:'video', prompt, path, aspect, duration, resolution, references, wait:true`. Use `wait:false` only when background execution is requested or useful independent work remains. The result reports the lane, model, and `laneSource` (`requested` = a lane explicitly selected in the call (a model alone is not), `pinned` = the user's configured default for video, `remembered` = their last choice, `first` = the default lane or the first available signed-in lane) that actually ran.
35
+ 5. A background or resumed job: `media action:'status' job:<id> path:<file>` at spaced intervals; the call that sees `done` writes the file. A running job is pending, not completion and not a failure: do not regenerate it or tight-loop status calls; `cancel` only when the user moves on.
36
+ 6. **Hard rule — look before using**: frame-sample or play the file; check the opening frame, that the one change happens, the ending frame, and that no text, logo, or extra subject appeared.
37
+ 7. The result is also a Studio asset (`assetId`); the file at `path` is the copy the work uses.
38
+
39
+ ## 3. Prompt and controls
40
+ The prompt is an active shot, not a list of elements: labeled items, never keywords or JSON. A detailed user prompt is normalized and nothing is added; a generic one gets only shot design, camera intent, and a concrete environment — never extra characters, props, brands, or story beats. Exclusions: `no text, no logos, no watermarks`; `no faces` only for a field for copy or on request.
41
+ - `aspect` from the destination (a feed or story clip `9:16`, an embed or presentation `16:9`, a tile `1:1`), a listed value; the still generated first uses the same aspect.
42
+ - `duration` fits the beats and the model's list or range. References beyond one shorten the allowed duration on most models — read the model's controls before promising a length.
43
+ - `resolution`: lowest listed for a draft that checks motion and pacing, highest for delivery.
44
+ - `references:` take file paths (cap: `maxReferences`): the user's own file, a still from the `image` skill, or an earlier clip's frame — never a downloaded third-party image. Each gets an index and a role in the prompt; identity references (face, product, logo) go first.
45
+
46
+ ## 4. Saving, reporting, provenance
47
+ - **Never overwrite**: an existing file gets a sibling (`clip-v2.mp4`) unless replacement was asked; output goes beside the work that uses it or where the user named.
48
+ - **Report** the final path, the lane/model that actually ran, duration, resolution, aspect, and the prompt.
49
+ - **Declare generated media where it lands** (caption, note, or message) with lane/model and prompt; never present it as footage of a real event, place, or person.
50
+
51
+ ## 5. Failure and cost
52
+ A `generate` error carries `lanes` (signed in for video) or `available` (models): choose from it and call again, or stop. Video costs far more credit than a still: one call per job, drafts at low resolution, no speculative variants.
53
+
54
+ ## Resources
55
+ `references/prompting.md` — read before constructing a video prompt, and when extending a clip.
@@ -0,0 +1,31 @@
1
+ # Video prompting
2
+
3
+ ## Before prompt construction
4
+
5
+ Name these, in this order, each as one focused labeled sentence or two:
6
+
7
+ 1. *Shot design*: the opening frame, one motivated reveal or change, and a settling final frame.
8
+ 2. *Camera intent*: the move that serves the scene — macro push-in for a product, orbit for space, handheld for documentary, crane for scale, static when nothing needs to move. Never a default "slow dolly in".
9
+ 3. *Production*: concrete material and texture, a motivated light source, depth layers (foreground / mid / background), lens framing.
10
+ 4. *Sound*: music style, no music, room tone, or effects only.
11
+ 5. *Dialogue*: the exact line in its original language, or "no dialogue".
12
+ 6. *Ending frame*: final pose, camera state, last sound cue — clear enough to be the first frame of a next clip.
13
+
14
+ Pacing follows duration:
15
+
16
+ | Duration | Beats |
17
+ |---|---|
18
+ | 1-4 s | one action |
19
+ | 5-7 s | setup, turn, hold |
20
+ | 8-10 s | two connected beats |
21
+ | 11-15 s | a three-beat arc |
22
+
23
+ Pick the duration from the model's list to fit the beats.
24
+
25
+ Rejected slop words: "cinematic", "volumetric lighting", "neon glow", "AAA trailer", "shot on RED", and unmotivated dark or moody defaults. Write what the camera actually sees.
26
+
27
+ Normalize a detailed user prompt into these items without adding to it. Reference images are named in the prompt by index and role, e.g. `Image 1: the scene to animate. Image 2: subject identity.`, followed by the relationship.
28
+
29
+ ## When extending a clip
30
+
31
+ Start a new job; do not stretch the old one. Use the previous clip's last frame as the reference (index and role in the prompt, e.g. `Image 1: the opening frame, continuing the previous clip`). Only the pose carries over: motion, trajectory, and camera do not, so restate the camera intent and the motion from that pose. The previous clip's ending frame item is the opening frame of this one; write the new ending frame so it can seed a further clip.