saeeol 1.2.9 → 1.3.1

package/test/tool/task.test.ts

@@ -0,0 +1,612 @@
+import { afterEach, describe, expect } from "bun:test"
+import { Effect, Layer } from "effect"
+import { Agent } from "../../src/agent/agent"
+import { Config } from "@/config/config"
+import { CrossSpawnSpawner } from "@saeeol/core/cross-spawn-spawner"
+import { Instance } from "../../src/project/instance"
+import { Session } from "@/session/session"
+import { MessageV2 } from "../../src/session/message-v2"
+import type { SessionPrompt } from "../../src/session/prompt"
+import { MessageID, PartID, SessionID } from "../../src/session/schema" 
+import { ModelID, ProviderID } from "../../src/provider/schema"
+import { TaskTool, type TaskPromptOps } from "../../src/tool/task"
+import { Truncate } from "@/tool/truncate"
+import { ToolRegistry } from "@/tool/registry"
+import { disposeAllInstances, provideTmpdirInstance } from "../fixture/fixture"
+import { testEffect } from "../lib/effect"
+
+afterEach(async () => {
+  await disposeAllInstances()
+})
+
+const ref = {
+  providerID: ProviderID.make("test"),
+  modelID: ModelID.make("test-model"),
+}
+
+const it = testEffect(
+  Layer.mergeAll(
+    Agent.defaultLayer,
+    Config.defaultLayer,
+    CrossSpawnSpawner.defaultLayer,
+    Session.defaultLayer,
+    Truncate.defaultLayer,
+    ToolRegistry.defaultLayer,
+  ),
+)
+
+const seed = Effect.fn("TaskToolTest.seed")(function* (title = "Pinned") {
+  const session = yield* Session.Service
+  const chat = yield* session.create({ title })
+  const user = yield* session.updateMessage({
+    id: MessageID.ascending(),
+    role: "user",
+    sessionID: chat.id,
+    agent: "build",
+    model: ref,
+    time: { created: Date.now() },
+  })
+  const assistant: MessageV2.Assistant = {
+    id: MessageID.ascending(),
+    role: "assistant",
+    parentID: user.id,
+    sessionID: chat.id,
+    mode: "build",
+    agent: "build",
+    cost: 0,
+    path: { cwd: "/tmp", root: "/tmp" },
+    tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+    modelID: ref.modelID,
+    providerID: ref.providerID,
+    time: { created: Date.now() },
+  }
+  yield* session.updateMessage(assistant)
+  return { chat, assistant }
+})
+function stubOps(opts?: {
+  onPrompt?: (input: SessionPrompt.PromptInput) => void
+  text?: string
+  sessions?: Session.Interface
+  childCost?: number
+}): TaskPromptOps {
+  return {
+    cancel() {},
+    resolvePromptParts: (template) => Effect.succeed([{ type: "text" as const, text: template }]),
+    prompt: (input) =>
+      Effect.gen(function* () {
+        opts?.onPrompt?.(input)
+        const rep = reply(input, opts?.text ?? "done")
+        if (opts?.sessions && opts?.childCost != null) {
+          yield* opts.sessions.updateMessage({ ...rep.info, cost: opts.childCost })
+        }
+        return rep
+      }),
+  }
+}
+
+function reply(input: SessionPrompt.PromptInput, text: string): MessageV2.WithParts {
+  const id = MessageID.ascending()
+  return {
+    info: {
+      id,
+      role: "assistant",
+      parentID: input.messageID ?? MessageID.ascending(),
+      sessionID: input.sessionID,
+      mode: input.agent ?? "general",
+      agent: input.agent ?? "general",
+      cost: 0,
+      path: { cwd: "/tmp", root: "/tmp" },
+      tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+      modelID: input.model?.modelID ?? ref.modelID,
+      providerID: input.model?.providerID ?? ref.providerID,
+      time: { created: Date.now() },
+      finish: "stop",
+    },
+    parts: [
+      {
+        id: PartID.ascending(),
+        messageID: id,
+        sessionID: input.sessionID,
+        type: "text",
+        text,
+      },
+    ],
+  }
+}
+
+describe("tool.task", () => {
+  it.live("description sorts subagents by name and is stable across calls", () =>
+    provideTmpdirInstance(
+      () =>
+        Effect.gen(function* () {
+          const agent = yield* Agent.Service
+          const build = yield* agent.get("build")
+          const registry = yield* ToolRegistry.Service
+          const get = Effect.fnUntraced(function* () {
+            const tools = yield* registry.tools({ ...ref, agent: build })
+            return tools.find((tool) => tool.id === TaskTool.id)?.description ?? ""
+          })
+          const first = yield* get()
+          const second = yield* get()
+
+          expect(first).toBe(second)
+
+          const alpha = first.indexOf("- alpha: Alpha agent")
+          const explore = first.indexOf("- explore:")
+          const general = first.indexOf("- general:")
+          const zebra = first.indexOf("- zebra: Zebra agent")
+
+          expect(alpha).toBeGreaterThan(-1)
+          expect(explore).toBeGreaterThan(alpha)
+          expect(general).toBeGreaterThan(explore)
+          expect(zebra).toBeGreaterThan(general)
+        }),
+      {
+        config: {
+          agent: {
+            zebra: {
+              description: "Zebra agent",
+              mode: "subagent",
+            },
+            alpha: {
+              description: "Alpha agent",
+              mode: "subagent",
+            },
+          },
+        },
+      },
+    ),
+  )
+
+  it.live("description hides denied subagents for the caller", () =>
+    provideTmpdirInstance(
+      () =>
+        Effect.gen(function* () {
+          const agent = yield* Agent.Service
+          const build = yield* agent.get("build")
+          const registry = yield* ToolRegistry.Service
+          const description =
+            (yield* registry.tools({ ...ref, agent: build })).find((tool) => tool.id === TaskTool.id)?.description ?? ""
+
+          expect(description).toContain("- alpha: Alpha agent")
+          expect(description).not.toContain("- zebra: Zebra agent")
+        }),
+      {
+        config: {
+          permission: {
+            task: {
+              "*": "allow",
+              zebra: "deny",
+            },
+          },
+          agent: {
+            zebra: {
+              description: "Zebra agent",
+              mode: "subagent",
+            },
+            alpha: {
+              description: "Alpha agent",
+              mode: "subagent",
+            },
+          },
+        },
+      },
+    ),
+  )
+
+  it.live("execute resumes an existing task session from task_id", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const sessions = yield* Session.Service
+        const { chat, assistant } = yield* seed()
+        const child = yield* sessions.create({ parentID: chat.id, title: "Existing child" })
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        let seen: SessionPrompt.PromptInput | undefined
+        const promptOps = stubOps({ text: "resumed", onPrompt: (input) => (seen = input) })
+
+        const result = yield* def.execute(
+          {
+            description: "inspect bug",
+            prompt: "look into the cache key path",
+            subagent_type: "general",
+            task_id: child.id,
+          },
+          {
+            sessionID: chat.id,
+            messageID: assistant.id,
+            agent: "build",
+            abort: new AbortController().signal,
+            extra: { promptOps },
+            messages: [],
+            metadata: () => Effect.void,
+            ask: () => Effect.void,
+          },
+        )
+
+        const kids = yield* sessions.children(chat.id)
+        expect(kids).toHaveLength(1)
+        expect(kids[0]?.id).toBe(child.id)
+        expect(result.metadata.sessionId).toBe(child.id)
+        expect(result.output).toContain(`task_id: ${child.id}`)
+        expect(seen?.sessionID).toBe(child.id)
+      }),
+    ),
+  )
+
+  it.live("execute asks by default and skips checks when bypassed", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const { chat, assistant } = yield* seed()
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        const calls: unknown[] = []
+        const promptOps = stubOps()
+
+        const exec = (extra?: Record<string, any>) =>
+          def.execute(
+            {
+              description: "inspect bug",
+              prompt: "look into the cache key path",
+              subagent_type: "general",
+            },
+            {
+              sessionID: chat.id,
+              messageID: assistant.id,
+              agent: "build",
+              abort: new AbortController().signal,
+              extra: { promptOps, ...extra },
+              messages: [],
+              metadata: () => Effect.void,
+              ask: (input) =>
+                Effect.sync(() => {
+                  calls.push(input)
+                }),
+            },
+          )
+
+        yield* exec()
+        yield* exec({ bypassAgentCheck: true })
+
+        expect(calls).toHaveLength(1)
+        expect(calls[0]).toEqual({
+          permission: "task",
+          patterns: ["general"],
+          always: ["*"],
+          metadata: {
+            description: "inspect bug",
+            subagent_type: "general",
+          },
+        })
+      }),
+    ),
+  )
+
+  it.live("execute creates a child when task_id does not exist", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const sessions = yield* Session.Service
+        const { chat, assistant } = yield* seed()
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        let seen: SessionPrompt.PromptInput | undefined
+        const promptOps = stubOps({ text: "created", onPrompt: (input) => (seen = input) })
+
+        const result = yield* def.execute(
+          {
+            description: "inspect bug",
+            prompt: "look into the cache key path",
+            subagent_type: "general",
+            task_id: "ses_missing",
+          },
+          {
+            sessionID: chat.id,
+            messageID: assistant.id,
+            agent: "build",
+            abort: new AbortController().signal,
+            extra: { promptOps },
+            messages: [],
+            metadata: () => Effect.void,
+            ask: () => Effect.void,
+          },
+        )
+
+        const kids = yield* sessions.children(chat.id)
+        expect(kids).toHaveLength(1)
+        expect(kids[0]?.id).toBe(result.metadata.sessionId)
+        expect(result.metadata.sessionId).not.toBe("ses_missing")
+        expect(result.output).toContain(`task_id: ${result.metadata.sessionId}`)
+        expect(seen?.sessionID).toBe(result.metadata.sessionId)
+      }),
+    ),
+  )
+
+  it.live("execute shapes child permissions for task, todowrite, and primary tools", () =>
+    provideTmpdirInstance(
+      () =>
+        Effect.gen(function* () {
+          const sessions = yield* Session.Service
+          const { chat, assistant } = yield* seed()
+          const tool = yield* TaskTool
+          const def = yield* tool.init()
+          let seen: SessionPrompt.PromptInput | undefined
+          const promptOps = stubOps({ onPrompt: (input) => (seen = input) })
+
+          const result = yield* def.execute(
+            {
+              description: "inspect bug",
+              prompt: "look into the cache key path",
+              subagent_type: "reviewer",
+            },
+            {
+              sessionID: chat.id,
+              messageID: assistant.id,
+              agent: "build",
+              abort: new AbortController().signal,
+              extra: { promptOps },
+              messages: [],
+              metadata: () => Effect.void,
+              ask: () => Effect.void,
+            },
+          )
+
+          const child = yield* sessions.get(result.metadata.sessionId)
+          expect(child.parentID).toBe(chat.id)
+          expect(child.permission).toEqual(
+            expect.arrayContaining([
+              {
+                permission: "todowrite",
+                pattern: "*",
+                action: "deny",
+              },
+              {
+                permission: "bash",
+                pattern: "*",
+                action: "allow",
+              },
+              {
+                permission: "read",
+                pattern: "*",
+                action: "allow",
+              },
+              {
+                permission: "task",
+                pattern: "*",
+                action: "deny",
+              },
+            ]),
+          )
+          expect(seen?.tools).toEqual({
+            todowrite: false,
+            task: false, 
+            bash: false,
+            read: false,
+          })
+        }),
+      {
+        config: {
+          agent: {
+            reviewer: {
+              mode: "subagent",
+              permission: {
+                task: "allow",
+              },
+            },
+          },
+          experimental: {
+            primary_tools: ["bash", "read"],
+            openTelemetry: true, 
+          },
+        },
+      },
+    ),
+  )
+})
+const assistantCost = Effect.fn("TaskToolTest.assistantCost")(function* (sessionID: string) {
+  const sessions = yield* Session.Service
+  const msgs = yield* sessions.messages({ sessionID: SessionID.make(sessionID) })
+  return msgs.reduce((sum, m) => sum + (m.info.role === "assistant" ? m.info.cost : 0), 0)
+})
+
+describe("tool.task cost propagation", () => {
+  it.live("propagates subagent cost to parent assistant message", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const sessions = yield* Session.Service
+        const { chat, assistant } = yield* seed()
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        const promptOps = stubOps({ sessions, childCost: 0.25 })
+
+        yield* def.execute(
+          {
+            description: "inspect bug",
+            prompt: "look into the cache key path",
+            subagent_type: "general",
+          },
+          {
+            sessionID: chat.id,
+            messageID: assistant.id,
+            agent: "build",
+            abort: new AbortController().signal,
+            extra: { promptOps },
+            messages: [],
+            metadata: () => Effect.void,
+            ask: () => Effect.void,
+          },
+        )
+        const parent = yield* Effect.sync(() => MessageV2.get({ sessionID: chat.id, messageID: assistant.id }))
+        expect(parent.info.role).toBe("assistant")
+        if (parent.info.role !== "assistant") return
+        expect(parent.info.cost).toBeCloseTo(0.25, 6)
+      }),
+    ),
+  )
+
+  it.live("propagates recursively through nested subagent costs", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const sessions = yield* Session.Service
+        const { chat, assistant } = yield* seed()
+        // Pre-create a child with its own assistant already bearing a grandchild cost.
+        const child = yield* sessions.create({ parentID: chat.id, title: "grandchild-accumulated" })
+        yield* sessions.updateMessage({
+          id: MessageID.ascending(),
+          role: "assistant",
+          parentID: assistant.id,
+          sessionID: child.id,
+          mode: "build",
+          agent: "general",
+          cost: 0.4,
+          path: { cwd: "/tmp", root: "/tmp" },
+          tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+          modelID: ref.modelID,
+          providerID: ref.providerID,
+          time: { created: Date.now() },
+        })
+
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        // Resuming into the same child via task_id and the stub tacks on another 0.15.
+        const promptOps = stubOps({ sessions, childCost: 0.15 })
+
+        yield* def.execute(
+          {
+            description: "inspect bug",
+            prompt: "look into the cache key path",
+            subagent_type: "general",
+            task_id: child.id,
+          },
+          {
+            sessionID: chat.id,
+            messageID: assistant.id,
+            agent: "build",
+            abort: new AbortController().signal,
+            extra: { promptOps },
+            messages: [],
+            metadata: () => Effect.void,
+            ask: () => Effect.void,
+          },
+        )
+        const parent = yield* Effect.sync(() => MessageV2.get({ sessionID: chat.id, messageID: assistant.id }))
+        if (parent.info.role !== "assistant") return
+        // Only the delta since the start of this invocation propagates.
+        expect(parent.info.cost).toBeCloseTo(0.15, 6)
+        // Child session keeps the full cumulative total (0.4 pre-existing + 0.15 this run).
+        expect(yield* assistantCost(child.id)).toBeCloseTo(0.55, 6)
+      }),
+    ),
+  )
+
+  it.live("resumed task_id only propagates the delta", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const sessions = yield* Session.Service
+        const { chat, assistant } = yield* seed()
+        const child = yield* sessions.create({ parentID: chat.id, title: "resume target" })
+        yield* sessions.updateMessage({
+          id: MessageID.ascending(),
+          role: "assistant",
+          parentID: assistant.id,
+          sessionID: child.id,
+          mode: "build",
+          agent: "general",
+          cost: 0.1,
+          path: { cwd: "/tmp", root: "/tmp" },
+          tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+          modelID: ref.modelID,
+          providerID: ref.providerID,
+          time: { created: Date.now() },
+        })
+
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        const promptOps = stubOps({ sessions, childCost: 0.05 })
+
+        yield* def.execute(
+          {
+            description: "inspect bug",
+            prompt: "continue investigation",
+            subagent_type: "general",
+            task_id: child.id,
+          },
+          {
+            sessionID: chat.id,
+            messageID: assistant.id,
+            agent: "build",
+            abort: new AbortController().signal,
+            extra: { promptOps },
+            messages: [],
+            metadata: () => Effect.void,
+            ask: () => Effect.void,
+          },
+        )
+        const parent = yield* Effect.sync(() => MessageV2.get({ sessionID: chat.id, messageID: assistant.id }))
+        if (parent.info.role !== "assistant") return
+        // Delta-only: only the 0.05 from this run, not 0.15 including the pre-existing 0.10.
+        expect(parent.info.cost).toBeCloseTo(0.05, 6)
+      }),
+    ),
+  )
+
+  it.live("propagates partial cost on abort", () =>
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const sessions = yield* Session.Service
+        const { chat, assistant } = yield* seed()
+        const tool = yield* TaskTool
+        const def = yield* tool.init()
+        const abort = new AbortController()
+        // Stub that persists a partial cost, then aborts — mimics interrupted run after tokens billed.
+        const ops: TaskPromptOps = {
+          cancel() {},
+          resolvePromptParts: (template) => Effect.succeed([{ type: "text" as const, text: template }]),
+          prompt: (input) =>
+            Effect.gen(function* () {
+              const info: MessageV2.Assistant = {
+                id: MessageID.ascending(),
+                role: "assistant",
+                parentID: input.messageID ?? MessageID.ascending(),
+                sessionID: input.sessionID,
+                mode: "general",
+                agent: "general",
+                cost: 0.07,
+                path: { cwd: "/tmp", root: "/tmp" },
+                tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+                modelID: ref.modelID,
+                providerID: ref.providerID,
+                time: { created: Date.now() },
+              }
+              yield* sessions.updateMessage(info)
+              abort.abort()
+              return yield* Effect.interrupt
+            }),
+        }
+
+        yield* def
+          .execute(
+            {
+              description: "partial",
+              prompt: "will abort",
+              subagent_type: "general",
+            },
+            {
+              sessionID: chat.id,
+              messageID: assistant.id,
+              agent: "build",
+              abort: abort.signal,
+              extra: { promptOps: ops },
+              messages: [],
+              metadata: () => Effect.void,
+              ask: () => Effect.void,
+            },
+          )
+          .pipe(Effect.exit)
+
+        const parent = yield* Effect.sync(() => MessageV2.get({ sessionID: chat.id, messageID: assistant.id }))
+        if (parent.info.role !== "assistant") return
+        expect(parent.info.cost).toBeCloseTo(0.07, 6)
+      }),
+    ),
+  )
+})

package/test/tool/tool-define.test.ts

@@ -0,0 +1,99 @@
+import { describe, test, expect } from "bun:test"
+import { Effect, Layer, ManagedRuntime, Schema } from "effect"
+import { Agent } from "../../src/agent/agent"
+import { MessageID, SessionID } from "../../src/session/schema"
+import { Tool } from "@/tool/tool"
+import { Truncate } from "@/tool/truncate"
+
+const runtime = ManagedRuntime.make(Layer.mergeAll(Truncate.defaultLayer, Agent.defaultLayer))
+
+const params = Schema.Struct({ input: Schema.String })
+
+function makeTool(id: string, executeFn?: () => void) {
+  return {
+    description: "test tool",
+    parameters: params,
+    execute() {
+      executeFn?.()
+      return Effect.succeed({ title: "test", output: "ok", metadata: {} })
+    },
+  }
+}
+
+describe("Tool.define", () => {
+  test("object-defined tool does not mutate the original init object", async () => {
+    const original = makeTool("test")
+    const originalExecute = original.execute
+
+    const info = await runtime.runPromise(Tool.define("test-tool", Effect.succeed(original)))
+
+    await Effect.runPromise(info.init())
+    await Effect.runPromise(info.init())
+    await Effect.runPromise(info.init())
+
+    expect(original.execute).toBe(originalExecute)
+  })
+
+  test("effect-defined tool returns fresh objects and is unaffected", async () => {
+    const info = await runtime.runPromise(
+      Tool.define(
+        "test-fn-tool",
+        Effect.succeed(() => Effect.succeed(makeTool("test"))),
+      ),
+    )
+
+    const first = await Effect.runPromise(info.init())
+    const second = await Effect.runPromise(info.init())
+
+    expect(first).not.toBe(second)
+  })
+
+  test("object-defined tool returns distinct objects per init() call", async () => {
+    const info = await runtime.runPromise(Tool.define("test-copy", Effect.succeed(makeTool("test"))))
+
+    const first = await Effect.runPromise(info.init())
+    const second = await Effect.runPromise(info.init())
+
+    expect(first).not.toBe(second)
+  })
+
+  test("execute receives decoded parameters", async () => {
+    const parameters = Schema.Struct({
+      count: Schema.NumberFromString.pipe(Schema.optional, Schema.withDecodingDefaultType(Effect.succeed(5))),
+    })
+    const calls: Array<Schema.Schema.Type<typeof parameters>> = []
+    const info = await runtime.runPromise(
+      Tool.define(
+        "test-decoded",
+        Effect.succeed({
+          description: "test tool",
+          parameters,
+          execute(args: Schema.Schema.Type<typeof parameters>) {
+            calls.push(args)
+            return Effect.succeed({ title: "test", output: "ok", metadata: { truncated: false } })
+          },
+        }),
+      ),
+    )
+    const ctx: Tool.Context = {
+      sessionID: SessionID.descending(),
+      messageID: MessageID.ascending(),
+      agent: "build",
+      abort: new AbortController().signal,
+      messages: [],
+      metadata() {
+        return Effect.void
+      },
+      ask() {
+        return Effect.void
+      },
+    }
+    const tool = await Effect.runPromise(info.init())
+    const execute = tool.execute as unknown as (args: unknown, ctx: Tool.Context) => ReturnType<typeof tool.execute>
+
+    await Effect.runPromise(execute({}, ctx))
+    await Effect.runPromise(execute({ count: "7" }, ctx))
+
+    expect(calls).toEqual([{ count: 5 }, { count: 7 }])
+  })
+})