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.

Name the Bot on a hop that was refused, retried or lost, not only on one that landed (#306)

kevinb85f361f8e75
diff --git a/CHANGELOG.md b/CHANGELOG.md
index cd564de4..b4876146 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,14 @@ Newest first. `Unreleased` is what is on `main` and not yet tagged.
## Unreleased
+### A hop the boundary refused now names the Bot that was refused
+
+The audit page renders its Bot column from `payload.bot` and nothing else. `agent.handoff_offered`
+and `agent.handoff_delivered` were given that key; the four rows either side of them — a hop
+refused, a hop retried, and a hop that failed for good — were not, so they showed a dash where the
+Bot belongs. Those are the rows somebody actually opens the trail for: a hop that happened is visible
+in the transcript anyway, and a refused or lost one is visible nowhere else. All four now name the
+asking Bot, exactly as the accepted pair and `agent.escalated` already did.
### A failed tool refresh no longer leaves a connector offering nothing
Refreshing a connector's tools replaced the list with a delete and then an insert, as two separate
diff --git a/server/src/agents/handoff-runner.ts b/server/src/agents/handoff-runner.ts
index 5b7ab8db..c59a3bc7 100644
--- a/server/src/agents/handoff-runner.ts
+++ b/server/src/agents/handoff-runner.ts
@@ -300,6 +300,9 @@ export function createHandoffRunner(options: {
targetId: work.toBotId,
...(work.actorId ? { actorUserId: work.actorId } : {}),
payload: {
+ // See the same key on `agent.handoff_delivered` below: the Audit screen's Bot
+ // column reads `payload.bot`, so a row without it names no Bot.
+ bot: work.fromBotId,
from: work.fromBotId,
to: work.toBotId,
run: work.runId,
@@ -374,6 +377,8 @@ export function createHandoffRunner(options: {
targetId: work.toBotId,
...(work.actorId ? { actorUserId: work.actorId } : {}),
payload: {
+ // See the same key on `agent.handoff_delivered` below.
+ bot: work.fromBotId,
from: work.fromBotId,
to: work.toBotId,
run: work.runId,
@@ -456,6 +461,10 @@ export function createHandoffRunner(options: {
targetId: work.toBotId,
...(work.actorId ? { actorUserId: work.actorId } : {}),
payload: {
+ // See the same key on `agent.handoff_delivered` above. This row is the one a
+ // person's unanswered question ends on, so a Bot column showing a dash on it is
+ // the worst place in the set to have one.
+ bot: work.fromBotId,
from: work.fromBotId,
to: work.toBotId,
run: work.runId,
diff --git a/server/src/agents/handoff.ts b/server/src/agents/handoff.ts
index 0e23818c..58c32848 100644
--- a/server/src/agents/handoff.ts
+++ b/server/src/agents/handoff.ts
@@ -110,6 +110,12 @@ export function createHandoffDesk(options: {
targetId: from.botId,
...(from.actorId ? { actorUserId: from.actorId } : {}),
payload: {
+ // The same key `agent.handoff_offered` sets below, and for the same reason: the Audit
+ // screen renders `payload.bot` and nothing else in its Bot column, so a row without it
+ // shows a dash. The accepted row was given this and its refusal was not, which left the
+ // refusal — the one the trail says matters more, because a hop that happened is visible in
+ // the transcript and a refused one is invisible everywhere else — naming no Bot at all.
+ bot: from.botId,
from: from.botId,
// As the model named it, capped: untrusted input, kept because "who did it reach for" is the
// useful half of the question.
diff --git a/server/tests/agent-handoff-runner.test.ts b/server/tests/agent-handoff-runner.test.ts
index 61c5c56e..7f8a8f1f 100644
--- a/server/tests/agent-handoff-runner.test.ts
+++ b/server/tests/agent-handoff-runner.test.ts
@@ -199,6 +199,40 @@ describe("delivering a hop", () => {
});
});
+ /*
+ * And so do the rows either side of it, which is the half the assertion above did not reach.
+ *
+ * A delivery is the outcome that is also visible in the transcript. A hop that was retried or
+ * that failed is visible nowhere else at all, so those are the rows somebody actually comes to
+ * this screen for — and they were the ones rendering a dash where the Bot's name belongs.
+ */
+ test("a hop that was retried or that failed names the Bot too", async () => {
+ const retried = runner({
+ claimed: [
+ { kind: "bot.message", key: "run-1:abc", payload: WORK, attempts: 2 },
+ ],
+ });
+ await retried.runner.sweep();
+
+ expect(
+ retried.written.find(
+ (event) => event.eventType === "agent.handoff_retried",
+ )?.payload,
+ ).toMatchObject({ bot: WORK.fromBotId, from: WORK.fromBotId });
+
+ const failed = runner({
+ deliver: async () => {
+ throw new Error("the gateway was unreachable");
+ },
+ });
+ await failed.runner.sweep();
+
+ expect(
+ failed.written.find((event) => event.eventType === "agent.handoff_failed")
+ ?.payload,
+ ).toMatchObject({ bot: WORK.fromBotId, from: WORK.fromBotId });
+ });
+
/* Releasing an unusable row would put it back on the queue for ever. */
test("a row that is not a hop is finished rather than released", async () => {
const { runner: sweep, calls } = runner({
diff --git a/server/tests/agent-handoff.test.ts b/server/tests/agent-handoff.test.ts
index da3abae6..748e232b 100644
--- a/server/tests/agent-handoff.test.ts
+++ b/server/tests/agent-handoff.test.ts
@@ -338,8 +338,57 @@ describe("handing work to another Bot", () => {
expect(refused.events[0]?.payload).toMatchObject({
reason: "not_granted",
run: "run-1",
+ /*
+ * And the refusal names the Bot too, which is the half this was missing.
+ *
+ * The accepted row above was given `bot` and its refusal was not, so the pair the trail calls
+ * "both outcomes" rendered one Bot and one dash. On the row the notes call the more important
+ * of the two: a hop that happened is visible in the transcript, and a refused one is
+ * invisible everywhere except here.
+ */
+ bot: "assistant",
});
});
+
+ /*
+ * Every way a hop can be refused, not only the one the pair above happens to use.
+ *
+ * `refuse` is one function and all five reasons go through it, so this could not drift per reason
+ * — but that is the argument for asserting it once across all of them rather than trusting it.
+ */
+ test("every refusal names the Bot that was refused", async () => {
+ const cases: Array<[string, ReturnType<typeof desk>]> = [
+ ["no_task", desk()],
+ ["not_granted", desk({ granted: false })],
+ ["unknown_bot", desk()],
+ ["depth", desk({ caps: { maxDepth: 0, maxPerRun: 3 } })],
+ ["fan_out", desk({ caps: { maxDepth: 2, maxPerRun: 0 } })],
+ ];
+ const envelopes: Record<string, { target: string; task: string }> = {
+ no_task: { target: "researcher", task: "" },
+ not_granted: { target: "researcher", task: "t" },
+ unknown_bot: { target: "nobody-by-that-name", task: "t" },
+ depth: { target: "researcher", task: "t" },
+ fan_out: { target: "researcher", task: "t" },
+ };
+
+ for (const [name, harness] of cases) {
+ const envelope = envelopes[name] as { target: string; task: string };
+ const outcome = await harness.desk.send({
+ from: FROM,
+ target: envelope.target,
+ envelope: { task: envelope.task },
+ });
+
+ expect(outcome.ok).toBe(false);
+ expect(harness.events.map((event) => event.eventType)).toEqual([
+ "agent.handoff_refused",
+ ]);
+ // The asking Bot, the same one `agent.handoff_offered` records, so the two rows of a pair
+ // read as one Bot's two possible outcomes rather than as one Bot and a dash.
+ expect(harness.events[0]?.payload).toMatchObject({ bot: "assistant" });
+ }
+ });
});
/*