Q60: step limits end sessions as checkpoints; only sessions without progress spend the budget
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
8775c34650
commit
88463e4396
6 files changed
+166
-35
No files matched your search
@@ -727,7 +727,7 @@ describe("task execution service", () => {
|
||||
spendRun(); spendRun(); spendRun();
|
||||
expect(connection.transaction(() => execution.enqueueHeartbeatRuns())).toEqual([]);
|
||||
const stalled = connection.db.prepare("select status, stalled_reason from tasks where id = ?").get(taskId) as { status: string; stalled_reason: string };
|
||||
expect(stalled).toEqual({ status: "in_progress", stalled_reason: "Stalled: 3 sessions ran without reaching done. Edit, reassign or reopen the task to give it a fresh budget." });
|
||||
expect(stalled).toEqual({ status: "in_progress", stalled_reason: "Stalled: 3 sessions made no progress. Edit, reassign or reopen the task to give it a fresh budget." });
|
||||
expect(connection.db.prepare("select count(*) as n from audit_events where entity_id = ? and action = 'stalled'").get(taskId)).toEqual({ n: 1 });
|
||||
expect(notified).toBe(1);
|
||||
|
||||
@@ -743,6 +743,42 @@ describe("task execution service", () => {
|
||||
connection.close();
|
||||
});
|
||||
|
||||
test("sessions that make progress do not spend the run budget, up to a ceiling of sessions", () => {
|
||||
const connection = createConnection();
|
||||
const organization = new OrganizationService(connection.db);
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault);
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let orgId = "";
|
||||
let taskId = "";
|
||||
connection.transaction(() => {
|
||||
orgId = organization.createOrganization({ organizationName: "Acme", employeeName: "Dana Lee", mission: "m", goal: "g" });
|
||||
const providerId = openAiProviderId(connection, orgId);
|
||||
providers.saveCredential({ orgId, providerId, secret: "sk-live-provider", model: "gpt-5.4" });
|
||||
const agentId = organization.createAgent({
|
||||
orgId, name: "Ada", role: "Engineer", department: "Engineering", modelProviderId: providerId, modelId: "gpt-5.4",
|
||||
budgetCents: 10000, managerAgentId: null, heartbeatEnabled: true,
|
||||
});
|
||||
taskId = organization.createTask({
|
||||
orgId, title: "Big build", description: "d", status: TASK_STATUS.IN_PROGRESS, assigneeAgentId: agentId,
|
||||
risk: RISK_LEVEL.LOW, projectId: null, goalId: null, workspaceId: null, priority: TASK_PRIORITY.NORMAL, requiresFilesystem: false,
|
||||
});
|
||||
});
|
||||
const productiveSession = (): boolean => {
|
||||
const [runId] = connection.transaction(() => execution.enqueueHeartbeatRuns());
|
||||
if (!runId) return false;
|
||||
connection.db.prepare("update task_runs set state = 'completed' where id = ?").run(runId);
|
||||
connection.db.prepare("insert into run_actions(id, org_id, run_id, task_id, tool_name, arguments_excerpt, result_excerpt, created_at) values (?, ?, ?, ?, 'write_deliverable', '{}', 'Wrote src/app.ts', ?)")
|
||||
.run(`rac_${runId}`, orgId, runId, taskId, new Date().toISOString());
|
||||
return true;
|
||||
};
|
||||
let sessions = 0;
|
||||
while (productiveSession()) sessions += 1;
|
||||
expect(sessions).toBe(12);
|
||||
expect(connection.db.prepare("select stalled_reason from tasks where id = ?").get(taskId))
|
||||
.toEqual({ stalled_reason: "Stalled: 12 sessions ran without reaching done. Edit, reassign or reopen the task to give it a fresh budget." });
|
||||
connection.close();
|
||||
});
|
||||
|
||||
test("an open task with a completed run keeps getting sessions until done or the budget is spent", () => {
|
||||
const connection = createConnection();
|
||||
const organization = new OrganizationService(connection.db);
|
||||
|
||||
Reference in new issue
Block a user