Q83: a model passes an agent check before it does agent work
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
4e232b241f
commit
665bf3cae1
35 files changed
+350
-47
No files matched your search
@@ -9,6 +9,7 @@ import { providerAdapters } from "../../src/main/services/provider-adapters";
|
||||
import { TaskExecutionService } from "../../src/main/services/task-execution-service";
|
||||
import { RISK_LEVEL, TASK_PRIORITY, TASK_STATUS } from "../../src/shared/domain";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -27,7 +28,7 @@ describe("agent model", () => {
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-model-${randomUUID()}.db`));
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
const tasks: Record<string, string> = {};
|
||||
@@ -53,7 +54,7 @@ describe("agent model", () => {
|
||||
const turn = async (): Promise<AgentTurnResult> => ({ text: "done", toolCalls: [], rawResponseJson: "{}" });
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-model-az-${randomUUID()}.db`));
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
let taskId = "";
|
||||
@@ -83,7 +84,7 @@ describe("agent model", () => {
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-model-pin-${randomUUID()}.db`));
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
let taskId = "";
|
||||
|
||||
@@ -11,6 +11,7 @@ import { SnapshotService } from "../../src/main/services/snapshot-service";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import { canonicalMentionHandle } from "../../src/shared/mention-handles";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput, ToolCall } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -59,7 +60,7 @@ function scriptedTeam(): (input: ProviderToolTurnInput) => Promise<AgentTurnResu
|
||||
describe("inter-agent reply", () => {
|
||||
test("an @mention wakes the teammate, who replies in the same thread", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-reply-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, scriptedTeam());
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(scriptedTeam()));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
const snapshot = new SnapshotService(connection.db);
|
||||
|
||||
@@ -12,6 +12,7 @@ import { TaskExecutionService } from "../../src/main/services/task-execution-ser
|
||||
import { SnapshotService } from "../../src/main/services/snapshot-service";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnMessage, AgentTurnResult, ProviderToolTurnInput, ToolCall } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -89,7 +90,7 @@ describe("autonomous bootstrap run", () => {
|
||||
const workspacePath = join(tmpdir(), `ac-auto-ws-${randomUUID()}`);
|
||||
mkdirSync(workspacePath, { recursive: true });
|
||||
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, scriptedCeo());
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(scriptedCeo()));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const snapshot = new SnapshotService(connection.db);
|
||||
|
||||
@@ -183,7 +184,7 @@ describe("autonomous bootstrap run", () => {
|
||||
test("a goal created after onboarding reaches the CEO and runs autonomously", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-goal-${randomUUID()}.db`));
|
||||
// A minimal well-behaved agent: does its session, closes its task, then stops.
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, async (input: ProviderToolTurnInput) => {
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(async (input: ProviderToolTurnInput) => {
|
||||
const firstUser = input.messages.find((message) => message.role === "user");
|
||||
const taskId = firstUser && firstUser.role === "user" ? /Task id: (\S+)/.exec(firstUser.text)?.[1] : undefined;
|
||||
const alreadyClosed = input.messages.some(
|
||||
@@ -201,7 +202,7 @@ describe("autonomous bootstrap run", () => {
|
||||
toolCalls: [],
|
||||
rawResponseJson: JSON.stringify({ id: "resp_goal" }),
|
||||
};
|
||||
});
|
||||
}));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
|
||||
@@ -11,6 +11,7 @@ import { ChatService } from "../../src/main/services/chat-service";
|
||||
import { ensureCeoDirectSpace } from "../../src/main/services/ceo-direct-space";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput, ToolCall } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -67,7 +68,7 @@ describe("founder⇄CEO chat", () => {
|
||||
test("a founder DM wakes the CEO, who sees who is asking and replies in the pinned conversation", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-ceo-chat-${randomUUID()}.db`));
|
||||
const capturedWakePrompts: string[] = [];
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, scriptedCeo(capturedWakePrompts));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(scriptedCeo(capturedWakePrompts)));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
const chat = new ChatService(connection.db);
|
||||
@@ -143,7 +144,7 @@ describe("founder⇄CEO chat", () => {
|
||||
|
||||
test("a CEO who answers in plain text still has the reply delivered into the founder's thread", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-ceo-plain-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, plainTextCeo());
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(plainTextCeo()));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
const chat = new ChatService(connection.db);
|
||||
|
||||
@@ -10,6 +10,7 @@ import { providerAdapters } from "../../src/main/services/provider-adapters";
|
||||
import { TaskExecutionService } from "../../src/main/services/task-execution-service";
|
||||
import { canonicalMentionHandle } from "../../src/shared/mention-handles";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput, ToolCall } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -37,7 +38,7 @@ function scriptedCeoReply(): (input: ProviderToolTurnInput) => Promise<AgentTurn
|
||||
describe("channel anchor wake", () => {
|
||||
test("a plain user message in a channel wakes the most senior member, who replies", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-anchor-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, scriptedCeoReply());
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(scriptedCeoReply()));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
const chat = new ChatService(connection.db);
|
||||
|
||||
@@ -14,6 +14,7 @@ import { dispatchAgentTool, type AgentToolContext } from "../../src/main/service
|
||||
import { ensureCeoDirectSpace } from "../../src/main/services/ceo-direct-space";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -34,7 +35,7 @@ async function founderLine() {
|
||||
return { text: `Reply ${wakePrompts.length}.`, toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-chatctx-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
const chat = new ChatService(connection.db);
|
||||
@@ -153,7 +154,7 @@ describe("chat wake context", () => {
|
||||
}
|
||||
return { text: "Switching to Postgres.", toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let ceoId = "";
|
||||
|
||||
@@ -9,6 +9,7 @@ import { providerAdapters } from "../../src/main/services/provider-adapters";
|
||||
import { TaskExecutionService } from "../../src/main/services/task-execution-service";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -17,7 +18,7 @@ const testVault = {
|
||||
|
||||
function company(turn: (input: ProviderToolTurnInput, db: DatabaseConnection["db"]) => AgentTurnResult) {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-goalchg-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, async (input) => turn(input, connection.db));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(async (input) => turn(input, connection.db)));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
|
||||
@@ -12,6 +12,7 @@ import { TaskExecutionService } from "../../src/main/services/task-execution-ser
|
||||
import { dispatchAgentTool } from "../../src/main/services/agent-tools";
|
||||
import { claimNextQueuedRun, startTaskQueueWorker } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput, ToolCall } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -75,7 +76,7 @@ describe("interactive execution lane", () => {
|
||||
}
|
||||
return turn("done");
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, script);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(script));
|
||||
providers.saveCredential({ orgId, providerId, secret: "sk-test-key", model: "gpt-5.4" });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
|
||||
@@ -162,7 +163,7 @@ describe("interactive execution lane", () => {
|
||||
}
|
||||
return turn("done");
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, script);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(script));
|
||||
providers.saveCredential({ orgId, providerId, secret: "sk-test-key", model: "gpt-5.4" });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
|
||||
@@ -227,7 +228,7 @@ describe("interactive execution lane", () => {
|
||||
const alwaysOverloaded = async (): Promise<AgentTurnResult> => {
|
||||
throw new Error("The engine is currently overloaded, please try again later");
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, alwaysOverloaded);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(alwaysOverloaded));
|
||||
providers.saveCredential({ orgId, providerId, secret: "sk-test-key", model: "gpt-5.4" });
|
||||
const taskExecution = new TaskExecutionService(connection.db, providers);
|
||||
const wakeDana = () => dispatchAgentTool({ db: connection.db, organization, chat }, {
|
||||
|
||||
@@ -11,6 +11,7 @@ import { ChatService } from "../../src/main/services/chat-service";
|
||||
import { ensureCeoDirectSpace } from "../../src/main/services/ceo-direct-space";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -24,7 +25,7 @@ describe("session checkpoints", () => {
|
||||
? { text: "Audit done: all clean.", toolCalls: [], rawResponseJson: "{}" }
|
||||
: { text: "Looking into it.", toolCalls: [{ id: `c_${randomUUID()}`, name: "read_messages", arguments: {} }], rawResponseJson: "{}" };
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-checkpoint-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
|
||||
@@ -9,6 +9,7 @@ import { providerAdapters } from "../../src/main/services/provider-adapters";
|
||||
import { TaskExecutionService } from "../../src/main/services/task-execution-service";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -41,7 +42,7 @@ describe("run journal", () => {
|
||||
throw new Error("Provider dropped the connection mid-run");
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-journal-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
|
||||
@@ -11,6 +11,7 @@ import { ChatService } from "../../src/main/services/chat-service";
|
||||
import { ensureCeoDirectSpace } from "../../src/main/services/ceo-direct-space";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -26,7 +27,7 @@ async function company(reply: (prompt: string) => AgentTurnResult) {
|
||||
return reply(prompt);
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-memory-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
|
||||
@@ -13,6 +13,7 @@ import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import { postJson } from "../../src/main/services/provider-http";
|
||||
import { RUN_STOPPED_MESSAGE, stopRun, withRunControl } from "../../src/main/services/run-control";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -21,7 +22,7 @@ const testVault = {
|
||||
|
||||
function company(turn: (input: ProviderToolTurnInput) => Promise<AgentTurnResult>) {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-stop-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
@@ -44,7 +45,7 @@ describe("stopping runs", () => {
|
||||
return { text: "Announcing.", toolCalls: [{ id: "c1", name: "send_message", arguments: { channel: "general", body: "Should never be posted." } }], rawResponseJson: "{}" };
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-stop-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
@@ -115,7 +116,7 @@ describe("stopping runs", () => {
|
||||
return { text: "done", toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-deadline-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers, undefined, undefined, undefined, 200);
|
||||
let orgId = "";
|
||||
|
||||
@@ -9,6 +9,7 @@ import { providerAdapters } from "../../src/main/services/provider-adapters";
|
||||
import { TaskExecutionService } from "../../src/main/services/task-execution-service";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import type { AgentTurnResult } from "../../src/main/services/provider-tool-turn";
|
||||
import { markModelChecked } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -39,6 +40,7 @@ describe("money budgets", () => {
|
||||
ceoId = summary.ceoAgentId;
|
||||
});
|
||||
const providerId = (connection.db.prepare("select model_provider_id as id from agents where id = ?").get(ceoId) as { id: string }).id;
|
||||
markModelChecked(connection.db, providerId, "gpt-5.4");
|
||||
providers.setPrices({ orgId, providerId, inputDollarsPerMtok: 2.5, cachedInputDollarsPerMtok: null, outputDollarsPerMtok: 10 });
|
||||
connection.db.prepare("update agents set budget_cents = 2000 where id = ?").run(ceoId);
|
||||
|
||||
|
||||
@@ -11,6 +11,7 @@ import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import { ChatService } from "../../src/main/services/chat-service";
|
||||
import { ensureCeoDirectSpace } from "../../src/main/services/ceo-direct-space";
|
||||
import type { AgentTurnResult } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -24,7 +25,7 @@ function company() {
|
||||
return { text: "done", toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-stale-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
|
||||
@@ -13,6 +13,7 @@ import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import { canonicalMentionHandle } from "../../src/shared/mention-handles";
|
||||
import { RISK_LEVEL, TASK_PRIORITY, TASK_STATUS } from "../../src/shared/domain";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput, ToolCall } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -49,7 +50,7 @@ describe("task prompt scope", () => {
|
||||
return { text: step.text, toolCalls: step.toolCalls, rawResponseJson: "{}" };
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-scope-${randomUUID()}.db`));
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
@@ -90,7 +91,7 @@ describe("task prompt scope", () => {
|
||||
calls += 1;
|
||||
return { text: "done", toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
@@ -125,7 +126,7 @@ describe("task prompt scope", () => {
|
||||
test("a manual write run waits while another run is changing the same workspace", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-wsplock-${randomUUID()}.db`));
|
||||
const turn = async (): Promise<AgentTurnResult> => ({ text: "done", toolCalls: [], rawResponseJson: "{}" });
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
|
||||
@@ -13,6 +13,7 @@ import { dispatchAgentTool } from "../../src/main/services/agent-tools";
|
||||
import { claimNextQueuedRun } from "../../src/main/app/task-queue-worker";
|
||||
import { RISK_LEVEL, TASK_PRIORITY, TASK_STATUS } from "../../src/shared/domain";
|
||||
import type { AgentTurnResult, ProviderToolTurnInput } from "../../src/main/services/provider-tool-turn";
|
||||
import { answeringModelCheck } from "../helpers/model-check";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
@@ -28,7 +29,7 @@ function company() {
|
||||
prompts.push(first.text);
|
||||
return { text: "done", toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, answeringModelCheck(turn));
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
const organization = new OrganizationService(connection.db);
|
||||
|
||||
Reference in new issue
Block a user