Q52: goal closures stop linked work, and running sessions hear goal changes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
d9cb86a999
commit
69e991ff93
5 files changed
+134
-1
No files matched your search
@@ -0,0 +1,33 @@
|
||||
import type { DatabaseSync } from "node:sqlite";
|
||||
|
||||
interface GoalState {
|
||||
title: string;
|
||||
description: string;
|
||||
status: string;
|
||||
}
|
||||
|
||||
/** Returns a notice whenever the run's goal is rewritten or closed after the run started, once per change. */
|
||||
export function createGoalChangePoller(db: DatabaseSync, goalId: string | null): () => string | null {
|
||||
if (!goalId) return () => null;
|
||||
const read = () => db.prepare("select title, description, status from goals where id = ?").get(goalId) as GoalState | undefined;
|
||||
let seen = read();
|
||||
return () => {
|
||||
const now = read();
|
||||
if (!seen || !now || (now.title === seen.title && now.description === seen.description && now.status === seen.status)) return null;
|
||||
const before = seen;
|
||||
seen = now;
|
||||
console.info(`[task-run] goal ${goalId} changed mid-run (${now.status}); notifying the session`);
|
||||
if (now.status !== before.status && now.status !== "active") {
|
||||
return `The goal for this work ("${now.title}") is now ${now.status}. Stop starting new work on it: wrap up, report what you finished, and set this task's status.`;
|
||||
}
|
||||
return `The goal for this work changed while you were working. It now reads: "${now.title}" — ${now.description} Re-check your plan against it.`;
|
||||
};
|
||||
}
|
||||
|
||||
/** Joins mid-run notices from several sources into one follow-up, or null when none has news. */
|
||||
export function combineMidRunPolls(...polls: Array<() => string | null>): () => string | null {
|
||||
return () => {
|
||||
const notices = polls.map((poll) => poll()).filter((notice): notice is string => notice !== null);
|
||||
return notices.length > 0 ? notices.join("\n\n") : null;
|
||||
};
|
||||
}
|
||||
@@ -53,6 +53,7 @@ export class HeartbeatScheduler {
|
||||
and coalesce(u.request_count, 0) < p.monthly_request_limit
|
||||
and not exists (select 1 from task_runs r where r.task_id = t.id and r.state in ('queued', 'running'))
|
||||
and not exists (select 1 from tasks c where c.parent_task_id = t.id and c.status <> 'done')
|
||||
and (t.origin is not null or t.goal_id is null or exists (select 1 from goals g where g.id = t.goal_id and g.status = 'active'))
|
||||
and (a.heartbeat_interval_sec <= 0 or not exists (
|
||||
select 1 from task_runs hb where hb.agent_id = a.id and hb.trigger = 'heartbeat' and hb.state <> 'superseded'
|
||||
and hb.created_at > strftime('%Y-%m-%dT%H:%M:%fZ', ?, '-' || a.heartbeat_interval_sec || ' seconds')
|
||||
|
||||
@@ -11,6 +11,7 @@ import { deliverChatReplyIfMissing } from "./chat-agent-wake-service";
|
||||
import { SkillService } from "./skill-service";
|
||||
import { runAgentLoop, runTools } from "./agent-run-loop";
|
||||
import type { RunModel } from "./skill-runtime";
|
||||
import { combineMidRunPolls, createGoalChangePoller } from "./goal-change-poller";
|
||||
import type { AgentToolContext } from "./agent-tools";
|
||||
import {
|
||||
buildAgentInstructions,
|
||||
@@ -121,6 +122,10 @@ export class TaskExecutionService {
|
||||
if (!sameWorkspaceAccess(run.workspace_mount_json, workspaceAccess(task, workspace))) {
|
||||
return this.supersedeRun(orgId, runId, run, "the task's workspace access changed after this run was queued.");
|
||||
}
|
||||
const goal = task.origin === null && task.goal_id
|
||||
? this.db.prepare("select status from goals where id = ?").get(task.goal_id) as { status: string } | undefined
|
||||
: undefined;
|
||||
if (goal && goal.status !== "active") return this.supersedeRun(orgId, runId, run, `the task's goal is ${goal.status}, so its work stops.`);
|
||||
const openSubtask = this.db.prepare("select title from tasks where parent_task_id = ? and status <> 'done' limit 1").get(task.id) as { title: string } | undefined;
|
||||
if (openSubtask) return this.supersedeRun(orgId, runId, run, `the task has open subtasks ("${openSubtask.title}"); it runs again once they are done.`);
|
||||
const approvalProblem = executionApprovalProblem(this.db, orgId, task);
|
||||
@@ -187,6 +192,8 @@ export class TaskExecutionService {
|
||||
.toolkitsForAgent(orgId, agent.id)
|
||||
.map(connectorDisplayName);
|
||||
const language = this.appLanguage();
|
||||
// Baseline taken before any await, so a goal closed while tools load still reaches the session.
|
||||
const goalChanges = createGoalChangePoller(this.db, task.goal_id);
|
||||
const externalTools = await loadComposioTools(this.db, this.providers.integrationKeys, orgId, agent.id);
|
||||
const result = await runAgentLoop(
|
||||
{
|
||||
@@ -223,7 +230,10 @@ export class TaskExecutionService {
|
||||
externalTools,
|
||||
onModelTurn: contextRecorder.commit,
|
||||
inputBudgetTokens: inputBudgetTokens(provider.context_window_tokens),
|
||||
pollNewMessages: createRunMessagePoller(this.db, { orgId, onInjected: contextRecorder.addMidRun, agentId: agent.id, sourceThreadId: chatContext?.threadId ?? null, runStartedAt }),
|
||||
pollNewMessages: combineMidRunPolls(
|
||||
goalChanges,
|
||||
createRunMessagePoller(this.db, { orgId, onInjected: contextRecorder.addMidRun, agentId: agent.id, sourceThreadId: chatContext?.threadId ?? null, runStartedAt }),
|
||||
),
|
||||
},
|
||||
);
|
||||
providerRequestId = result.lastRequestId;
|
||||
|
||||
@@ -0,0 +1,69 @@
|
||||
import { randomUUID } from "node:crypto";
|
||||
import { tmpdir } from "node:os";
|
||||
import { join } from "node:path";
|
||||
import { describe, expect, test } from "vitest";
|
||||
import { DatabaseConnection } from "../../src/main/database/connection";
|
||||
import { OnboardingService } from "../../src/main/services/onboarding-service";
|
||||
import { ProviderService } from "../../src/main/services/provider-service";
|
||||
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";
|
||||
|
||||
const testVault = {
|
||||
encryptSecret: (secret: string) => Buffer.from(secret, "utf8"),
|
||||
decryptSecret: (ciphertext: Buffer) => ciphertext.toString("utf8"),
|
||||
};
|
||||
|
||||
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 onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
connection.transaction(() => {
|
||||
orgId = onboarding.bootstrap({
|
||||
companyName: "FlowPilot", companyDescription: "CSM analytics.", employeeName: "Yang", workspacePath: "",
|
||||
executionMode: "ai_ceo", autoApproveHires: true, reviewDeliverables: false,
|
||||
goalTitle: "Ship the MVP.", goalDescription: "A working product.",
|
||||
providerKey: "openai", providerSecret: "sk-test-key", providerModel: "gpt-5.4",
|
||||
}).orgId;
|
||||
});
|
||||
const goalId = (connection.db.prepare("select id from goals where org_id = ?").get(orgId) as { id: string }).id;
|
||||
return { connection, execution, orgId, goalId };
|
||||
}
|
||||
|
||||
describe("goal changes", () => {
|
||||
test("work under an achieved goal stops: its queued run is superseded and the heartbeat leaves it alone", async () => {
|
||||
let calls = 0;
|
||||
const h = company(() => { calls += 1; return { text: "done", toolCalls: [], rawResponseJson: "{}" }; });
|
||||
const kickoff = claimNextQueuedRun(h.connection)!;
|
||||
h.connection.db.prepare("update goals set status = 'achieved' where id = ?").run(h.goalId);
|
||||
await h.execution.executeRun(h.orgId, kickoff.id);
|
||||
expect(h.connection.db.prepare("select state, error_message from task_runs where id = ?").get(kickoff.id))
|
||||
.toEqual({ state: "superseded", error_message: "Superseded: the task's goal is achieved, so its work stops." });
|
||||
expect(h.connection.transaction(() => h.execution.enqueueHeartbeatRuns())).toEqual([]);
|
||||
expect(calls).toBe(0);
|
||||
h.connection.close();
|
||||
});
|
||||
|
||||
test("a running session hears about a goal rewrite before its next step", async () => {
|
||||
let followUp: string | undefined;
|
||||
let goalId = "";
|
||||
const h = company((input, db) => {
|
||||
const last = input.messages.at(-1);
|
||||
if (last?.role === "tool") {
|
||||
followUp = last.followUp;
|
||||
return { text: "Re-planned.", toolCalls: [], rawResponseJson: "{}" };
|
||||
}
|
||||
// Same updated_at on purpose: the notice must follow the content, not the timestamp.
|
||||
db.prepare("update goals set title = 'Ship the paid beta.', description = 'Charge from day one.' where id = ?").run(goalId);
|
||||
return { text: "Checking messages.", toolCalls: [{ id: "c1", name: "read_messages", arguments: {} }], rawResponseJson: "{}" };
|
||||
});
|
||||
goalId = h.goalId;
|
||||
const kickoff = claimNextQueuedRun(h.connection)!;
|
||||
await h.execution.executeRun(h.orgId, kickoff.id);
|
||||
expect(followUp).toContain('The goal for this work changed while you were working. It now reads: "Ship the paid beta." — Charge from day one. Re-check your plan against it.');
|
||||
h.connection.close();
|
||||
});
|
||||
});
|
||||
@@ -0,0 +1,20 @@
|
||||
import { DatabaseSync } from "node:sqlite";
|
||||
import { describe, expect, test } from "vitest";
|
||||
import { combineMidRunPolls, createGoalChangePoller } from "../../src/main/services/goal-change-poller";
|
||||
|
||||
describe("goal change poller", () => {
|
||||
test("reports a closure and a rewrite once each, against the state seen when the run began", () => {
|
||||
const db = new DatabaseSync(":memory:");
|
||||
db.exec("create table goals(id text primary key, title text, description text, status text); insert into goals values ('g1', 'Ship', 'MVP', 'active');");
|
||||
const poll = createGoalChangePoller(db, "g1");
|
||||
expect(poll()).toBeNull();
|
||||
db.exec("update goals set status = 'achieved' where id = 'g1'");
|
||||
expect(poll()).toBe('The goal for this work ("Ship") is now achieved. Stop starting new work on it: wrap up, report what you finished, and set this task\'s status.');
|
||||
expect(poll()).toBeNull();
|
||||
db.exec("update goals set description = 'Paid beta' where id = 'g1'");
|
||||
expect(poll()).toContain("It now reads: \"Ship\" — Paid beta");
|
||||
expect(createGoalChangePoller(db, null)()).toBeNull();
|
||||
expect(combineMidRunPolls(() => "a", () => null, () => "b")()).toBe("a\n\nb");
|
||||
db.close();
|
||||
});
|
||||
});
|
||||
Reference in new issue
Block a user