Nan Gao 4fcb4bc366
feat(subagents): persist and display subagent step history (#3779) (#3845)
* feat(subagents): persist and display subagent step history (#3779)

Capture both assistant turns and tool outputs during subagent execution,
stream them in task_running events, and persist them as subagent.* run
events so the subtask card's step timeline survives a reload.

Backend:
- step_events.py: pure layer (capture_step_message, build_subagent_step,
  subagent_run_event) shared by streaming and persistence
- executor.py: capture ToolMessage outputs, not just AIMessage turns
- worker.py: persist task_* custom events to RunEventStore (category
  "subagent" keeps them out of the thread feed; list_events backfills)

Frontend:
- core/tasks/steps.ts + api.ts: SubtaskStep model, messageToStep,
  eventsToSteps, mergeSteps, fetchSubtaskSteps
- subtask card accumulates live steps and backfills on expand
- carry run_id onto history content messages for the events endpoint

* fix(subagents): show AI turns in subtask card + paginate step backfill (#3779)

Two follow-ups to the subagent step-history feature:

Problem 1 — reload backfill could silently truncate the step timeline because
list_events capped at 500 events (seq-ASC) across the whole run. Add task_id
filtering + an after_seq forward cursor to list_events (all three stores +
abstract base + the /events route), and make fetchSubtaskSteps page through one
task's subagent.step events until a short page. No schema migration: the DB
filter rides the existing run-scoped index via event_metadata["task_id"].

Problem 2 — the card only rendered tool steps, so persisted AI turns were never
shown. Replace toolStepsForDisplay with stepsForDisplay: interleave AI reasoning
turns (with text) and tool steps by message_index, drop blank-text AI turns, and
drop the trailing final-answer AI turn when completed (already shown as result).
Card renders AI steps as muted clamped markdown with a sparkles icon.

Tests: store task_id/after_seq filtering + pagination across memory/db/jsonl,
the /events route forwarding, stepsForDisplay rules, and fetchSubtaskSteps
pagination. Docs updated in both AGENTS.md.

* make format

* fix(subagents): capture full multi-tool step tail, batch step persistence, cap tool-call args (#3779)

Address PR review findings on the subagent step-history feature:

1. executor.py streamed on stream_mode="values" and captured only
   messages[-1] per chunk, so a multi-tool-call turn (ToolNode appends
   one ToolMessage per call in a single super-step) lost all but the last
   tool output in both the live task_running stream and the persisted
   history. Replace with capture_new_step_messages, which walks the
   newly-appended tail (and still re-checks the trailing message on
   no-growth chunks so id-less in-place replacements survive).

2. worker.py persisted each step with the store's low-frequency put()
   (a per-thread advisory lock per call); a deep subagent (max_turns=150)
   emits hundreds of steps on the hot stream loop. Replace with
   _SubagentEventBuffer, which batches via put_batch (flush on terminal
   subagent.end, at FLUSH_THRESHOLD, and in the worker finally).

3. build_subagent_step capped only text; tool_calls[].args were copied
   verbatim, so a large write_file/bash payload produced an unbounded
   subagent.step row. Cap each call's serialized args at
   SUBAGENT_STEP_MAX_CHARS, flagged args_truncated.

Tests updated/added for all three; AGENTS.md refreshed.

* fix(subagents): merge backfill into latest subtask state; reuse message_content_to_text (#3779)

Address the remaining two PR review findings:

4. subtask-card's fetchSubtaskSteps().then(updateSubtask) closed over a
   stale tasks snapshot: a late-resolving backfill wrote setTasks({...stale}),
   clobbering SSE steps/status and sibling subtasks that arrived during the
   fetch. useUpdateSubtask now reads/writes through a tasksRef mirroring the
   latest state (ref-to-latest), and the pure per-subtask transition is
   extracted to core/tasks/subtask-update.ts::computeNextSubtask (unit-tested).

5. step_events._content_to_text duplicated deerflow.utils.messages.
   message_content_to_text; call the shared helper instead (guarding None
   content with 'or ""' so a tool-call-only turn still renders as "").

Tests added for computeNextSubtask and the None-content case; AGENTS.md docs updated.
2026-07-02 07:43:09 +08:00

221 lines
6.1 KiB
TypeScript

import { describe, expect, it } from "@rstest/core";
import {
eventsToSteps,
mergeSteps,
messageToStep,
stepsForDisplay,
} from "@/core/tasks/steps";
describe("messageToStep", () => {
it("normalizes an AI message into an ai step with tool calls", () => {
const step = messageToStep(
{
type: "ai",
id: "ai-1",
content: "Let me search.",
tool_calls: [{ name: "web_search", args: { query: "x" }, id: "c1" }],
},
1,
);
expect(step.kind).toBe("ai");
expect(step.message_index).toBe(1);
expect(step.text).toBe("Let me search.");
expect(step.tool_calls).toEqual([
{ name: "web_search", args: { query: "x" } },
]);
expect(step.tool_name).toBeUndefined();
});
it("normalizes a tool message into a tool step with its output", () => {
const step = messageToStep(
{ type: "tool", id: "t-1", name: "web_search", content: "results" },
2,
);
expect(step.kind).toBe("tool");
expect(step.tool_name).toBe("web_search");
expect(step.text).toBe("results");
expect(step.tool_calls).toBeUndefined();
});
it("flattens list-of-blocks content to text", () => {
const step = messageToStep(
{
type: "ai",
content: [
{ type: "text", text: "first" },
{ type: "text", text: "second" },
],
},
1,
);
expect(step.text).toContain("first");
expect(step.text).toContain("second");
});
});
describe("mergeSteps", () => {
it("appends a new step", () => {
const a = messageToStep({ type: "ai", content: "a" }, 1);
const b = messageToStep({ type: "tool", name: "x", content: "b" }, 2);
expect(mergeSteps([a], [b])).toEqual([a, b]);
});
it("dedupes by message_index, preferring the incoming step", () => {
const old = messageToStep({ type: "ai", content: "old" }, 1);
const fresh = messageToStep({ type: "ai", content: "fresh" }, 1);
const merged = mergeSteps([old], [fresh]);
expect(merged).toHaveLength(1);
expect(merged[0]!.text).toBe("fresh");
});
it("keeps steps ordered by message_index", () => {
const s1 = messageToStep({ type: "ai", content: "1" }, 1);
const s2 = messageToStep({ type: "ai", content: "2" }, 2);
const s3 = messageToStep({ type: "ai", content: "3" }, 3);
const merged = mergeSteps([s3], [s1, s2]);
expect(merged.map((s) => s.message_index)).toEqual([1, 2, 3]);
});
});
describe("stepsForDisplay", () => {
it("keeps tool steps and AI steps that have text, ordered by message_index", () => {
const steps = [
messageToStep(
{ type: "tool", name: "web_search", content: "big result body" },
2,
),
messageToStep(
{
type: "ai",
content: "Let me search",
tool_calls: [{ name: "web_search", args: {} }],
},
1,
),
];
const display = stepsForDisplay(steps, "in_progress");
expect(display.map((s) => s.message_index)).toEqual([1, 2]);
expect(display.map((s) => s.kind)).toEqual(["ai", "tool"]);
});
it("drops AI steps with blank text even if they have tool_calls", () => {
const steps = [
messageToStep(
{
type: "ai",
content: " ",
tool_calls: [{ name: "web_search", args: {} }],
},
1,
),
messageToStep({ type: "tool", name: "read_file", content: "x" }, 2),
];
expect(
stepsForDisplay(steps, "in_progress").map((s) => s.message_index),
).toEqual([2]);
});
it("drops the trailing final AI answer when completed (already shown as result)", () => {
const steps = [
messageToStep({ type: "tool", name: "web_search", content: "x" }, 1),
messageToStep({ type: "ai", content: "The final answer is 42." }, 2),
];
expect(stepsForDisplay(steps, "completed").map((s) => s.kind)).toEqual([
"tool",
]);
});
it("keeps the trailing AI step while still in progress", () => {
const steps = [
messageToStep({ type: "tool", name: "web_search", content: "x" }, 1),
messageToStep({ type: "ai", content: "Thinking about the answer..." }, 2),
];
expect(stepsForDisplay(steps, "in_progress").map((s) => s.kind)).toEqual([
"tool",
"ai",
]);
});
it("returns empty for undefined", () => {
expect(stepsForDisplay(undefined, "in_progress")).toEqual([]);
});
});
describe("eventsToSteps", () => {
const events = [
{
event_type: "subagent.start",
content: { task_id: "call_1", description: "research" },
metadata: { task_id: "call_1" },
},
{
event_type: "subagent.step",
content: {
task_id: "call_1",
message_index: 2,
kind: "tool",
tool_name: "web_search",
text: "results",
truncated: false,
},
metadata: { task_id: "call_1", message_index: 2 },
},
{
event_type: "subagent.step",
content: {
task_id: "call_1",
message_index: 1,
kind: "ai",
text: "searching",
tool_calls: [{ name: "web_search", args: {} }],
},
metadata: { task_id: "call_1", message_index: 1 },
},
{
event_type: "subagent.step",
content: { task_id: "other", message_index: 1, kind: "ai", text: "nope" },
metadata: { task_id: "other", message_index: 1 },
},
{
event_type: "subagent.end",
content: { task_id: "call_1", status: "completed", result: "done" },
metadata: { task_id: "call_1" },
},
];
it("maps subagent.step events for the task into ordered steps", () => {
const steps = eventsToSteps(events, "call_1");
expect(steps.map((s) => s.message_index)).toEqual([1, 2]);
expect(steps[0]!.kind).toBe("ai");
expect(steps[1]!.kind).toBe("tool");
expect(steps[1]!.tool_name).toBe("web_search");
});
it("ignores steps belonging to other tasks and non-step events", () => {
const steps = eventsToSteps(events, "call_1");
expect(steps.every((s) => s.text !== "nope")).toBe(true);
expect(steps).toHaveLength(2);
});
it("returns empty array when no events match", () => {
expect(eventsToSteps(events, "missing")).toEqual([]);
expect(eventsToSteps([], "call_1")).toEqual([]);
});
});