Compare commits

..
2 Commits
Author SHA1 Message Date
Miguel Palhas 9f5b71d563 Merge remote-tracking branch 'origin/main' into aeolian
ci / web (push) Failing after 29s
ci / rust (push) Failing after 51s
ci / flake (push) Successful in 3m9s
2026-09-12 10:42:10 +01:00
Miguel PalhasandClaude Opus 4.6 de603ccd20 feat(maestro): split tool groups on text boundaries
Tool rows interleaved with agent text now produce separate
collapsed groups instead of one per turn. The timeline reads
text → group → text → group, matching chronological order.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-09-12 10:42:06 +01:00
3 changed files with 107 additions and 61 deletions
+11 -8
View File
@@ -189,14 +189,17 @@ choices, not daemon ones.
presentation for `Notice`. `Plan` is the exception: both clients omit it
from the transcript; the TUI may show the current plan in its status line,
while the web client deliberately has no stats or plan rail.
- The web client groups two or more tool rows after the same `User` row into
one collapsed turn activity block. `AgentText` and `AgentThought` rows stay
outside the group as standalone transcript items — the operator sees all
reasoning without expanding. Resolved approval rows between tools are
absorbed into the group. The collapsed header reads "N tool calls · Label
'arg'" where the label and argument come from the last tool row's
`splitToolTitle` — the operator sees which command is current at a glance.
Opening the group reveals tool rows in chronological order.
- The web client groups each contiguous run of two or more tool rows into a
collapsed activity block. An `AgentText` or `AgentThought` row between
tools splits them into separate groups, so the timeline reads
text → group → text → group rather than lumping every tool in a turn into
one box. Text and thought rows stay outside groups as standalone transcript
items — the operator sees all reasoning without expanding. Resolved
approval rows between tools within a run are absorbed into the group. The
collapsed header reads "N tool calls · Label 'arg'" where the label and
argument come from the last tool row's `splitToolTitle` — the operator
sees which command is current at a glance. Opening the group reveals tool
rows in chronological order.
- The web client's user, agent, thought, tool-detail, and notice text blocks
copy their complete source text when clicked or keyboard-activated.
- The web client consumes `available_commands_update` notices as the live
+44 -10
View File
@@ -115,8 +115,8 @@ describe("activity grouping", () => {
it("groups a turn's tools once there are two of them", () => {
const rows = [text("user:1", "user", "go"), tool("tool:1"), tool("tool:2")];
const groups = activityGroupsByTurn(rows, null, false);
expect([...groups.keys()]).toEqual(["turn:user:1"]);
expect(groups.get("turn:user:1")).toMatchObject({ toolCount: 2 });
expect([...groups.keys()]).toEqual(["turn:user:1:0"]);
expect(groups.get("turn:user:1:0")).toMatchObject({ toolCount: 2 });
});
it("leaves a single tool call standalone", () => {
@@ -124,7 +124,7 @@ describe("activity grouping", () => {
expect(activityGroupsByTurn(rows, null, false).size).toBe(0);
});
it("keeps intermediate text outside the group as standalone items", () => {
it("splits tools into separate runs when text appears between them", () => {
const rows = [
text("user:1", "user", "go"),
tool("tool:1"),
@@ -132,9 +132,23 @@ describe("activity grouping", () => {
tool("tool:2"),
text("agent:2", "agent_text", "done"),
];
const group = activityGroupsByTurn(rows, null, false).get("turn:user:1")!;
expect(group.rows.map((r) => r.id)).toEqual(["tool:1", "tool:2"]);
expect(group.toolCount).toBe(2);
// Each run has only one tool, so neither qualifies for grouping.
expect(activityGroupsByTurn(rows, null, false).size).toBe(0);
});
it("groups contiguous tools separated by text into distinct groups", () => {
const rows = [
text("user:1", "user", "go"),
tool("tool:1"),
tool("tool:2"),
text("agent:1", "agent_text", "working"),
tool("tool:3"),
tool("tool:4"),
];
const groups = activityGroupsByTurn(rows, null, false);
expect([...groups.keys()]).toEqual(["turn:user:1:0", "turn:user:1:1"]);
expect(groups.get("turn:user:1:0")!.rows.map((r) => r.id)).toEqual(["tool:1", "tool:2"]);
expect(groups.get("turn:user:1:1")!.rows.map((r) => r.id)).toEqual(["tool:3", "tool:4"]);
});
it("keeps thoughts outside the group as standalone items", () => {
@@ -145,7 +159,7 @@ describe("activity grouping", () => {
text("thought:1", "agent_thought", "hmm"),
text("thought:2", "agent_thought", "aha"),
];
const group = activityGroupsByTurn(rows, null, false).get("turn:user:1")!;
const group = activityGroupsByTurn(rows, null, false).get("turn:user:1:0")!;
expect(group.rows.map((r) => r.id)).toEqual(["tool:1", "tool:2"]);
});
@@ -155,7 +169,7 @@ describe("activity grouping", () => {
row("tool:1", "tool_call", { title: "Read(src/main.ts)", status: "completed" }),
row("tool:2", "tool_call", { title: "Execute(git status)", kind: "execute", status: "completed" }),
];
const group = activityGroupsByTurn(rows, null, false).get("turn:user:1")!;
const group = activityGroupsByTurn(rows, null, false).get("turn:user:1:0")!;
expect(group.lastToolLabel).toBe("Execute");
expect(group.lastToolArg).toBe("git status");
});
@@ -169,7 +183,7 @@ describe("activity grouping", () => {
row("approval:2", "approval"),
tool("tool:3"),
];
const group = activityGroupsByTurn(rows, "approval:2", false).get("turn:user:1")!;
const group = activityGroupsByTurn(rows, "approval:2", false).get("turn:user:1:0")!;
expect(group.rows.map((r) => r.id)).toEqual(["tool:1", "approval:1", "tool:2", "tool:3"]);
});
});
@@ -180,11 +194,31 @@ describe("transcript render list", () => {
const items = transcriptItems(rows, null, false);
expect(items.map((item) => (item.type === "row" ? item.row.id : item.id))).toEqual([
"user:1",
"turn:user:1",
"turn:user:1:0",
"agent:1",
]);
});
it("interleaves groups and text in timeline order", () => {
const rows = [
text("user:1", "user", "go"),
tool("tool:1"),
tool("tool:2"),
text("agent:1", "agent_text", "working"),
tool("tool:3"),
tool("tool:4"),
text("agent:2", "agent_text", "done"),
];
const items = transcriptItems(rows, null, false);
expect(items.map((item) => (item.type === "row" ? item.row.id : item.id))).toEqual([
"user:1",
"turn:user:1:0",
"agent:1",
"turn:user:1:1",
"agent:2",
]);
});
it("omits plan rows, state-only notices and standalone resolved approvals", () => {
const rows = [
text("user:1", "user", "go"),
+52 -43
View File
@@ -155,57 +155,68 @@ export interface ActivityGroup {
}
/**
* Group tool rows (and resolved approvals between them) by turn. Text and
* thought rows stay outside the group as standalone transcript items — the
* operator sees all reasoning without expanding anything.
* Group contiguous tool rows (and resolved approvals between them) into
* collapsible blocks. An `AgentText` or `AgentThought` row between tools
* splits the run — each contiguous stretch of tools becomes its own group,
* so the timeline reads text → group → text → group rather than lumping
* every tool in a turn into one box.
*
* Returns groups keyed by `turn:<userId>`.
* A group must contain at least two tools to qualify; a single tool stays
* standalone. Returns groups keyed by `turn:<userId>:<seq>`.
*/
export function activityGroupsByTurn(
rows: TranscriptRow[],
pendingApprovalId: string | null,
_turnActive: boolean,
): Map<string, ActivityGroup> {
// First: count tools per turn to know which turns qualify for grouping.
const toolCounts = new Map<string, number>();
let turnId = "before-first-message";
for (const row of rows) {
if (row.kind === "user") turnId = row.id;
if (row.kind !== "tool_call" && row.kind !== "tool_diff") continue;
const gid = `turn:${turnId}`;
toolCounts.set(gid, (toolCounts.get(gid) ?? 0) + 1);
}
// Second: build groups for turns with 2+ tools.
const groups = new Map<string, ActivityGroup>();
turnId = "before-first-message";
let pending: TranscriptRow[] = [];
let currentGid = `turn:${turnId}`;
let turnId = "before-first-message";
let runSeq = 0;
let runTools: TranscriptRow[] = [];
let runPending: TranscriptRow[] = [];
const flushRun = () => {
if (runTools.length >= 2) {
const gid = `turn:${turnId}:${runSeq}`;
const tool = (runTools[runTools.length - 1].data ?? {}) as { title?: string; name?: string | null; kind?: string };
const { label, arg } = splitToolTitle(tool.title, tool.name ?? undefined, tool.kind);
groups.set(gid, {
toolCount: runTools.length,
rows: [...runTools],
lastToolLabel: label,
lastToolArg: arg,
});
runSeq++;
}
runTools = [];
runPending = [];
};
for (const row of rows) {
if (row.kind === "user") {
flushRun();
turnId = row.id;
currentGid = `turn:${turnId}`;
pending = [];
runSeq = 0;
continue;
}
// Text and thought rows break the current tool run.
if (row.kind === "agent_text" || row.kind === "agent_thought") {
flushRun();
continue;
}
const count = toolCounts.get(currentGid) ?? 0;
if (count < 2) continue;
const isResolvedApproval = row.kind === "approval" && row.id !== pendingApprovalId;
if (isResolvedApproval) {
pending.push(row);
runPending.push(row);
continue;
}
if (row.kind === "tool_call" || row.kind === "tool_diff") {
const group = groups.get(currentGid) ?? { toolCount: count, rows: [], lastToolLabel: "", lastToolArg: "" };
group.rows.push(...pending);
pending = [];
group.rows.push(row);
const tool = (row.data ?? {}) as { title?: string; name?: string | null; kind?: string };
const { label, arg } = splitToolTitle(tool.title, tool.name ?? undefined, tool.kind);
group.lastToolLabel = label;
group.lastToolArg = arg;
groups.set(currentGid, group);
// Flush any pending resolved approvals into the run before this tool.
runTools.push(...runPending);
runPending = [];
runTools.push(row);
}
}
flushRun();
return groups;
}
@@ -217,10 +228,10 @@ export type TranscriptItem =
| { type: "group"; id: string; group: ActivityGroup };
/**
* The transcript in render order: rows, with each turn's tool activity folded
* into a single group item where one exists. Resolved approvals not absorbed
* into a group are dropped — the tool row itself already identifies the
* action. `plan` rows and state-only notices are dropped too (both clients
* The transcript in render order: rows, with each contiguous tool stretch
* folded into a single group item where one exists. Resolved approvals not
* absorbed into a group are dropped — the tool row itself already identifies
* the action. `plan` rows and state-only notices are dropped too (both clients
* omit `Plan` from the transcript). A turn footer only survives as the final
* item, so old footers stop reading like transcript content.
*/
@@ -230,21 +241,19 @@ export function transcriptItems(
turnActive: boolean,
): TranscriptItem[] {
const groups = activityGroupsByTurn(rows, pendingApprovalId, turnActive);
const grouped = new Set<string>();
for (const group of groups.values()) for (const row of group.rows) grouped.add(row.id);
// Reverse index: row id → group id that contains it.
const rowToGroup = new Map<string, string>();
for (const [groupId, group] of groups) for (const r of group.rows) rowToGroup.set(r.id, groupId);
const items: TranscriptItem[] = [];
let turnId = "before-first-message";
const emitted = new Set<string>();
for (const row of rows) {
if (row.kind === "user") turnId = row.id;
const groupId = `turn:${turnId}`;
const group = groups.get(groupId);
if (group && grouped.has(row.id)) {
const groupId = rowToGroup.get(row.id);
if (groupId !== undefined) {
// The group renders once, at its first member's position.
if (!emitted.has(groupId)) {
emitted.add(groupId);
items.push({ type: "group", id: groupId, group });
items.push({ type: "group", id: groupId, group: groups.get(groupId)! });
}
continue;
}