evOo-local/packages/adapters/src/executor-capabilities.test.ts
Claude 848abc757a
Merge capabilities: least-privilege builtin tools per bot and schedules on behalf
The per-bot tool policy now carries the main bot's own tools: delegate_task
and update_user_profile are the main-only builtins, so the separate filters
from the profile and delegation branches are gone. One set of exposed tool
names drives both the executor's call guard and the prompt. The schedule
tool says new routines default to the user's time zone, which makes the
extra prompt line unnecessary. Tests that relied on an empty tool list
meaning every tool, or on the tool's own refusal text, now pass the tool
explicitly or expect the executor's refusal.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MRtGG9ARfqqhma3t85Ro2u
2026-10-02 23:53:19 +00:00

280 lines
10 KiB
TypeScript

import type { AgentRunRequest } from "@rakazo/adapter-kit";
import { describe, expect, it, vi } from "vitest";
import type * as ComputerLifecycleModule from "./computer-lifecycle.js";
import { createRunExecutor, MAIN_ONLY_BUILTIN_TOOLS } from "./executor.js";
vi.mock("./computer-lifecycle.js", async (importOriginal) => ({
...(await importOriginal<typeof ComputerLifecycleModule>()),
acquireComputerExecutionLease: async () => null,
provisionComputer: async () => ({ id: "computer-1", kind: "desktop" }),
}));
type BotPolicy = { kind?: string; restrictBuiltinTools?: boolean; allowedBuiltinTools?: string[] };
type Call = { name: string; args?: Record<string, unknown> };
type Effect = { id: string; status: string; idempotencyKey: string } & Record<string, unknown>;
/** In-memory external effects, enough for the idempotency record of a mutating builtin tool. */
function effectStore() {
const effects: Effect[] = [];
const find = (where: { id?: string; idempotencyKey?: string }) =>
effects.find((effect) =>
where.id ? effect.id === where.id : effect.idempotencyKey === where.idempotencyKey,
);
return {
findMany: vi.fn(async () => []),
findUnique: vi.fn(async ({ where }: { where: { id?: string; idempotencyKey?: string } }) => {
const effect = find(where);
return effect ? { ...effect } : null;
}),
create: vi.fn(async ({ data }: { data: Omit<Effect, "id"> }) => {
const effect = { ...data, id: `effect-${effects.length + 1}` } as Effect;
effects.push(effect);
return { ...effect };
}),
update: vi.fn(async ({ where, data }: { where: { id: string }; data: Partial<Effect> }) => {
Object.assign(find(where) ?? {}, data);
}),
updateMany: vi.fn(
async ({ where, data }: { where: { id: string; status: string }; data: Partial<Effect> }) => {
const effect = find(where);
if (!effect || effect.status !== where.status) return { count: 0 };
Object.assign(effect, data);
return { count: 1 };
},
),
};
}
async function runBot(policy: BotPolicy, calls: Call[] = [], models: Record<string, unknown> = {}) {
const run = {
id: "run-1",
botId: "bot-1",
threadId: "thread-1",
taskId: "task-1",
spaceId: "space-1",
userId: "user-1",
status: "queued",
trigger: "user",
leaseFence: 0,
};
const routineFindMany = vi.fn(async () => []);
const prisma = {
run: {
findUnique: vi.fn(async () => run),
findUniqueOrThrow: vi.fn(async () => run),
updateMany: vi.fn(async ({ data }: { data: Record<string, unknown> }) => {
Object.assign(run, data);
return { count: 1 };
}),
},
bot: {
findUniqueOrThrow: vi.fn(async () => ({
id: run.botId,
name: "Assistant",
title: "Assistant",
description: "Test assistant",
computerId: "computer-1",
computer: { id: "computer-1", scope: "dedicated" },
kind: "specialist",
restrictBuiltinTools: false,
allowedBuiltinTools: [],
...policy,
})),
findMany: vi.fn(async () => []),
},
attempt: {
create: vi.fn(async () => ({ id: "attempt-1" })),
update: vi.fn(),
updateMany: vi.fn(async () => ({ count: 1 })),
},
thread: { findUniqueOrThrow: vi.fn(async () => ({ id: run.threadId, groupId: null })) },
message: { findMany: vi.fn(async () => []) },
task: { findUniqueOrThrow: vi.fn(async () => ({ id: run.taskId, prompt: "Work" })) },
connection: { findMany: vi.fn(async () => []) },
spaceModelPreference: { findFirst: vi.fn(async () => null) },
userModelCredential: { findFirst: vi.fn(async () => null) },
deploymentSettings: {
findUnique: vi.fn(async () => ({
defaultModelProvider: "scripted",
defaultModelId: "scripted",
})),
},
taughtSkill: { findMany: vi.fn(async () => []) },
agentSecret: { findMany: vi.fn(async () => []) },
agentSkill: { findMany: vi.fn(async () => []) },
scratchpadItem: { findMany: vi.fn(async () => []) },
actionApprovalRule: { findMany: vi.fn(async () => []) },
actionAutoReviewPreference: { findUnique: vi.fn(async () => ({ enabled: false })) },
externalEffect: effectStore(),
routine: { findMany: routineFindMany },
// The user block every run states.
user: { findUnique: vi.fn(async () => ({ name: "Ada" })) },
userProfile: { findUnique: vi.fn(async () => null) },
userFact: { findMany: vi.fn(async () => []) },
...models,
};
let request: AgentRunRequest | undefined;
const results: unknown[] = [];
const jobs = { enqueue: vi.fn(async () => undefined) };
const executor = createRunExecutor({
prisma,
runtime: {
describe: () => ({ capabilities: { scripted: false } }),
run: async function* (next: AgentRunRequest) {
request = next;
for (const [index, call] of calls.entries()) {
results.push(await next.executeTool!(call.name, call.args ?? {}, `call-${index}`));
}
yield { type: "done" as const, text: "Done" };
},
},
sandbox: { describe: () => ({ capabilities: { graphical: false } }) },
memory: { read: async () => ({ documents: [] }) },
memoryProviders: { resolve: async () => null },
events: {
append: vi.fn(async () => undefined),
pauseRunForInput: vi.fn(async () => true),
finalizeRun: vi.fn(async () => ({ continuationRunId: null })),
},
jobs,
secrets: [],
} as unknown as Parameters<typeof createRunExecutor>[0]);
await executor.continueRun(run.id, "worker-1");
if (!request) throw new Error("the runtime was never started");
const tools = request.tools;
return { names: tools.map((tool) => tool.name), tools, results, routineFindMany };
}
describe("bot capabilities in runs", () => {
it("exposes every builtin to an unrestricted bot", async () => {
const { names } = await runBot({});
expect(names).toEqual(expect.arrayContaining(["write_file", "shell", "schedule_list"]));
});
it("exposes only the allowlisted builtins and refuses calls to any other", async () => {
const { names, results, routineFindMany } = await runBot(
{ restrictBuiltinTools: true, allowedBuiltinTools: ["schedule_list", "web_search"] },
[
{ name: "write_file", args: { path: "notes/x.txt", content: "nope" } },
{ name: "schedule_list" },
],
);
expect(names).toEqual(["web_search", "schedule_list"]);
expect(results[0]).toEqual({ error: "Tool write_file is not available in this run." });
expect(results[1]).toEqual({ routines: [] });
expect(routineFindMany).toHaveBeenCalledOnce();
});
it("exposes no builtins when the allowlist is empty", async () => {
const { names, results } = await runBot({ restrictBuiltinTools: true }, [
{ name: "schedule_list" },
]);
expect(names).toEqual([]);
expect(results[0]).toEqual({ error: "Tool schedule_list is not available in this run." });
});
it("refuses a tool no run ever exposes", async () => {
const { results } = await runBot({}, [{ name: "made_up_tool" }]);
expect(results[0]).toEqual({ error: "Tool made_up_tool is not available in this run." });
});
it("offers schedule_create's bot option to the main bot only", async () => {
const properties = async (policy: BotPolicy) =>
Object.keys(
((await runBot(policy)).tools.find((tool) => tool.name === "schedule_create")?.inputSchema
.properties ?? {}) as object,
);
expect(await properties({ kind: "main" })).toContain("bot");
expect(await properties({})).not.toContain("bot");
});
it("withholds main-only tools from every other bot", async () => {
const mainOnly = MAIN_ONLY_BUILTIN_TOOLS as Set<string>;
// No main-only tool ships yet in this package; schedule_list stands in for one.
mainOnly.add("schedule_list");
try {
const specialist = await runBot({}, [{ name: "schedule_list" }]);
expect(specialist.names).not.toContain("schedule_list");
expect(specialist.results[0]).toEqual({
error: "Tool schedule_list is not available in this run.",
});
const main = await runBot({ kind: "main" }, [{ name: "schedule_list" }]);
expect(main.names).toContain("schedule_list");
expect(main.results[0]).toEqual({ routines: [] });
} finally {
mainOnly.delete("schedule_list");
}
});
it("schedules for a named specialist when the main bot asks, in that bot's space and thread", async () => {
const routineCreate = vi.fn(async ({ data }: { data: Record<string, unknown> }) => ({
id: "routine-1",
...data,
}));
const { results } = await runBot(
{ kind: "main" },
[
{
name: "schedule_create",
args: { name: "Digest", prompt: "Summarize", delayMinutes: 5, bot: "Atlas" },
},
],
{
bot: {
findUniqueOrThrow: vi.fn(async () => ({
id: "bot-1",
name: "Chief",
title: "",
description: "",
computerId: "computer-1",
computer: { id: "computer-1", scope: "dedicated" },
kind: "main",
restrictBuiltinTools: false,
allowedBuiltinTools: [],
})),
findMany: vi.fn(async () => [
{
id: "atlas-1",
name: "Atlas",
title: "",
description: "",
spaceId: "space-2",
space: { name: "Studio" },
thread: { id: "atlas-dm" },
},
]),
},
chatGroup: { findFirst: vi.fn(async () => null) },
routine: { findMany: vi.fn(async () => []), create: routineCreate },
},
);
expect(results[0]).toMatchObject({ ok: true, routineId: "routine-1", bot: "Atlas" });
expect(routineCreate).toHaveBeenCalledWith(
expect.objectContaining({
data: expect.objectContaining({
spaceId: "space-2",
botId: "atlas-1",
threadId: "atlas-dm",
createdByBotId: "bot-1",
}),
}),
);
});
it("refuses the bot option from any other bot", async () => {
const routineCreate = vi.fn();
const { results } = await runBot(
{},
[
{
name: "schedule_create",
args: { name: "x", prompt: "y", delayMinutes: 5, bot: "Atlas" },
},
],
{ routine: { findMany: vi.fn(async () => []), create: routineCreate } },
);
expect(results[0]).toEqual({ error: "Only the main agent can schedule for another bot." });
expect(routineCreate).not.toHaveBeenCalled();
});
});