Q6: feed messages that arrive mid-run into the next model turn
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
46dd77ddda
commit
0fb4ca9ae6
11 files changed
+161
-11
No files matched your search
@@ -19,10 +19,15 @@ const testVault = {
|
||||
|
||||
async function founderLine() {
|
||||
const wakePrompts: string[] = [];
|
||||
const wakeInputs: ProviderToolTurnInput["messages"][] = [];
|
||||
let onWake: (turnIndex: number) => void = () => undefined;
|
||||
const turn = async (input: ProviderToolTurnInput): Promise<AgentTurnResult> => {
|
||||
const first = input.messages.find((message) => message.role === "user");
|
||||
const prompt = first && first.role === "user" ? first.text : "";
|
||||
if (/thread_id="/.test(prompt) && input.messages.length === 1) wakePrompts.push(prompt);
|
||||
if (!/thread_id="/.test(prompt)) return { text: "Kickoff noted.", toolCalls: [], rawResponseJson: "{}" };
|
||||
if (input.messages.length === 1) wakePrompts.push(prompt);
|
||||
wakeInputs.push(input.messages);
|
||||
onWake(wakeInputs.length);
|
||||
return { text: `Reply ${wakePrompts.length}.`, toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-chatctx-${randomUUID()}.db`));
|
||||
@@ -60,7 +65,8 @@ async function founderLine() {
|
||||
const row = connection.db.prepare("select id from chat_messages where body_md = ?").get(bodyMd) as { id: string };
|
||||
chat.editMessage({ orgId, messageId: row.id, bodyMd: newBody, actorType: "employee", actorId: founderId });
|
||||
};
|
||||
return { connection, wakePrompts, say, runWake, edit };
|
||||
const onWakeTurn = (hook: (turnIndex: number) => void) => { onWake = hook; };
|
||||
return { connection, wakePrompts, wakeInputs, say, runWake, edit, onWakeTurn };
|
||||
}
|
||||
|
||||
describe("chat wake context", () => {
|
||||
@@ -99,4 +105,72 @@ describe("chat wake context", () => {
|
||||
expect(fresh).not.toContain("Reply 1.");
|
||||
h.connection.close();
|
||||
});
|
||||
|
||||
test("a message that arrives mid-run reaches the next model turn", async () => {
|
||||
const connection = new DatabaseConnection(join(tmpdir(), `ac-midrun-${randomUUID()}.db`));
|
||||
const chat = new ChatService(connection.db);
|
||||
const seenFollowUps: string[] = [];
|
||||
let founderId = "";
|
||||
let orgId = "";
|
||||
let spaceId = "";
|
||||
let threadId = "";
|
||||
let wakeTurns = 0;
|
||||
const turn = async (input: ProviderToolTurnInput): Promise<AgentTurnResult> => {
|
||||
const first = input.messages.find((message) => message.role === "user");
|
||||
if (!(first && first.role === "user" && /thread_id="/.test(first.text))) return { text: "Kickoff noted.", toolCalls: [], rawResponseJson: "{}" };
|
||||
wakeTurns += 1;
|
||||
const last = input.messages[input.messages.length - 1];
|
||||
if (last.role === "tool" && last.followUp) seenFollowUps.push(last.followUp);
|
||||
if (wakeTurns === 1) {
|
||||
chat.sendMessage({ orgId, spaceId, threadId, senderType: "employee", senderId: founderId, bodyMd: "Stop — use Postgres instead.", attachmentFileIds: [] });
|
||||
return { text: "Looking.", toolCalls: [{ id: "c1", name: "read_messages", arguments: {} }], rawResponseJson: "{}" };
|
||||
}
|
||||
return { text: "Switching to Postgres.", toolCalls: [], rawResponseJson: "{}" };
|
||||
};
|
||||
const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
|
||||
const onboarding = new OnboardingService({ db: connection.db, saveProviderCredential: (input) => providers.saveCredential(input) });
|
||||
const execution = new TaskExecutionService(connection.db, providers);
|
||||
let ceoId = "";
|
||||
connection.transaction(() => {
|
||||
const summary = 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 = summary.orgId;
|
||||
ceoId = summary.ceoAgentId;
|
||||
});
|
||||
founderId = (connection.db.prepare("select id from employees where org_id = ?").get(orgId) as { id: string }).id;
|
||||
spaceId = ensureCeoDirectSpace(connection.db, orgId)!;
|
||||
const kickoff = claimNextQueuedRun(connection);
|
||||
await execution.executeRun(kickoff!.org_id, kickoff!.id);
|
||||
const firstId = chat.sendMessage({ orgId, spaceId, threadId: null, senderType: "employee", senderId: founderId, bodyMd: "Use MySQL.", attachmentFileIds: [] });
|
||||
threadId = (connection.db.prepare("select thread_id from chat_messages where id = ?").get(firstId) as { thread_id: string }).thread_id;
|
||||
const wake = connection.db.prepare(
|
||||
"select id from task_runs where org_id = ? and agent_id = ? and trigger = 'chat_mention' and state = 'queued'",
|
||||
).get(orgId, ceoId) as { id: string };
|
||||
await execution.executeRun(orgId, wake.id);
|
||||
|
||||
expect(seenFollowUps).toHaveLength(1);
|
||||
expect(seenFollowUps[0]).toContain("Messages that arrived while you were working");
|
||||
expect(seenFollowUps[0]).toContain("Stop — use Postgres instead.");
|
||||
expect(seenFollowUps[0]).not.toContain("Use MySQL.");
|
||||
connection.close();
|
||||
});
|
||||
|
||||
test("a correction that lands while the model answers gets one more turn", async () => {
|
||||
const h = await founderLine();
|
||||
const threadId = h.say("Use MySQL.", null);
|
||||
h.onWakeTurn((turnIndex) => {
|
||||
if (turnIndex === 1) h.say("Stop — use Postgres instead.", threadId);
|
||||
});
|
||||
await h.runWake();
|
||||
const second = h.wakeInputs[1];
|
||||
expect(second.at(-2)).toMatchObject({ role: "assistant", text: "Reply 1." });
|
||||
const last = second.at(-1)!;
|
||||
expect(last.role).toBe("user");
|
||||
expect(last.role === "user" && last.text).toContain("Stop — use Postgres instead.");
|
||||
h.connection.close();
|
||||
});
|
||||
});
|
||||
Reference in new issue
Block a user