Skip to content

Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

Refuse a blank question put to a person, the way a blank task handed to a Bot is refused (#476)

kevina0a745ce6e09
diff --git a/CHANGELOG.md b/CHANGELOG.md
index b1814703..73e29559 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,17 @@ Newest first. `Unreleased` is what is on `main` and not yet tagged.
## Unreleased
+### A Bot cannot end its turn by asking a person nothing
+
+`ask_person` is how a Bot stops and puts something to a person instead of guessing, and a call with
+no question in it was already meant to come back as a sentence telling it to say what it needs. That
+only happened when the field was missing altogether. A question that was present and empty was
+carried out: the Bot was told its question had been put to somebody, its turn ended there, and the
+trail took an escalation row with nothing in its question — the row an administrator counts these by,
+saying a person was asked something that was never said. On a deployment whose escalation route is a
+duty desk rather than the person already in the conversation, it is a page to somebody with no
+question on it. A blank question is now refused with the sentence that was already written for it,
+and a question typed with room around it is recorded as the question rather than as the spacing.
### A routine scheduled for Sunday says Sundays, whichever number it was written with
Crontab has always let Sunday be either 0 or 7, the scheduler here takes both, and a routine written
diff --git a/server/src/agents/escalation.ts b/server/src/agents/escalation.ts
index 017c27d7..590071bc 100644
--- a/server/src/agents/escalation.ts
+++ b/server/src/agents/escalation.ts
@@ -102,12 +102,31 @@ export function escalationTool(options: {
return "That was not put to anybody: say what you need a person to answer.";
}
+ /*
+ * A question field that is present and empty is a call with nothing in it.
+ *
+ * `z.string()` accepts "" and a run of spaces, so the refusal above — which is the sentence
+ * written for exactly this — only ever fired when the field was missing altogether. Spelled
+ * the other way it went straight through: the Bot was told its question had been put to
+ * somebody, the turn ended on that, and the trail took an `agent.escalated` row with nothing
+ * in its question, which is the row an operator counts escalations by. Where a route is a duty
+ * desk rather than the person already here, it is a page to somebody with no question on it.
+ *
+ * `message_bot`, which this competes with for the same decision, refuses a blank task and says
+ * so. This is the other half of that, and the trimmed text is what travels, for the same
+ * reason a handoff sends the trimmed task: what was recorded should be what was asked.
+ */
+ const question = parsed.data.question.trim();
+ if (!question) {
+ return "That was not put to anybody: say what you need a person to answer.";
+ }
+
const outcome = await route({
actorId: from.actorId,
botId: from.botId,
...(from.threadId ? { threadId: from.threadId } : {}),
runId: from.runId,
- question: parsed.data.question,
+ question,
...(parsed.data.why ? { why: parsed.data.why } : {}),
});
@@ -128,7 +147,7 @@ export function escalationTool(options: {
payload: {
bot: from.botId,
run: from.runId,
- question: parsed.data.question,
+ question,
...(parsed.data.why ? { why: parsed.data.why } : {}),
...("reached" in outcome
? { reached: outcome.reached }
diff --git a/server/tests/agent-escalation.test.ts b/server/tests/agent-escalation.test.ts
index d1c92eef..80ecaae6 100644
--- a/server/tests/agent-escalation.test.ts
+++ b/server/tests/agent-escalation.test.ts
@@ -128,6 +128,58 @@ describe("asking a person", () => {
expect(said).toContain("say what you need");
});
+
+ /*
+ * The same call, spelled the other way.
+ *
+ * A question field that is present and empty is a call with nothing in it too, and the refusal
+ * above is the sentence written for it. `message_bot`, the tool this competes with for the same
+ * decision, refuses a blank task and says so; this is the other half of that.
+ */
+ test.each([
+ ["an empty question", ""],
+ ["a question of spaces", " "],
+ ])("%s is refused as a sentence", async (_name, question) => {
+ const tool = escalationTool({ from: FROM, route: askTheirOwnPerson });
+
+ const said = await tool.execute({ question });
+
+ expect(said).toContain("say what you need");
+ expect(said).not.toContain(PUT_TO);
+ });
+
+ test("a blank question reaches nobody and leaves no row saying it did", async () => {
+ const { written, store } = recorder();
+ let reached = 0;
+ const tool = escalationTool({
+ from: FROM,
+ route: async () => {
+ reached += 1;
+ return { reached: "the on-call engineer" };
+ },
+ auditStore: store,
+ });
+
+ await tool.execute({ question: " " });
+
+ // An `agent.escalated` row with no question in it is the row an operator counts escalations by,
+ // saying a person was asked something that was never said.
+ expect(reached).toBe(0);
+ expect(written).toEqual([]);
+ });
+
+ test("a question with room around it is recorded as the question", async () => {
+ const { written, store } = recorder();
+ const tool = escalationTool({
+ from: FROM,
+ route: askTheirOwnPerson,
+ auditStore: store,
+ });
+
+ await tool.execute({ question: " which account? " });
+
+ expect(written[0]?.payload).toMatchObject({ question: "which account?" });
+ });
});
/*