mirror of
https://github.com/fabro-sh/fabro.git
synced 2026-10-08 03:10:26 +00:00
Merge remote-tracking branch 'origin/main' into feat/stage-chat-view
# Conflicts: # apps/fabro-web/app/routes/run-stages.test.ts # apps/fabro-web/app/routes/run-stages.tsx
This commit is contained in:
Bryan Helmkamp
commit
e739f86f6a
7 changed files with 1064 additions and 259 deletions
|
|
@ -443,7 +443,10 @@ const THREAD_CATEGORY_COLOR: Record<ThreadCategory, string> = {
|
|||
|
||||
export type ThreadDnaSelection =
|
||||
| { kind: "single"; turnIndex: number }
|
||||
| { kind: "group"; childTurnIndices: number[] };
|
||||
| {
|
||||
kind: "group";
|
||||
childTurnIndices: readonly [number, number, ...number[]];
|
||||
};
|
||||
|
||||
export interface ThreadDnaItem {
|
||||
category: ThreadCategory;
|
||||
|
|
@ -456,13 +459,15 @@ export interface ThreadDnaItem {
|
|||
const INSTANT_MARKER_PX = 4;
|
||||
const MIN_DURATION_PX = 3;
|
||||
|
||||
function selectionKey(s: ThreadDnaSelection): string {
|
||||
return s.kind === "single"
|
||||
? `s:${s.turnIndex}`
|
||||
: `g:${s.childTurnIndices.join(",")}`;
|
||||
export function threadSelectionId(selection: ThreadDnaSelection): number {
|
||||
const turnIndex =
|
||||
selection.kind === "single"
|
||||
? selection.turnIndex
|
||||
: selection.childTurnIndices[0];
|
||||
return turnIndex * 2 + (selection.kind === "group" ? 1 : 0);
|
||||
}
|
||||
|
||||
function selectionsEqual(
|
||||
export function threadSelectionsEqual(
|
||||
a: ThreadDnaSelection,
|
||||
b: ThreadDnaSelection | null,
|
||||
): boolean {
|
||||
|
|
@ -504,19 +509,19 @@ export function ThreadDnaStrip({
|
|||
selection: ThreadDnaSelection | null;
|
||||
onSelect: (s: ThreadDnaSelection) => void;
|
||||
}) {
|
||||
const [hover, setHover] = useState<{ key: string; rect: DOMRect } | null>(
|
||||
const [hover, setHover] = useState<{ id: number; rect: DOMRect } | null>(
|
||||
null,
|
||||
);
|
||||
const visibleItems = useMemo(
|
||||
() => sampleStripItems(items, STRIP_MAX_MARKERS, (item) =>
|
||||
selectionsEqual(item.selection, selection)
|
||||
threadSelectionsEqual(item.selection, selection)
|
||||
),
|
||||
[items, selection],
|
||||
);
|
||||
const visibleItemByKey = useMemo(
|
||||
const visibleItemById = useMemo(
|
||||
() =>
|
||||
new Map(
|
||||
visibleItems.map((item) => [selectionKey(item.selection), item]),
|
||||
visibleItems.map((item) => [threadSelectionId(item.selection), item]),
|
||||
),
|
||||
[visibleItems],
|
||||
);
|
||||
|
|
@ -542,7 +547,7 @@ export function ThreadDnaStrip({
|
|||
|
||||
const hoveredItem =
|
||||
hover != null
|
||||
? visibleItemByKey.get(hover.key) ?? null
|
||||
? visibleItemById.get(hover.id) ?? null
|
||||
: null;
|
||||
|
||||
return (
|
||||
|
|
@ -552,10 +557,10 @@ export function ThreadDnaStrip({
|
|||
>
|
||||
<div className="relative h-full">
|
||||
{visibleItems.map((item) => {
|
||||
const key = selectionKey(item.selection);
|
||||
const id = threadSelectionId(item.selection);
|
||||
const isInstant = item.durationMs <= 0;
|
||||
const isSelected = selectionsEqual(item.selection, selection);
|
||||
const isHovered = hover?.key === key;
|
||||
const isSelected = threadSelectionsEqual(item.selection, selection);
|
||||
const isHovered = hover?.id === id;
|
||||
const leftPct = (item.startMs / totalMs) * 100;
|
||||
const baseColor = THREAD_CATEGORY_COLOR[item.category];
|
||||
|
||||
|
|
@ -585,18 +590,18 @@ export function ThreadDnaStrip({
|
|||
|
||||
return (
|
||||
<button
|
||||
key={key}
|
||||
key={id}
|
||||
type="button"
|
||||
aria-label={`${THREAD_CATEGORY_LABEL[item.category]} · ${item.label}`}
|
||||
aria-pressed={isSelected}
|
||||
onMouseEnter={(e) =>
|
||||
setHover({
|
||||
key,
|
||||
id,
|
||||
rect: e.currentTarget.getBoundingClientRect(),
|
||||
})
|
||||
}
|
||||
onMouseLeave={() =>
|
||||
setHover((cur) => (cur?.key === key ? null : cur))
|
||||
setHover((cur) => (cur?.id === id ? null : cur))
|
||||
}
|
||||
onClick={() => onSelect(item.selection)}
|
||||
className="absolute cursor-pointer rounded-[2px] border-0 p-0 transition-all duration-100 ease-out"
|
||||
|
|
|
|||
|
|
@ -5,12 +5,21 @@ import {
|
|||
buildChatItems,
|
||||
buildStageActivity,
|
||||
buildThreadDnaItems,
|
||||
EVENT_KINDS,
|
||||
eventsTabLabel,
|
||||
eventsToActivity,
|
||||
filterDisplayItems,
|
||||
filterThreadDnaItems,
|
||||
formatStageModelUsageLabel,
|
||||
groupConsecutiveTools,
|
||||
searchableText,
|
||||
selectStageRenderer,
|
||||
turnSummary,
|
||||
visibleTurnCount,
|
||||
type DisplayItem,
|
||||
type EventKind,
|
||||
} from "./run-stages";
|
||||
import { threadSelectionId } from "../components/event-debug";
|
||||
|
||||
function envelope(seq: number, partial: Partial<EventEnvelope>): EventEnvelope {
|
||||
return {
|
||||
|
|
@ -42,6 +51,26 @@ function toolTurn(opts: {
|
|||
};
|
||||
}
|
||||
|
||||
function expectToolGroup(
|
||||
item: DisplayItem | undefined,
|
||||
): Extract<DisplayItem, { kind: "group" }> {
|
||||
expect(item?.kind).toBe("group");
|
||||
if (item?.kind !== "group") {
|
||||
throw new Error("expected a tool group");
|
||||
}
|
||||
return item;
|
||||
}
|
||||
|
||||
function expectSingleItem(
|
||||
item: DisplayItem | undefined,
|
||||
): Extract<DisplayItem, { kind: "single" }> {
|
||||
expect(item?.kind).toBe("single");
|
||||
if (item?.kind !== "single") {
|
||||
throw new Error("expected a single display item");
|
||||
}
|
||||
return item;
|
||||
}
|
||||
|
||||
describe("eventsToActivity", () => {
|
||||
test("filters events by stage_id (verify@1 vs verify@2 do not cross-contaminate)", () => {
|
||||
const events: EventEnvelope[] = [
|
||||
|
|
@ -73,25 +102,35 @@ describe("eventsToActivity", () => {
|
|||
|
||||
const firstVisit = eventsToActivity(events, "verify@1");
|
||||
expect(firstVisit).toEqual([
|
||||
{ kind: "system", ts: "2026-04-09T12:00:00Z", content: "first visit prompt" },
|
||||
{
|
||||
kind: "system",
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
content: "first visit prompt",
|
||||
},
|
||||
{
|
||||
kind: "assistant",
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
content: "first visit reply",
|
||||
inputTokens: 0,
|
||||
outputTokens: 0,
|
||||
toolCallCount: null,
|
||||
},
|
||||
]);
|
||||
|
||||
const secondVisit = eventsToActivity(events, "verify@2");
|
||||
expect(secondVisit).toEqual([
|
||||
{ kind: "system", ts: "2026-04-09T12:00:00Z", content: "second visit prompt" },
|
||||
{
|
||||
kind: "system",
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
content: "second visit prompt",
|
||||
},
|
||||
{
|
||||
kind: "assistant",
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
content: "second visit reply",
|
||||
inputTokens: 0,
|
||||
outputTokens: 0,
|
||||
toolCallCount: null,
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
|
@ -350,6 +389,7 @@ describe("eventsToActivity", () => {
|
|||
content: "Refactored auth module",
|
||||
inputTokens: 120,
|
||||
outputTokens: 30,
|
||||
toolCallCount: null,
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
|
@ -397,6 +437,7 @@ describe("eventsToActivity", () => {
|
|||
content: "Done.",
|
||||
inputTokens: 10,
|
||||
outputTokens: 5,
|
||||
toolCallCount: null,
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
|
@ -423,6 +464,7 @@ describe("eventsToActivity", () => {
|
|||
content: "All clear.",
|
||||
inputTokens: 0,
|
||||
outputTokens: 4,
|
||||
toolCallCount: null,
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
|
@ -484,7 +526,10 @@ describe("eventsToActivity", () => {
|
|||
describe("groupConsecutiveTools", () => {
|
||||
type Filtered = Parameters<typeof groupConsecutiveTools>[0];
|
||||
|
||||
function entry(turn: ReturnType<typeof toolTurn> | { kind: "system"; ts: string; content: string } | { kind: "assistant"; ts: string; content: string; inputTokens: number; outputTokens: number }, index: number): Filtered[number] {
|
||||
function entry(
|
||||
turn: Filtered[number]["turn"],
|
||||
index: number,
|
||||
): Filtered[number] {
|
||||
return { turn, index };
|
||||
}
|
||||
|
||||
|
|
@ -493,15 +538,32 @@ describe("groupConsecutiveTools", () => {
|
|||
});
|
||||
|
||||
test("single tool turn becomes a single, not a group", () => {
|
||||
const t = toolTurn({ ts: "2026-04-09T12:00:00Z", toolName: "shell", durationMs: 100 });
|
||||
const t = toolTurn({
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
toolName: "shell",
|
||||
durationMs: 100,
|
||||
});
|
||||
expect(groupConsecutiveTools([entry(t, 0)])).toEqual([
|
||||
{ kind: "single", turn: t, turnIndex: 0 },
|
||||
{
|
||||
kind: "single",
|
||||
turn: t,
|
||||
turnIndex: 0,
|
||||
selection: { kind: "single", turnIndex: 0 },
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
||||
test("two consecutive same-tool successes form a group of 2", () => {
|
||||
const a = toolTurn({ ts: "2026-04-09T12:00:00Z", toolName: "shell", durationMs: 1000 });
|
||||
const b = toolTurn({ ts: "2026-04-09T12:00:01Z", toolName: "shell", durationMs: 2000 });
|
||||
const a = toolTurn({
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
toolName: "shell",
|
||||
durationMs: 1000,
|
||||
});
|
||||
const b = toolTurn({
|
||||
ts: "2026-04-09T12:00:01Z",
|
||||
toolName: "shell",
|
||||
durationMs: 2000,
|
||||
});
|
||||
const result = groupConsecutiveTools([entry(a, 0), entry(b, 1)]);
|
||||
expect(result).toEqual([
|
||||
{
|
||||
|
|
@ -513,11 +575,12 @@ describe("groupConsecutiveTools", () => {
|
|||
{ turn: a, turnIndex: 0 },
|
||||
{ turn: b, turnIndex: 1 },
|
||||
],
|
||||
selection: { kind: "group", childTurnIndices: [0, 1] },
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
||||
test("five consecutive same-tool successes form one group; durations summed; ts is first", () => {
|
||||
test("five consecutive same-tool successes form one group spanning earliest start to latest end", () => {
|
||||
const turns = [0, 1, 2, 3, 4].map((i) =>
|
||||
toolTurn({
|
||||
ts: `2026-04-09T12:00:0${i}Z`,
|
||||
|
|
@ -528,21 +591,104 @@ describe("groupConsecutiveTools", () => {
|
|||
const filtered = turns.map((t, i) => entry(t, i));
|
||||
const result = groupConsecutiveTools(filtered);
|
||||
expect(result).toHaveLength(1);
|
||||
const item = result[0];
|
||||
expect(item.kind).toBe("group");
|
||||
if (item.kind === "group") {
|
||||
expect(item.ts).toBe("2026-04-09T12:00:00Z");
|
||||
expect(item.durationMs).toBe(15000);
|
||||
expect(item.children.map((c) => c.turnIndex)).toEqual([0, 1, 2, 3, 4]);
|
||||
}
|
||||
const item = expectToolGroup(result[0]);
|
||||
expect(item.ts).toBe("2026-04-09T12:00:00Z");
|
||||
// last child starts at 4s and runs 5s → ends at 9s. The summed 15s is
|
||||
// not elapsed time; overlapping calls would double-count.
|
||||
expect(item.durationMs).toBe(9000);
|
||||
expect(item.children.map((c) => c.turnIndex)).toEqual([0, 1, 2, 3, 4]);
|
||||
});
|
||||
|
||||
test("group bounds ignore array order and use the earliest start / latest end", () => {
|
||||
// Children listed in completion order: the second one started first and
|
||||
// the first one finished last.
|
||||
const late = toolTurn({
|
||||
ts: "2026-04-09T12:00:05Z",
|
||||
toolName: "shell",
|
||||
durationMs: 4000,
|
||||
});
|
||||
const early = toolTurn({
|
||||
ts: "2026-04-09T12:00:02Z",
|
||||
toolName: "shell",
|
||||
durationMs: 500,
|
||||
});
|
||||
const result = groupConsecutiveTools([entry(late, 0), entry(early, 1)]);
|
||||
expect(result).toHaveLength(1);
|
||||
const item = expectToolGroup(result[0]);
|
||||
expect(item.ts).toBe("2026-04-09T12:00:02Z");
|
||||
// earliest start 2s, latest end 5s + 4s = 9s → 7s elapsed.
|
||||
expect(item.durationMs).toBe(7000);
|
||||
});
|
||||
|
||||
test("parallel children collapse to their overlapping wall-clock span", () => {
|
||||
const a = toolTurn({
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
toolName: "shell",
|
||||
durationMs: 3000,
|
||||
});
|
||||
const b = toolTurn({
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
toolName: "shell",
|
||||
durationMs: 2000,
|
||||
});
|
||||
const c = toolTurn({
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
toolName: "shell",
|
||||
durationMs: 1000,
|
||||
});
|
||||
const result = groupConsecutiveTools([
|
||||
entry(a, 0),
|
||||
entry(b, 1),
|
||||
entry(c, 2),
|
||||
]);
|
||||
const item = expectToolGroup(result[0]);
|
||||
// Three calls issued together: elapsed is the slowest, not the sum.
|
||||
expect(item.durationMs).toBe(3000);
|
||||
});
|
||||
|
||||
test("a group of unparseable timestamps falls back to zero elapsed", () => {
|
||||
const a = toolTurn({
|
||||
ts: "not-a-timestamp",
|
||||
toolName: "shell",
|
||||
durationMs: 10,
|
||||
});
|
||||
const b = toolTurn({
|
||||
ts: "also-bad",
|
||||
toolName: "shell",
|
||||
durationMs: 10,
|
||||
});
|
||||
const result = groupConsecutiveTools([entry(a, 0), entry(b, 1)]);
|
||||
const item = expectToolGroup(result[0]);
|
||||
expect(item.ts).toBe("not-a-timestamp");
|
||||
expect(item.durationMs).toBe(0);
|
||||
});
|
||||
|
||||
test("a different tool between same-tool calls breaks the group boundary", () => {
|
||||
const a = toolTurn({ ts: "2026-04-09T12:00:00Z", toolName: "shell", durationMs: 1 });
|
||||
const b = toolTurn({ ts: "2026-04-09T12:00:01Z", toolName: "shell", durationMs: 1 });
|
||||
const c = toolTurn({ ts: "2026-04-09T12:00:02Z", toolName: "read_file", durationMs: 1 });
|
||||
const d = toolTurn({ ts: "2026-04-09T12:00:03Z", toolName: "shell", durationMs: 1 });
|
||||
const e = toolTurn({ ts: "2026-04-09T12:00:04Z", toolName: "shell", durationMs: 1 });
|
||||
const a = toolTurn({
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
toolName: "shell",
|
||||
durationMs: 1,
|
||||
});
|
||||
const b = toolTurn({
|
||||
ts: "2026-04-09T12:00:01Z",
|
||||
toolName: "shell",
|
||||
durationMs: 1,
|
||||
});
|
||||
const c = toolTurn({
|
||||
ts: "2026-04-09T12:00:02Z",
|
||||
toolName: "read_file",
|
||||
durationMs: 1,
|
||||
});
|
||||
const d = toolTurn({
|
||||
ts: "2026-04-09T12:00:03Z",
|
||||
toolName: "shell",
|
||||
durationMs: 1,
|
||||
});
|
||||
const e = toolTurn({
|
||||
ts: "2026-04-09T12:00:04Z",
|
||||
toolName: "shell",
|
||||
durationMs: 1,
|
||||
});
|
||||
const result = groupConsecutiveTools([
|
||||
entry(a, 0),
|
||||
entry(b, 1),
|
||||
|
|
@ -551,20 +697,22 @@ describe("groupConsecutiveTools", () => {
|
|||
entry(e, 4),
|
||||
]);
|
||||
expect(result.map((r) => r.kind)).toEqual(["group", "single", "group"]);
|
||||
if (result[0].kind === "group") {
|
||||
expect(result[0].children.map((c) => c.turnIndex)).toEqual([0, 1]);
|
||||
}
|
||||
if (result[1].kind === "single") {
|
||||
expect(result[1].turnIndex).toBe(2);
|
||||
}
|
||||
if (result[2].kind === "group") {
|
||||
expect(result[2].children.map((c) => c.turnIndex)).toEqual([3, 4]);
|
||||
}
|
||||
expect(expectToolGroup(result[0]).children.map((c) => c.turnIndex)).toEqual(
|
||||
[0, 1],
|
||||
);
|
||||
expect(expectSingleItem(result[1]).turnIndex).toBe(2);
|
||||
expect(expectToolGroup(result[2]).children.map((c) => c.turnIndex)).toEqual(
|
||||
[3, 4],
|
||||
);
|
||||
});
|
||||
|
||||
test("an errored tool call is never grouped and breaks the run", () => {
|
||||
const a = toolTurn({ ts: "2026-04-09T12:00:00Z", toolName: "shell" });
|
||||
const errored = toolTurn({ ts: "2026-04-09T12:00:01Z", toolName: "shell", isError: true });
|
||||
const errored = toolTurn({
|
||||
ts: "2026-04-09T12:00:01Z",
|
||||
toolName: "shell",
|
||||
isError: true,
|
||||
});
|
||||
const c = toolTurn({ ts: "2026-04-09T12:00:02Z", toolName: "shell" });
|
||||
const d = toolTurn({ ts: "2026-04-09T12:00:03Z", toolName: "shell" });
|
||||
const result = groupConsecutiveTools([
|
||||
|
|
@ -574,12 +722,10 @@ describe("groupConsecutiveTools", () => {
|
|||
entry(d, 3),
|
||||
]);
|
||||
expect(result.map((r) => r.kind)).toEqual(["single", "single", "group"]);
|
||||
if (result[1].kind === "single") {
|
||||
expect(result[1].turn).toBe(errored);
|
||||
}
|
||||
if (result[2].kind === "group") {
|
||||
expect(result[2].children.map((c) => c.turnIndex)).toEqual([2, 3]);
|
||||
}
|
||||
expect(expectSingleItem(result[1]).turn).toBe(errored);
|
||||
expect(expectToolGroup(result[2]).children.map((c) => c.turnIndex)).toEqual(
|
||||
[2, 3],
|
||||
);
|
||||
});
|
||||
|
||||
test("non-tool turns flush the buffer correctly", () => {
|
||||
|
|
@ -591,6 +737,7 @@ describe("groupConsecutiveTools", () => {
|
|||
content: "thinking",
|
||||
inputTokens: 0,
|
||||
outputTokens: 0,
|
||||
toolCallCount: null,
|
||||
};
|
||||
const c = toolTurn({ ts: "2026-04-09T12:00:03Z", toolName: "shell" });
|
||||
const result = groupConsecutiveTools([
|
||||
|
|
@ -600,12 +747,10 @@ describe("groupConsecutiveTools", () => {
|
|||
entry(c, 3),
|
||||
]);
|
||||
expect(result.map((r) => r.kind)).toEqual(["group", "single", "single"]);
|
||||
if (result[0].kind === "group") {
|
||||
expect(result[0].children.map((c) => c.turnIndex)).toEqual([0, 1]);
|
||||
}
|
||||
if (result[2].kind === "single") {
|
||||
expect(result[2].turnIndex).toBe(3);
|
||||
}
|
||||
expect(expectToolGroup(result[0]).children.map((c) => c.turnIndex)).toEqual(
|
||||
[0, 1],
|
||||
);
|
||||
expect(expectSingleItem(result[2]).turnIndex).toBe(3);
|
||||
});
|
||||
});
|
||||
|
||||
|
|
@ -664,7 +809,13 @@ describe("buildChatItems", () => {
|
|||
const TS = "2026-04-09T12:00:00Z";
|
||||
|
||||
function assistant(content: string) {
|
||||
return { kind: "assistant" as const, ts: TS, content, inputTokens: 0, outputTokens: 0 };
|
||||
return {
|
||||
kind: "assistant" as const,
|
||||
ts: TS,
|
||||
content,
|
||||
inputTokens: 0,
|
||||
outputTokens: 0,
|
||||
};
|
||||
}
|
||||
|
||||
function chatTool(toolName: string, isError = false) {
|
||||
|
|
@ -702,7 +853,11 @@ describe("buildChatItems", () => {
|
|||
});
|
||||
|
||||
test("non-tool turns break tool batches", () => {
|
||||
const steer = { kind: "steer" as const, ts: TS, content: "focus on the API" };
|
||||
const steer = {
|
||||
kind: "steer" as const,
|
||||
ts: TS,
|
||||
content: "focus on the API",
|
||||
};
|
||||
const items = buildChatItems([chatTool("shell"), steer, chatTool("shell")]);
|
||||
expect(items).toEqual([
|
||||
{ kind: "tools", ts: TS, count: 1, errored: 0 },
|
||||
|
|
@ -773,7 +928,11 @@ describe("buildStageActivity pending tools", () => {
|
|||
event: "agent.tool.started",
|
||||
stage_id: "plan@2",
|
||||
node_id: "plan",
|
||||
properties: { tool_call_id: "call-1", tool_name: "shell", arguments: {} },
|
||||
properties: {
|
||||
tool_call_id: "call-1",
|
||||
tool_name: "shell",
|
||||
arguments: {},
|
||||
},
|
||||
}),
|
||||
];
|
||||
expect(buildStageActivity(events, "plan@1").pendingTools).toEqual([]);
|
||||
|
|
@ -858,6 +1017,7 @@ describe("buildThreadDnaItems", () => {
|
|||
kind: "single" as const,
|
||||
turnIndex,
|
||||
turn: { kind: "system" as const, ts, content },
|
||||
selection: { kind: "single" as const, turnIndex },
|
||||
};
|
||||
}
|
||||
|
||||
|
|
@ -871,7 +1031,9 @@ describe("buildThreadDnaItems", () => {
|
|||
content: "hi",
|
||||
inputTokens: 0,
|
||||
outputTokens: 0,
|
||||
toolCallCount: null,
|
||||
},
|
||||
selection: { kind: "single" as const, turnIndex },
|
||||
};
|
||||
}
|
||||
|
||||
|
|
@ -893,6 +1055,7 @@ describe("buildThreadDnaItems", () => {
|
|||
isError: false,
|
||||
durationMs,
|
||||
},
|
||||
selection: { kind: "single" as const, turnIndex },
|
||||
};
|
||||
}
|
||||
|
||||
|
|
@ -901,6 +1064,7 @@ describe("buildThreadDnaItems", () => {
|
|||
kind: "single" as const,
|
||||
turnIndex,
|
||||
turn: { kind: "steer" as const, ts, content: "do this" },
|
||||
selection: { kind: "single" as const, turnIndex },
|
||||
};
|
||||
}
|
||||
|
||||
|
|
@ -976,40 +1140,37 @@ describe("buildThreadDnaItems", () => {
|
|||
});
|
||||
});
|
||||
|
||||
test("tool group spans first child's start to last child's end", () => {
|
||||
const child1 = {
|
||||
turnIndex: 0,
|
||||
turn: {
|
||||
kind: "tool" as const,
|
||||
ts: "2026-04-09T12:00:10Z",
|
||||
toolName: "shell",
|
||||
input: "",
|
||||
result: "",
|
||||
isError: false,
|
||||
durationMs: 1000,
|
||||
},
|
||||
};
|
||||
const child2 = {
|
||||
turnIndex: 1,
|
||||
turn: {
|
||||
kind: "tool" as const,
|
||||
ts: "2026-04-09T12:00:12Z",
|
||||
toolName: "shell",
|
||||
input: "",
|
||||
result: "",
|
||||
isError: false,
|
||||
durationMs: 2000,
|
||||
},
|
||||
};
|
||||
const group = {
|
||||
kind: "group" as const,
|
||||
test("a group's bar reuses the same wall-clock bounds the row shows", () => {
|
||||
// Children in completion order, so the group's start is not children[0].
|
||||
const late = {
|
||||
kind: "tool" as const,
|
||||
ts: "2026-04-09T12:00:12Z",
|
||||
toolName: "shell",
|
||||
ts: "2026-04-09T12:00:10Z",
|
||||
durationMs: 3000,
|
||||
children: [child1, child2],
|
||||
input: "",
|
||||
result: "",
|
||||
isError: false,
|
||||
durationMs: 2000,
|
||||
};
|
||||
const items = buildThreadDnaItems([group], RUN_START);
|
||||
// span = 12s + 2s − 10s = 4s, not the summed 3s.
|
||||
const early = {
|
||||
kind: "tool" as const,
|
||||
ts: "2026-04-09T12:00:10Z",
|
||||
toolName: "shell",
|
||||
input: "",
|
||||
result: "",
|
||||
isError: false,
|
||||
durationMs: 1000,
|
||||
};
|
||||
const grouped = groupConsecutiveTools([
|
||||
{ turn: late, index: 0 },
|
||||
{ turn: early, index: 1 },
|
||||
]);
|
||||
const group = expectToolGroup(grouped[0]);
|
||||
|
||||
// span = 12s + 2s − 10s = 4s, not the summed 3s and not children[0]'s ts.
|
||||
expect(group.ts).toBe("2026-04-09T12:00:10Z");
|
||||
expect(group.durationMs).toBe(4000);
|
||||
|
||||
const items = buildThreadDnaItems(grouped, RUN_START);
|
||||
expect(items[0]).toMatchObject({
|
||||
category: "tool",
|
||||
startMs: 10_000,
|
||||
|
|
@ -1030,3 +1191,339 @@ describe("buildThreadDnaItems", () => {
|
|||
expect(items[1]).toMatchObject({ startMs: 0, durationMs: 5000 });
|
||||
});
|
||||
});
|
||||
|
||||
describe("tool-call-only agent responses", () => {
|
||||
test("retains an empty agent.message with its timestamp, billing, and tool-call count", () => {
|
||||
const events: EventEnvelope[] = [
|
||||
envelope(1, {
|
||||
event: "agent.message",
|
||||
ts: "2026-04-09T12:00:42Z",
|
||||
stage_id: "code@1",
|
||||
node_id: "code",
|
||||
properties: {
|
||||
text: "",
|
||||
billing: { input_tokens: 4200, output_tokens: 96 },
|
||||
tool_call_count: 2,
|
||||
},
|
||||
}),
|
||||
];
|
||||
|
||||
expect(eventsToActivity(events, "code@1")).toEqual([
|
||||
{
|
||||
kind: "assistant",
|
||||
ts: "2026-04-09T12:00:42Z",
|
||||
content: "",
|
||||
inputTokens: 4200,
|
||||
outputTokens: 96,
|
||||
toolCallCount: 2,
|
||||
},
|
||||
]);
|
||||
});
|
||||
|
||||
test("does not synthesize a prompt.completed turn after an empty agent.message", () => {
|
||||
const events: EventEnvelope[] = [
|
||||
envelope(1, {
|
||||
event: "agent.message",
|
||||
stage_id: "code@1",
|
||||
node_id: "code",
|
||||
properties: { text: "", tool_call_count: 1 },
|
||||
}),
|
||||
envelope(2, {
|
||||
event: "prompt.completed",
|
||||
stage_id: "code@1",
|
||||
node_id: "code",
|
||||
properties: {
|
||||
response: "",
|
||||
billing: { input_tokens: 1, output_tokens: 2 },
|
||||
},
|
||||
}),
|
||||
];
|
||||
|
||||
const turns = eventsToActivity(events, "code@1");
|
||||
expect(turns).toHaveLength(1);
|
||||
expect(turns[0]).toMatchObject({ kind: "assistant", toolCallCount: 1 });
|
||||
});
|
||||
|
||||
test("empty responses get nonblank summary copy and stay searchable by it", () => {
|
||||
const withTools = {
|
||||
kind: "assistant" as const,
|
||||
ts: "2026-04-09T12:00:00Z",
|
||||
content: "",
|
||||
inputTokens: 0,
|
||||
outputTokens: 0,
|
||||
toolCallCount: 3,
|
||||
};
|
||||
const withOneTool = { ...withTools, toolCallCount: 1 };
|
||||
const withoutCount = { ...withTools, toolCallCount: null };
|
||||
const whitespaceOnly = { ...withTools, content: " \n\t" };
|
||||
|
||||
expect(turnSummary(withTools)).toBe("Requested 3 tool calls");
|
||||
expect(turnSummary(withOneTool)).toBe("Requested 1 tool call");
|
||||
expect(turnSummary(withoutCount)).toBe("Model response contained no text");
|
||||
expect(turnSummary(whitespaceOnly)).toBe("Requested 3 tool calls");
|
||||
|
||||
expect(searchableText(withTools)).toContain("Requested 3 tool calls");
|
||||
expect(searchableText(whitespaceOnly)).toContain("Requested 3 tool calls");
|
||||
// Text-bearing responses keep searching their own content.
|
||||
expect(searchableText({ ...withTools, content: "all done" })).toBe(
|
||||
"all done",
|
||||
);
|
||||
});
|
||||
});
|
||||
|
||||
describe("tool batch boundaries", () => {
|
||||
const STAGE = "code@1";
|
||||
const RUN_START = "2026-04-09T12:00:00Z";
|
||||
|
||||
function modelResponse(
|
||||
seq: number,
|
||||
ts: string,
|
||||
toolCallCount: number,
|
||||
text = "",
|
||||
): EventEnvelope {
|
||||
return envelope(seq, {
|
||||
event: "agent.message",
|
||||
ts,
|
||||
stage_id: STAGE,
|
||||
node_id: "code",
|
||||
properties: {
|
||||
text,
|
||||
billing: { input_tokens: 1000, output_tokens: 20 },
|
||||
tool_call_count: toolCallCount,
|
||||
},
|
||||
});
|
||||
}
|
||||
|
||||
function shellCall(
|
||||
seq: number,
|
||||
callId: string,
|
||||
startTs: string,
|
||||
endTs: string,
|
||||
command: string,
|
||||
): EventEnvelope[] {
|
||||
return [
|
||||
envelope(seq, {
|
||||
event: "agent.tool.started",
|
||||
ts: startTs,
|
||||
stage_id: STAGE,
|
||||
node_id: "code",
|
||||
properties: {
|
||||
tool_call_id: callId,
|
||||
tool_name: "shell",
|
||||
arguments: { command },
|
||||
},
|
||||
}),
|
||||
envelope(seq + 1, {
|
||||
event: "agent.tool.completed",
|
||||
ts: endTs,
|
||||
stage_id: STAGE,
|
||||
node_id: "code",
|
||||
properties: { tool_call_id: callId, tool_name: "shell", output: "ok" },
|
||||
}),
|
||||
];
|
||||
}
|
||||
|
||||
// Anonymized reproduction: eight sub-100ms shell calls issued across five
|
||||
// model responses, each response separated by a minute or more of model
|
||||
// time and carrying no text of its own.
|
||||
const REPRO_EVENTS: EventEnvelope[] = [
|
||||
envelope(1, {
|
||||
event: "stage.prompt",
|
||||
ts: RUN_START,
|
||||
stage_id: STAGE,
|
||||
node_id: "code",
|
||||
properties: { text: "investigate the failure" },
|
||||
}),
|
||||
modelResponse(2, "2026-04-09T12:00:30Z", 2),
|
||||
...shellCall(
|
||||
3,
|
||||
"c1",
|
||||
"2026-04-09T12:00:30.010Z",
|
||||
"2026-04-09T12:00:30.060Z",
|
||||
"alpha",
|
||||
),
|
||||
...shellCall(
|
||||
5,
|
||||
"c2",
|
||||
"2026-04-09T12:00:30.070Z",
|
||||
"2026-04-09T12:00:30.140Z",
|
||||
"bravo",
|
||||
),
|
||||
modelResponse(7, "2026-04-09T12:01:30Z", 1),
|
||||
...shellCall(
|
||||
8,
|
||||
"c3",
|
||||
"2026-04-09T12:01:30.010Z",
|
||||
"2026-04-09T12:01:30.050Z",
|
||||
"charlie",
|
||||
),
|
||||
modelResponse(10, "2026-04-09T12:02:40Z", 1),
|
||||
...shellCall(
|
||||
11,
|
||||
"c4",
|
||||
"2026-04-09T12:02:40.010Z",
|
||||
"2026-04-09T12:02:40.090Z",
|
||||
"delta",
|
||||
),
|
||||
modelResponse(13, "2026-04-09T12:03:50Z", 2),
|
||||
...shellCall(
|
||||
14,
|
||||
"c5",
|
||||
"2026-04-09T12:03:50.010Z",
|
||||
"2026-04-09T12:03:50.060Z",
|
||||
"echo",
|
||||
),
|
||||
...shellCall(
|
||||
16,
|
||||
"c6",
|
||||
"2026-04-09T12:03:50.070Z",
|
||||
"2026-04-09T12:03:50.130Z",
|
||||
"foxtrot",
|
||||
),
|
||||
modelResponse(18, "2026-04-09T12:05:00Z", 2),
|
||||
...shellCall(
|
||||
19,
|
||||
"c7",
|
||||
"2026-04-09T12:05:00.010Z",
|
||||
"2026-04-09T12:05:00.060Z",
|
||||
"golf",
|
||||
),
|
||||
...shellCall(
|
||||
21,
|
||||
"c8",
|
||||
"2026-04-09T12:05:00.070Z",
|
||||
"2026-04-09T12:05:00.130Z",
|
||||
"hotel",
|
||||
),
|
||||
modelResponse(23, "2026-04-09T12:06:00Z", 0, "Done."),
|
||||
];
|
||||
|
||||
function reproItems(): DisplayItem[] {
|
||||
const turns = eventsToActivity(REPRO_EVENTS, STAGE);
|
||||
return groupConsecutiveTools(turns.map((turn, index) => ({ turn, index })));
|
||||
}
|
||||
|
||||
function visibleDna(
|
||||
items: DisplayItem[],
|
||||
kinds: readonly EventKind[],
|
||||
search: string,
|
||||
) {
|
||||
const all = buildThreadDnaItems(items, RUN_START);
|
||||
const visible = filterDisplayItems(items, kinds, search);
|
||||
return filterThreadDnaItems(all, visible);
|
||||
}
|
||||
|
||||
function groupSizes(items: DisplayItem[]): (number | "single")[] {
|
||||
return items
|
||||
.filter(
|
||||
(item) =>
|
||||
item.kind === "group" ||
|
||||
(item.kind === "single" && item.turn.kind === "tool"),
|
||||
)
|
||||
.map((item) => (item.kind === "group" ? item.children.length : "single"));
|
||||
}
|
||||
|
||||
test("eight shell calls across five responses keep their original batches", () => {
|
||||
const items = reproItems();
|
||||
expect(groupSizes(items)).toEqual([2, "single", "single", 2, 2]);
|
||||
// The bug produced a single `Bash x8` group.
|
||||
expect(
|
||||
items.some((item) => item.kind === "group" && item.children.length > 2),
|
||||
).toBe(false);
|
||||
});
|
||||
|
||||
test("the default visibility pass reuses the grouped item list", () => {
|
||||
const items = reproItems();
|
||||
expect(filterDisplayItems(items, EVENT_KINDS, "")).toBe(items);
|
||||
});
|
||||
|
||||
test("batches survive excluding Agent with the kind filter", () => {
|
||||
const items = reproItems();
|
||||
const withoutAgent = EVENT_KINDS.filter((k) => k !== "assistant");
|
||||
const visible = filterDisplayItems(items, withoutAgent, "");
|
||||
|
||||
expect(groupSizes(visible)).toEqual([2, "single", "single", 2, 2]);
|
||||
expect(
|
||||
visible.some(
|
||||
(item) => item.kind === "single" && item.turn.kind === "assistant",
|
||||
),
|
||||
).toBe(false);
|
||||
// Eight tool turns remain, just spread across the same five items.
|
||||
expect(visibleTurnCount(visible)).toBe(9); // 8 tool calls + the stage prompt
|
||||
});
|
||||
|
||||
test("search matching one child keeps its whole group and merges nothing", () => {
|
||||
const items = reproItems();
|
||||
const visible = filterDisplayItems(items, EVENT_KINDS, "alpha");
|
||||
|
||||
expect(visible).toHaveLength(1);
|
||||
const only = expectToolGroup(visible[0]);
|
||||
// "bravo" never matched the search but stays in the group for context.
|
||||
expect(only.children).toHaveLength(2);
|
||||
expect(only.children.map((c) => JSON.parse(c.turn.input).command)).toEqual([
|
||||
"alpha",
|
||||
"bravo",
|
||||
]);
|
||||
});
|
||||
|
||||
test("DNA charges the long gaps to Agent and keeps every tool batch sub-second", () => {
|
||||
const bars = buildThreadDnaItems(reproItems(), RUN_START);
|
||||
const agentBars = bars.filter((b) => b.category === "agent");
|
||||
const toolBars = bars.filter((b) => b.category === "tool");
|
||||
|
||||
expect(agentBars).toHaveLength(6);
|
||||
expect(toolBars).toHaveLength(5);
|
||||
for (const bar of toolBars) {
|
||||
expect(bar.durationMs).toBeLessThan(1000);
|
||||
}
|
||||
// First response: 30s of model time from the stage prompt.
|
||||
expect(agentBars[0]).toMatchObject({ startMs: 0, durationMs: 30_000 });
|
||||
// Second: from the end of the first batch (30.140s) to 90s.
|
||||
expect(agentBars[1]).toMatchObject({ startMs: 30_140, durationMs: 59_860 });
|
||||
// The first batch itself is 130ms, not the six minutes of the whole stage.
|
||||
expect(toolBars[0]).toMatchObject({ startMs: 30_010, durationMs: 130 });
|
||||
});
|
||||
|
||||
test("hiding tools does not inflate the adjacent Agent durations", () => {
|
||||
const items = reproItems();
|
||||
const unfiltered = buildThreadDnaItems(items, RUN_START).filter(
|
||||
(b) => b.category === "agent",
|
||||
);
|
||||
const withoutTools = visibleDna(
|
||||
items,
|
||||
EVENT_KINDS.filter((k) => k !== "tool"),
|
||||
"",
|
||||
).filter((b) => b.category === "agent");
|
||||
|
||||
expect(withoutTools).toEqual(unfiltered);
|
||||
});
|
||||
|
||||
test("hiding Agent does not inflate or merge the tool bars", () => {
|
||||
const items = reproItems();
|
||||
const unfiltered = buildThreadDnaItems(items, RUN_START).filter(
|
||||
(b) => b.category === "tool",
|
||||
);
|
||||
const withoutAgent = visibleDna(
|
||||
items,
|
||||
EVENT_KINDS.filter((k) => k !== "assistant"),
|
||||
"",
|
||||
).filter((b) => b.category === "tool");
|
||||
|
||||
expect(withoutAgent).toEqual(unfiltered);
|
||||
});
|
||||
|
||||
test("row and bar selection identifiers stay one-to-one", () => {
|
||||
const items = reproItems();
|
||||
const bars = buildThreadDnaItems(items, RUN_START);
|
||||
|
||||
expect(bars.map((b) => threadSelectionId(b.selection))).toEqual(
|
||||
items.map((item) => threadSelectionId(item.selection)),
|
||||
);
|
||||
|
||||
const group = expectToolGroup(items.find((item) => item.kind === "group"));
|
||||
expect(group.selection).toEqual({
|
||||
kind: "group",
|
||||
childTurnIndices: group.children.map((c) => c.turnIndex),
|
||||
});
|
||||
});
|
||||
});
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load diff
|
|
@ -40,6 +40,7 @@ When a run resumes after a node was cancelled or lost mid-flight, the replay now
|
|||
</Accordion>
|
||||
|
||||
<Accordion title="Improvements">
|
||||
- Added `claude-opus-5` to the first-party Anthropic model catalog; the `opus` and `claude-opus` aliases now resolve to Opus 5
|
||||
- Added `gpt-sol`, `gpt-terra`, and `gpt-luna` aliases for GPT-5.6 offerings
|
||||
- Added portable `glm`, `glm52`, `glm5.2`, `deepseek`, and `deepseek-flash` aliases across direct and OpenRouter offerings
|
||||
</Accordion>
|
||||
|
|
|
|||
|
|
@ -37,7 +37,8 @@ Fabro performs this selection once when creating a run and persists the chosen p
|
|||
| Model | Provider | Aliases | Context | Cost (in/out per Mtok) | Speed |
|
||||
|---|---|---|---|---|---|
|
||||
| `claude-fable-5` | anthropic | `fable`, `claude-fable` | 1M | $10.00 / $50.00 | n/a |
|
||||
| `claude-opus-4-8` | anthropic | `opus`, `claude-opus` | 1M | $5.00 / $25.00 | 25 tok/s |
|
||||
| `claude-opus-5` | anthropic | `opus`, `claude-opus` | 1M | $5.00 / $25.00 | n/a |
|
||||
| `claude-opus-4-8` | anthropic | | 1M | $5.00 / $25.00 | 25 tok/s |
|
||||
| `claude-opus-4-7` | anthropic | | 1M | $5.00 / $25.00 | 25 tok/s |
|
||||
| `claude-opus-4-6` | anthropic | | 1M | $5.00 / $25.00 | 25 tok/s |
|
||||
| `claude-sonnet-4-6` | anthropic | `sonnet`, `claude-sonnet` | 200K | $3.00 / $15.00 | 50 tok/s |
|
||||
|
|
|
|||
|
|
@ -299,6 +299,7 @@ mod tests {
|
|||
("claude-opus-4-6", "claude-opus-4-6", T::Anthropic, C::AnthropicMessages, B::Anthropic, P::Anthropic),
|
||||
("claude-opus-4-7", "claude-opus-4-7", T::Anthropic, C::AnthropicMessages, B::Anthropic, P::Anthropic),
|
||||
("claude-opus-4-8", "claude-opus-4-8", T::Anthropic, C::AnthropicMessages, B::Anthropic, P::Anthropic),
|
||||
("claude-opus-5", "claude-opus-5", T::Anthropic, C::AnthropicMessages, B::Anthropic, P::Anthropic),
|
||||
("claude-sonnet-4-5", "claude-sonnet-4-5", T::Anthropic, C::AnthropicMessages, B::Anthropic, P::Anthropic),
|
||||
("claude-sonnet-4-6", "claude-sonnet-4-6", T::Anthropic, C::AnthropicMessages, B::Anthropic, P::Anthropic),
|
||||
("gemini-3-flash-preview", "gemini-3-flash-preview", T::Gemini, C::GeminiGenerate, B::Gemini, P::Gemini),
|
||||
|
|
|
|||
|
|
@ -31,13 +31,44 @@ input_cost_per_mtok = 10.0
|
|||
output_cost_per_mtok = 50.0
|
||||
cache_input_cost_per_mtok = 1.0
|
||||
|
||||
[providers.anthropic.models."claude-opus-5"]
|
||||
display_name = "Claude Opus 5"
|
||||
family = "claude-5"
|
||||
training = "2026-05-01"
|
||||
knowledge_cutoff = "May 2026"
|
||||
aliases = ["opus", "claude-opus"]
|
||||
|
||||
[providers.anthropic.models."claude-opus-5".limits]
|
||||
context_window = 1000000
|
||||
max_output = 128000
|
||||
|
||||
[providers.anthropic.models."claude-opus-5".features]
|
||||
tools = true
|
||||
vision = true
|
||||
reasoning = true
|
||||
reasoning_effort = "levels"
|
||||
prompt_cache = true
|
||||
sampling_params = false
|
||||
|
||||
[providers.anthropic.models."claude-opus-5".controls]
|
||||
speed = ["fast"]
|
||||
|
||||
[providers.anthropic.models."claude-opus-5".costs]
|
||||
input_cost_per_mtok = 5.0
|
||||
output_cost_per_mtok = 25.0
|
||||
cache_input_cost_per_mtok = 0.5
|
||||
|
||||
[providers.anthropic.models."claude-opus-5".costs.speed.fast]
|
||||
input_cost_per_mtok = 10.0
|
||||
output_cost_per_mtok = 50.0
|
||||
cache_input_cost_per_mtok = 1.0
|
||||
|
||||
[providers.anthropic.models."claude-opus-4-8"]
|
||||
display_name = "Claude Opus 4.8"
|
||||
family = "claude-4"
|
||||
training = "2026-01-01"
|
||||
knowledge_cutoff = "Jan 2026"
|
||||
estimated_output_tps = 25
|
||||
aliases = ["opus", "claude-opus"]
|
||||
|
||||
[providers.anthropic.models."claude-opus-4-8".limits]
|
||||
context_window = 1000000
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue