tool-subagent.test.ts 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301
  1. import { describe, expect } from "bun:test"
  2. import { DateTime, Effect, Fiber, Layer, Schema, Stream } from "effect"
  3. import { Money } from "@opencode-ai/schema/money"
  4. import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
  5. import { LayerNode } from "@opencode-ai/core/effect/layer-node"
  6. import { makeGlobalNode } from "@opencode-ai/core/effect/app-node"
  7. import { Database } from "@opencode-ai/core/database/database"
  8. import { EventV2 } from "@opencode-ai/core/event"
  9. import { Location } from "@opencode-ai/core/location"
  10. import { ModelV2 } from "@opencode-ai/core/model"
  11. import { ProviderV2 } from "@opencode-ai/core/provider"
  12. import { AbsolutePath } from "@opencode-ai/core/schema"
  13. import { AgentV2 } from "@opencode-ai/core/agent"
  14. import { Job } from "@opencode-ai/core/job"
  15. import { LocationServiceMap } from "@opencode-ai/core/location-service-map"
  16. import { SessionV2 } from "@opencode-ai/core/session"
  17. import { SessionEvent } from "@opencode-ai/core/session/event"
  18. import { SessionExecution } from "@opencode-ai/core/session/execution"
  19. import { SessionPending } from "@opencode-ai/core/session/pending"
  20. import { SessionMessage } from "@opencode-ai/core/session/message"
  21. import { SessionRunnerModel } from "@opencode-ai/core/session/runner/model"
  22. import { SessionStore } from "@opencode-ai/core/session/store"
  23. import { PluginRuntime } from "@opencode-ai/core/plugin/runtime"
  24. import { PluginSupervisor } from "@opencode-ai/core/plugin/supervisor"
  25. import { SubagentTool } from "@opencode-ai/core/tool/subagent"
  26. import { ToolRegistry } from "@opencode-ai/core/tool/registry"
  27. import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
  28. import { tmpdir } from "./fixture/tmpdir"
  29. import { testEffect } from "./lib/effect"
  30. import { executeTool, settleTool, toolIdentity, waitForTool } from "./lib/tool"
  31. const childText = "child final response"
  32. const childModel = ModelV2.Ref.make({ id: ModelV2.ID.make("child"), providerID: ProviderV2.ID.make("test") })
  33. const parentModel = ModelV2.Ref.make({ id: ModelV2.ID.make("parent"), providerID: ProviderV2.ID.make("test") })
  34. const tokens = { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } }
  35. const outputSessionID = (value: unknown) => Schema.decodeUnknownSync(SubagentTool.Output)(value).sessionID
  36. const executionNode = makeGlobalNode({
  37. service: SessionExecution.Service,
  38. layer: Layer.effect(
  39. SessionExecution.Service,
  40. Effect.gen(function* () {
  41. const events = yield* EventV2.Service
  42. const store = yield* SessionStore.Service
  43. const completed = new Set<SessionV2.ID>()
  44. const complete = Effect.fn("SubagentTest.complete")(function* (sessionID: SessionV2.ID) {
  45. if (completed.has(sessionID)) return
  46. if ((yield* store.get(sessionID))?.title.includes("fail")) {
  47. yield* new SessionRunnerModel.ModelNotSelectedError({ sessionID })
  48. return
  49. }
  50. completed.add(sessionID)
  51. const assistantMessageID = SessionMessage.ID.create()
  52. yield* events.publish(SessionEvent.Step.Started, {
  53. sessionID,
  54. assistantMessageID,
  55. agent: AgentV2.ID.make("reviewer"),
  56. model: childModel,
  57. })
  58. yield* events.publish(SessionEvent.Text.Started, {
  59. sessionID,
  60. assistantMessageID,
  61. ordinal: 0,
  62. })
  63. yield* events.publish(SessionEvent.Text.Ended, {
  64. sessionID,
  65. assistantMessageID,
  66. ordinal: 0,
  67. text: childText,
  68. })
  69. yield* events.publish(SessionEvent.Step.Ended, {
  70. sessionID,
  71. assistantMessageID,
  72. finish: "stop",
  73. cost: Money.USD.zero,
  74. tokens,
  75. })
  76. })
  77. return SessionExecution.Service.of({
  78. active: Effect.succeed(new Set()),
  79. resume: complete,
  80. wake: () => Effect.void,
  81. interrupt: () => Effect.void,
  82. awaitIdle: (sessionID) => complete(sessionID).pipe(Effect.exit, Effect.asVoid),
  83. })
  84. }),
  85. ),
  86. deps: [EventV2.node, SessionStore.node],
  87. })
  88. const layer = AppNodeBuilder.build(
  89. LayerNode.group([
  90. Database.node,
  91. EventV2.node,
  92. Job.node,
  93. ToolOutputStore.cleanupNode,
  94. SessionV2.node,
  95. SessionExecution.node,
  96. PluginRuntime.providerNode,
  97. LocationServiceMap.node,
  98. ]),
  99. [[SessionExecution.node, executionNode]],
  100. )
  101. const it = testEffect(layer)
  102. const withSubagent = (location: Location.Ref) =>
  103. Effect.gen(function* () {
  104. const locations = yield* LocationServiceMap.Service
  105. yield* PluginSupervisor.Service.use((supervisor) => supervisor.flush).pipe(Effect.provide(locations.get(location)))
  106. yield* AgentV2.Service.use((agents) =>
  107. agents.transform((draft) => {
  108. // The caller identity used by executeTool; subagent permission asserts against it.
  109. draft.update(toolIdentity.agent, (agent) => {
  110. agent.mode = "primary"
  111. agent.permissions.push({ action: "*", resource: "*", effect: "allow" })
  112. })
  113. draft.update(AgentV2.ID.make("reviewer"), (agent) => {
  114. agent.mode = "subagent"
  115. agent.model = childModel
  116. })
  117. draft.update(AgentV2.ID.make("fallback"), (agent) => {
  118. agent.mode = "subagent"
  119. })
  120. draft.update(AgentV2.ID.make("primary"), (agent) => {
  121. agent.mode = "primary"
  122. })
  123. }),
  124. ).pipe(Effect.provide(locations.get(location)))
  125. })
  126. describe("SubagentTool", () => {
  127. it.live("registers globally while resolving agents from the caller location", () =>
  128. Effect.acquireRelease(
  129. Effect.promise(() => tmpdir()),
  130. (dir) => Effect.promise(() => dir[Symbol.asyncDispose]()),
  131. ).pipe(
  132. Effect.flatMap((dir) =>
  133. Effect.gen(function* () {
  134. const location = Location.Ref.make({ directory: AbsolutePath.make(dir.path) })
  135. const session = yield* SessionV2.Service
  136. const parent = yield* session.create({ location })
  137. yield* withSubagent(parent.location)
  138. const locations = yield* LocationServiceMap.Service
  139. const registry = yield* ToolRegistry.Service.pipe(Effect.provide(locations.get(parent.location)))
  140. yield* waitForTool(registry, SubagentTool.name)
  141. expect((yield* registry.materialize()).definitions.map((tool) => tool.name)).toContain(SubagentTool.name)
  142. expect(
  143. yield* executeTool(registry, {
  144. sessionID: parent.id,
  145. ...toolIdentity,
  146. call: {
  147. type: "tool-call",
  148. id: "call-primary",
  149. name: SubagentTool.name,
  150. input: { agent: "primary", description: "primary", prompt: "should fail" },
  151. },
  152. }),
  153. ).toEqual({ type: "error", value: "Agent primary cannot run as a subagent" })
  154. }),
  155. ),
  156. ),
  157. )
  158. it.live("runs a foreground child session and returns the final assistant text", () =>
  159. Effect.acquireRelease(
  160. Effect.promise(() => tmpdir()),
  161. (dir) => Effect.promise(() => dir[Symbol.asyncDispose]()),
  162. ).pipe(
  163. Effect.flatMap((dir) =>
  164. Effect.gen(function* () {
  165. const location = Location.Ref.make({ directory: AbsolutePath.make(dir.path) })
  166. const sessions = yield* SessionV2.Service
  167. const parent = yield* sessions.create({ location, model: parentModel })
  168. yield* withSubagent(parent.location)
  169. const locations = yield* LocationServiceMap.Service
  170. const registry = yield* ToolRegistry.Service.pipe(Effect.provide(locations.get(parent.location)))
  171. yield* waitForTool(registry, SubagentTool.name)
  172. const settled = yield* settleTool(registry, {
  173. sessionID: parent.id,
  174. ...toolIdentity,
  175. call: {
  176. type: "tool-call",
  177. id: "call-subagent",
  178. name: SubagentTool.name,
  179. input: { agent: "reviewer", description: "review", prompt: "review this" },
  180. },
  181. })
  182. expect(settled.output?.structured).toMatchObject({ status: "completed", output: childText })
  183. const child = yield* sessions.get(outputSessionID(settled.output?.structured))
  184. expect(child).toMatchObject({
  185. parentID: parent.id,
  186. location: parent.location,
  187. agent: "reviewer",
  188. model: childModel,
  189. })
  190. const fallback = yield* settleTool(registry, {
  191. sessionID: parent.id,
  192. ...toolIdentity,
  193. call: {
  194. type: "tool-call",
  195. id: "call-subagent-fallback",
  196. name: SubagentTool.name,
  197. input: { agent: "fallback", description: "fallback", prompt: "fallback" },
  198. },
  199. })
  200. const fallbackChild = yield* sessions.get(outputSessionID(fallback.output?.structured))
  201. expect(fallbackChild).toMatchObject({ parentID: parent.id, model: parentModel })
  202. }),
  203. ),
  204. ),
  205. )
  206. it.live("returns child runner failures as tool errors", () =>
  207. Effect.acquireRelease(
  208. Effect.promise(() => tmpdir()),
  209. (dir) => Effect.promise(() => dir[Symbol.asyncDispose]()),
  210. ).pipe(
  211. Effect.flatMap((dir) =>
  212. Effect.gen(function* () {
  213. const location = Location.Ref.make({ directory: AbsolutePath.make(dir.path) })
  214. const sessions = yield* SessionV2.Service
  215. const parent = yield* sessions.create({ location })
  216. yield* withSubagent(parent.location)
  217. const locations = yield* LocationServiceMap.Service
  218. const registry = yield* ToolRegistry.Service.pipe(Effect.provide(locations.get(parent.location)))
  219. yield* waitForTool(registry, SubagentTool.name)
  220. expect(
  221. yield* executeTool(registry, {
  222. sessionID: parent.id,
  223. ...toolIdentity,
  224. call: {
  225. type: "tool-call",
  226. id: "call-subagent-failure",
  227. name: SubagentTool.name,
  228. input: { agent: "reviewer", description: "fail review", prompt: "please fail" },
  229. },
  230. }),
  231. ).toEqual({ type: "error", value: expect.stringContaining("No model is available for session") })
  232. }),
  233. ),
  234. ),
  235. )
  236. it.live("notifies once when background work completes", () =>
  237. Effect.acquireRelease(
  238. Effect.promise(() => tmpdir()),
  239. (dir) => Effect.promise(() => dir[Symbol.asyncDispose]()),
  240. ).pipe(
  241. Effect.flatMap((dir) =>
  242. Effect.gen(function* () {
  243. const location = Location.Ref.make({ directory: AbsolutePath.make(dir.path) })
  244. const sessions = yield* SessionV2.Service
  245. const parent = yield* sessions.create({ location })
  246. yield* withSubagent(parent.location)
  247. const locations = yield* LocationServiceMap.Service
  248. const registry = yield* ToolRegistry.Service.pipe(Effect.provide(locations.get(parent.location)))
  249. yield* waitForTool(registry, SubagentTool.name)
  250. const events = yield* EventV2.Service
  251. const admitted = yield* events.subscribe(SessionEvent.InputAdmitted).pipe(
  252. Stream.filter((event) => event.data.sessionID === parent.id && event.data.input.type === "synthetic"),
  253. Stream.take(1),
  254. Stream.runCollect,
  255. Effect.forkScoped({ startImmediately: true }),
  256. )
  257. const settled = yield* settleTool(registry, {
  258. sessionID: parent.id,
  259. ...toolIdentity,
  260. call: {
  261. type: "tool-call",
  262. id: "call-background-subagent",
  263. name: SubagentTool.name,
  264. input: { agent: "reviewer", description: "background review", prompt: "review", background: true },
  265. },
  266. })
  267. const childID = outputSessionID(settled.output?.structured)
  268. expect(settled.output?.structured).toMatchObject({
  269. status: "running",
  270. output: expect.stringContaining(`id: ${childID}`),
  271. })
  272. const admission = Array.from(yield* Fiber.join(admitted))[0]
  273. expect(admission?.data.input.data.text).toContain(`<subagent id="${childID}" state="completed"`)
  274. const database = yield* Database.Service
  275. yield* SessionPending.promoteSteers(database.db, events, parent.id)
  276. const synthetic = (yield* sessions.context(parent.id)).filter((message) => message.type === "synthetic")
  277. expect(synthetic).toHaveLength(1)
  278. expect(synthetic[0]?.text).toContain(`<subagent id="${childID}" state="completed"`)
  279. expect(synthetic[0]?.text).toContain(childText)
  280. }),
  281. ),
  282. ),
  283. )
  284. })