fix(odin): dedupe tool-trace chips — one per tool, warning only if all attempts failed
A failed call followed by a successful retry showed two identical chips (one orange). Retries are loop mechanics: the trace now carries one entry per tool with ok = delivered-at-least-once, for live chips and persisted history alike (also keeps meta.tools under its 30-entry cap). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
@@ -534,6 +534,55 @@ describe("agenticChat loop (SDK mocked)", () => {
|
|||||||
expect(usageRows.length).toBe(2);
|
expect(usageRows.length).toBe(2);
|
||||||
});
|
});
|
||||||
|
|
||||||
|
it("dedupes the tool trace: failed attempt + successful retry = one ok chip", async () => {
|
||||||
|
const self: AiAuthCtx = {
|
||||||
|
userId: 999999996,
|
||||||
|
roleName: "viewer",
|
||||||
|
permissions: ["attendance.record"],
|
||||||
|
};
|
||||||
|
createMock.mockReset();
|
||||||
|
createMock
|
||||||
|
.mockResolvedValueOnce({
|
||||||
|
stop_reason: "tool_use",
|
||||||
|
content: [
|
||||||
|
{
|
||||||
|
type: "tool_use",
|
||||||
|
id: "toolu_a",
|
||||||
|
name: "get_attendance_summary",
|
||||||
|
input: { user_id: 1 }, // denied: someone else without manage
|
||||||
|
},
|
||||||
|
],
|
||||||
|
usage: { input_tokens: 10, output_tokens: 5 },
|
||||||
|
})
|
||||||
|
.mockResolvedValueOnce({
|
||||||
|
stop_reason: "tool_use",
|
||||||
|
content: [
|
||||||
|
{
|
||||||
|
type: "tool_use",
|
||||||
|
id: "toolu_b",
|
||||||
|
name: "get_attendance_summary",
|
||||||
|
input: {}, // retry: own data, succeeds
|
||||||
|
},
|
||||||
|
],
|
||||||
|
usage: { input_tokens: 10, output_tokens: 5 },
|
||||||
|
})
|
||||||
|
.mockResolvedValueOnce({
|
||||||
|
stop_reason: "end_turn",
|
||||||
|
content: [{ type: "text", text: "Hotovo." }],
|
||||||
|
usage: { input_tokens: 10, output_tokens: 5 },
|
||||||
|
});
|
||||||
|
const res = await agenticChat(
|
||||||
|
[{ role: "user", content: "Moje docházka?" }],
|
||||||
|
self,
|
||||||
|
);
|
||||||
|
expect(res.toolTrace).toEqual([
|
||||||
|
{ name: "get_attendance_summary", label: "Docházka", ok: true },
|
||||||
|
]);
|
||||||
|
await prisma.ai_usage.deleteMany({
|
||||||
|
where: { user_id: self.userId, kind: "agent" },
|
||||||
|
});
|
||||||
|
});
|
||||||
|
|
||||||
it("injects the caller's identity into the system prompt", async () => {
|
it("injects the caller's identity into the system prompt", async () => {
|
||||||
createMock.mockReset();
|
createMock.mockReset();
|
||||||
createMock.mockResolvedValueOnce({
|
createMock.mockResolvedValueOnce({
|
||||||
|
|||||||
@@ -381,6 +381,17 @@ export async function agenticChat(
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// One chip per tool: a failed attempt followed by a successful retry is
|
||||||
|
// loop mechanics, not information for the user. ok = the tool delivered
|
||||||
|
// data at least once; orange stays only when every attempt failed. Also
|
||||||
|
// keeps the persisted meta.tools comfortably under its 30-entry cap.
|
||||||
|
const dedupedTrace: ToolTraceEntry[] = [];
|
||||||
|
for (const t of trace) {
|
||||||
|
const seen = dedupedTrace.find((d) => d.name === t.name);
|
||||||
|
if (!seen) dedupedTrace.push({ ...t });
|
||||||
|
else seen.ok = seen.ok || t.ok;
|
||||||
|
}
|
||||||
|
|
||||||
const text = (res?.content ?? [])
|
const text = (res?.content ?? [])
|
||||||
.filter((b): b is Anthropic.TextBlock => b.type === "text")
|
.filter((b): b is Anthropic.TextBlock => b.type === "text")
|
||||||
.map((b) => b.text)
|
.map((b) => b.text)
|
||||||
@@ -399,7 +410,7 @@ export async function agenticChat(
|
|||||||
} else if (!reply) {
|
} else if (!reply) {
|
||||||
reply = "Nepodařilo se získat odpověď, zkuste to prosím znovu.";
|
reply = "Nepodařilo se získat odpověď, zkuste to prosím znovu.";
|
||||||
}
|
}
|
||||||
return { reply, toolTrace: trace };
|
return { reply, toolTrace: dedupedTrace };
|
||||||
}
|
}
|
||||||
|
|
||||||
export interface ExtractedInvoice {
|
export interface ExtractedInvoice {
|
||||||
|
|||||||
Reference in New Issue
Block a user