import { afterAll, beforeAll, describe, expect, test } from "bun:test" import { CodeModeTool } from "@opencode/core/codemode/tool" import { Permission } from "@opencode/core/permission" import { Tool } from "@opencode/core/tool" import { execute } from "@opencode/core/tool/runtime" import { Agent } from "@opencode/schema/agent" import { Session } from "@opencode/schema/session" import { SessionMessage } from "@opencode/schema/session-message" import type { Info } from "@opencode/schema/tool" import { Deferred, Effect, Fiber, Schema } from "effect" import { TestClock } from "effect/testing" import { it } from "./lib/effect" import { permissionService } from "./lib/permission" const context = { sessionID: Session.ID.make("ses_execute"), agent: Agent.ID.make("build"), messageID: SessionMessage.ID.make("msg_execute"), id: Tool.CallID.make("call_execute"), progress: () => Effect.void, } const allowAll = permissionService({ assert: () => Effect.void }) const createCodeMode = (tools: ReadonlyMap, permission = allowAll) => CodeModeTool.create({ tools }, (_, tool, input, context) => execute(tool, input, context), permission) test("execute describes invariant Code Mode behavior", () => { const description = createCodeMode(new Map()).description expect(description).toContain("Lua") expect(description).toContain("fetch") expect(description).toContain("search") expect(description).not.toContain("Promise.allSettled") }) test("execute rejects filesystem fetches", async () => { const result = await Effect.runPromise( createCodeMode(new Map()).execute({ code: 'return fetch("file:///blocked.txt")' }, context), ) expect(result.output.error).toBe(true) expect(result.output.output).toContain("fetch cannot be used for filesystem access.") }) test("execute rejects wrong-language source before admitting even its first tool call", async () => { let calls = 0 const touch: Info = { name: "touch", description: "Record a call", input: Schema.Struct({}), execute: () => Effect.sync(() => ({ content: String(++calls) })), } const result = await Effect.runPromise( createCodeMode(new Map([["touch", touch]])).execute( // The first statement is valid Lua. The JavaScript-only second statement must reject the whole chunk. { code: "tools.touch({}); const answer = 42; return answer" }, context, ), ) expect(result.output.error).toBe(true) expect(result.output.toolCalls).toEqual([]) expect(calls).toBe(0) }) describe("fetch permissions", () => { const hits: Array = [] let server: ReturnType beforeAll(() => { server = Bun.serve({ port: 0, fetch: async (request) => { const url = new URL(request.url) hits.push(`${request.method} ${url.pathname}`) if (url.pathname === "/redirect") return Response.redirect(`${url.origin}/blocked`, 302) return new Response(await request.text(), { status: 201, statusText: "Created", headers: { "x-test": "ok" } }) }, }) }) afterAll(() => server.stop(true)) test("fetch asks the webfetch permission for its URL as the execute call", async () => { hits.length = 0 const asked: Array = [] const url = `${server.url.origin}/allowed` const result = await Effect.runPromise( createCodeMode( new Map(), permissionService({ assert: (input) => Effect.sync(() => void asked.push(input)) }), ).execute({ code: `return fetch("${url}", { method = "POST", body = "posted" })` }, context), ) expect(JSON.parse(result.output.output)).toMatchObject({ status: 201, statusText: "Created", url, redirected: false, ok: true, headers: { "x-test": "ok" }, body: "posted", }) expect(hits).toEqual(["POST /allowed"]) expect(asked).toEqual([ { action: "webfetch", resources: [url], save: ["*"], metadata: { method: "POST", url }, sessionID: context.sessionID, agent: context.agent, source: { type: "tool", messageID: context.messageID, id: context.id }, }, ]) }) test("a redirect to a URL denied for webfetch is never requested", async () => { hits.length = 0 const denied = permissionService({ assert: () => Effect.void, assertNotDenied: (input) => Effect.fail(new Permission.BlockedError({ rules: [], permission: input.action, resources: input.resources })), }) const result = await Effect.runPromise( createCodeMode(new Map(), denied).execute({ code: `return fetch("${server.url.origin}/redirect")` }, context), ) expect(result.output.error).toBe(true) expect(result.output.output).toContain("Permission denied") expect(hits).toEqual(["GET /redirect"]) }) test("a rejection with feedback reaches the model as the execute error", async () => { hits.length = 0 const url = `${server.url.origin}/corrected` const corrected = permissionService({ assert: () => Effect.fail(new Permission.CorrectedError({ feedback: "use the docs site instead" })), }) const result = await Effect.runPromise( createCodeMode(new Map(), corrected).execute({ code: `return fetch("${url}")` }, context), ) expect(result.output.error).toBe(true) expect(result.output.output).toBe(`The user rejected fetching ${url}: use the docs site instead`) expect(hits).toEqual([]) }) test("a URL denied for webfetch fails execute without making a request", async () => { hits.length = 0 const url = `${server.url.origin}/denied` const denied = permissionService({ assert: (input) => Effect.fail(new Permission.BlockedError({ rules: [], permission: input.action, resources: input.resources })), }) const result = await Effect.runPromise( createCodeMode(new Map(), denied).execute({ code: `return fetch("${url}")` }, context), ) expect(result.output.error).toBe(true) expect(result.output.output).toContain("Permission denied") expect(hits).toEqual([]) }) }) test("canonical execution distinguishes declared, model-only, and raw schema outputs", async () => { const declared: Info = { name: "declared", description: "Declared", input: Schema.Struct({ value: Schema.String }), output: Schema.Struct({ value: Schema.String }), execute: ({ value }) => Effect.succeed({ output: { value } }), } const modelOnlyInput = Schema.Struct({}) const modelOnly = { name: "model_only", description: "Model only", input: modelOnlyInput, execute: () => Effect.succeed({ content: "visible only", metadata: { kind: "model" } }), } satisfies Info const raw: Info = { name: "raw", description: "Raw", input: {}, output: {}, execute: (input) => Effect.succeed({ output: input, content: "raw" }), } expect(await Effect.runPromise(execute(declared, { value: "encoded" }, context))).toEqual({ output: { value: "encoded" }, content: [{ type: "text", text: '{"value":"encoded"}' }], }) expect(await Effect.runPromise(execute(modelOnly, {}, context))).toEqual({ output: undefined, content: [{ type: "text", text: "visible only" }], metadata: { kind: "model" }, }) expect(await Effect.runPromise(execute(raw, { unchecked: true }, context))).toEqual({ output: { unchecked: true }, content: [{ type: "text", text: "raw" }], }) }) test("declared outputs cannot bypass validation and raw outputs stay JSON-compatible", async () => { const missing: Info = { name: "missing", description: "Missing output", input: Schema.Struct({}), output: Schema.String, execute: () => Effect.succeed({ content: "not an output" }), } const invalid: Info = { name: "invalid", description: "Invalid raw output", input: {}, output: {}, execute: () => Effect.succeed({ output: 1n, content: "not JSON" }), } expect((await Effect.runPromiseExit(execute(missing, {}, context))).toString()).toContain( "Tool did not return its declared output", ) expect((await Effect.runPromiseExit(execute(invalid, {}, context))).toString()).toContain( "Tool returned a non-JSON value", ) }) test("foreign typed failures settle as Tool.Error at the untrusted boundary", async () => { class ForeignFailure extends Schema.TaggedError()("Plugin.ForeignFailure", { message: Schema.String, }) {} const lying: Info = { name: "lying", description: "Fails with a non-Tool.Error typed failure", input: Schema.Struct({}), execute: () => new ForeignFailure({ message: "transport died" }) as never, } const error = await Effect.runPromise(execute(lying, {}, context).pipe(Effect.flip)) expect(error).toBeInstanceOf(Tool.Error) expect(error.message).toBe("transport died") }) test("execute supports callable namespace tools", async () => { const callable: Info = { name: "admin", description: "Administer Slack", input: Schema.Struct({}), output: Schema.String, options: { namespace: "slack" }, execute: () => Effect.succeed({ output: "admin" }), } const child: Info = { name: "create", description: "Create a Slack resource", input: Schema.Struct({}), output: Schema.String, options: { namespace: "slack.admin" }, execute: () => Effect.succeed({ output: "created" }), } const codeMode = createCodeMode( new Map([ ["slack_admin", callable], ["slack_admin_create", child], ]), ) const result = await Effect.runPromise( codeMode.execute({ code: "return {tools.slack.admin({}), tools.slack.admin.create({})}" }, context), ) expect(result.metadata).toEqual({ toolCalls: [ { tool: "slack.admin", status: "completed" }, { tool: "slack.admin.create", status: "completed" }, ], }) expect(result.content).toEqual([{ type: "text", text: '[\n "admin",\n "created"\n]' }]) }) test("executions share the prepared runtime of an inventory while keeping per-call state", async () => { let conversions = 0 const echo: Info = { name: "echo", description: "Echo a value", input: { "~standard": { version: 1, vendor: "test", validate: (value: unknown) => ({ value }), jsonSchema: { input: () => { conversions++ return { type: "object", properties: { value: { type: "string" } }, required: ["value"] } }, output: () => ({}), }, }, }, execute: (input, context) => Effect.succeed({ content: `${context.id}:${(input as { value: string }).value}` }), } const inventory = { tools: new Map([["echo", echo]]) } const codeMode = CodeModeTool.create(inventory, (_, tool, input, context) => execute(tool, input, context), allowAll) const run = (id: string) => { const progress: Array = [] return CodeModeTool.create(inventory, (_, tool, input, context) => execute(tool, input, context), allowAll) .execute( { code: `return tools.echo({ value = "${id}" })` }, { ...context, id: Tool.CallID.make(id), progress: (update) => Effect.sync(() => progress.push(update)) }, ) .pipe(Effect.map((result) => ({ output: result.output.output, progress }))) } expect(CodeModeTool.catalog(inventory).tools).toHaveLength(1) const results = await Effect.runPromise( Effect.all([run("first"), run("second")], { concurrency: "unbounded" }).pipe(Effect.provide(TestClock.layer())), ) expect(results).toEqual( ["first", "second"].map((id) => ({ output: `${id}:${id}`, progress: [{ toolCalls: [{ tool: "echo", status: "running", input: { value: id } }] }], })), ) expect( (await Effect.runPromise(codeMode.execute({ code: 'return tools.echo({ value = "x" })' }, context))).output .output, ).toBe("call_execute:x") expect(conversions).toBe(1) await Effect.runPromise( createCodeMode(new Map(inventory.tools)).execute({ code: 'return tools.echo({ value = "next" })' }, context), ) expect(conversions).toBe(2) }) const progressCodeMode = (started: Deferred.Deferred, wait: Effect.Effect) => createCodeMode( new Map([ [ "step", { name: "step", description: "Step", input: Schema.Struct({}), output: Schema.String, execute: () => Effect.succeed({ output: "ok" }), }, ], [ "wait", { name: "wait", description: "Wait", input: Schema.Struct({}), output: Schema.String, execute: () => Deferred.succeed(started, undefined).pipe(Effect.andThen(wait), Effect.as({ output: "ok" })), }, ], ]), ) const row = (tool: string, status: string) => ({ tool, status }) it.effect("execute sends at most one progress update per 100 ms with the latest rows", () => Effect.gen(function* () { const started = yield* Deferred.make() const progress: Array = [] const fiber = yield* progressCodeMode(started, Effect.sleep("250 millis")) .execute( { code: "for i = 1, 3 do tools.step({}) end; tools.wait({}); return tools.step({})" }, { ...context, progress: (update) => Effect.sync(() => progress.push(update)) }, ) .pipe(Effect.forkChild({ startImmediately: true })) yield* Deferred.await(started) yield* Effect.yieldNow expect(progress).toEqual([{ toolCalls: [row("step", "running")] }]) yield* TestClock.adjust("100 millis") const steps = [row("step", "completed"), row("step", "completed"), row("step", "completed")] expect(progress.slice(1)).toEqual([{ toolCalls: [...steps, row("wait", "running")] }]) // The wait ends 150 ms after the trailing update, so its row goes out at once; the last step's never does. yield* TestClock.adjust("150 millis") const result = yield* Fiber.join(fiber) yield* TestClock.adjust("1 second") expect(progress.slice(2)).toEqual([{ toolCalls: [...steps, row("wait", "completed")] }]) expect(result.metadata).toEqual({ toolCalls: [...steps, row("wait", "completed"), row("step", "completed")] }) }), ) it.effect("interrupted execute sends its latest progress rows before it settles", () => Effect.gen(function* () { const started = yield* Deferred.make() const progress: Array = [] const fiber = yield* progressCodeMode(started, Effect.never) .execute( { code: "tools.step({}); tools.step({}); return tools.wait({})" }, { ...context, progress: (update) => Effect.sync(() => progress.push(update)) }, ) .pipe(Effect.forkChild({ startImmediately: true })) yield* Deferred.await(started) yield* Fiber.interrupt(fiber) yield* TestClock.adjust("1 second") expect(progress).toEqual([ { toolCalls: [row("step", "running")] }, { toolCalls: [row("step", "completed"), row("step", "completed"), row("wait", "error")] }, ]) }), )