feat(agent-sessions): make the overview's tools a ledger you can act on

apps/clickhouse-builder-docs/src/sidebar-icons.tsx

@@ -1,7 +1,7 @@
 import type { ReactNode } from "react"
 
 // Nucleo geometry from Maple’s existing icon set (apps/web/src/components/icons).
-const icons: Record<string, ReactNode> = {
+const icons = {
 	"branch-fork": (
 		<>
 			{" "}
@@ -297,10 +297,16 @@ const icons: Record<string, ReactNode> = {
 			))}{" "}
 		</>
 	),
+} satisfies Record<string, ReactNode>
+
+type SidebarIconName = keyof typeof icons
+
+function isSidebarIconName(name: string): name is SidebarIconName {
+	return name in icons
 }
 
 export function sidebarIcon(name: string | undefined) {
-	const icon = name ? icons[name] : undefined
+	const icon = name !== undefined && isSidebarIconName(name) ? icons[name] : undefined
 	if (!icon) return undefined
 	return (
 		<svg viewBox="0 0 24 24" width="16" height="16" fill="none" aria-hidden="true" focusable="false">

apps/web/src/components/agent-sessions/session-detail/session-detail.test.tsx

@@ -437,12 +437,15 @@ describe("SessionOverview", () => {
 	})
 
 	// A mid-session failure the session recovered from is not a failed session —
-	// but it is exactly what the findings list exists to surface.
-	it("completes-with-findings when something failed mid-session, and opens it", () => {
+	// but it is exactly what the findings list exists to surface. There is no
+	// verdict line above it: the findings ARE the verdict, and a headline
+	// counting them said it twice.
+	it("leads with the findings when something failed mid-session, and opens one", () => {
 		const onSelectSpan = vi.fn()
 		render(<Overview onSelectSpan={onSelectSpan} />)
 
-		expect(screen.getByText(/Completed, with 1 finding/)).toBeTruthy()
+		expect(screen.queryByText(/^Completed/)).toBeNull()
+		expect(screen.getByText("Findings")).toBeTruthy()
 		fireEvent.click(screen.getByText("error · run_tests"))
 		expect(onSelectSpan).toHaveBeenCalledWith("tool-3")
 	})
@@ -493,16 +496,48 @@ describe("SessionOverview", () => {
 		expect(screen.getByText("No findings.")).toBeTruthy()
 	})
 
-	// A tool called ten times and failing every time reads nothing like one that
-	// never failed; the rail used to draw both as the same bar.
-	it("separates a tool's failed calls from its successful ones", () => {
-		render(<Overview />)
+	// The ledger's row is a summary; the calls behind it are the point. A mark is
+	// one call, and it opens that span rather than describing it.
+	it("puts every call on the session's clock and opens the span behind a mark", () => {
+		const onSelectSpan = vi.fn()
+		render(<Overview onSelectSpan={onSelectSpan} />)
+
+		fireEvent.click(screen.getByRole("button", { name: /^run_tests — turn 1, 14s in, 20.0s/ }))
+		expect(onSelectSpan).toHaveBeenCalledWith("tool-3")
+	})
+
+	// The description and the failure used to live in two different places — the
+	// rail disclosed one, the findings list carried the other. Expanding the tool
+	// is where a reader asks about the tool.
+	it("discloses a tool's definition and its failed calls when the row is expanded", () => {
+		const onSelectSpan = vi.fn()
+		const described = sessionOf([
+			agentSpan({ spanId: "d-agent", startMs: 0, durationMs: 30 * SECOND }),
+			toolSpan({
+				spanId: "d-tool",
+				parentSpanId: "d-agent",
+				startMs: SECOND,
+				durationMs: 4 * SECOND,
+				toolName: "reindex_shard",
+				statusCode: "Error",
+				statusMessage: "shard 3 is locked by a running merge",
+				genAi: { errorType: "SHARD_LOCKED", toolDescription: "Rebuild a shard's index." },
+			}),
+		])
+		render(<Overview turns={described.turns} summary={described.summary} onSelectSpan={onSelectSpan} />)
+
+		expect(screen.queryByText("Rebuild a shard's index.")).toBeNull()
+
+		fireEvent.click(screen.getByRole("button", { name: "reindex_shard" }))
+
+		expect(screen.getByText("Rebuild a shard's index.")).toBeTruthy()
+		// The findings list names the same failure; the disclosure is where a
+		// reader asking about this tool finds it.
+		expect(screen.getAllByText("SHARD_LOCKED").length).toBeGreaterThan(0)
+		expect(screen.getAllByText("shard 3 is locked by a running merge").length).toBe(2)
 
-		// run_tests: one call, and it errored.
-		expect(screen.getByTitle("1 failed")).toBeTruthy()
-		expect(screen.getByTitle("0 ok · 1 errored")).toBeTruthy()
-		// read_file and grep_repo ran clean, and say so by having nothing to say.
-		expect(screen.getAllByTitle("1 ok · 0 errored").length).toBe(2)
+		fireEvent.click(screen.getByRole("button", { name: /Open span/ }))
+		expect(onSelectSpan).toHaveBeenCalledWith("d-tool")
 	})
 
 	it("says no cost was reported rather than pricing tokens itself", () => {
@@ -1201,12 +1236,12 @@ describe("SessionViews", () => {
 	// height — which is what sent "Open in Traces view" nowhere near its row.
 	it("takes the view being left out of the page, not just out of sight", () => {
 		render(<Views view="overview" />)
-		expect(screen.getByText(/Completed, with/)).toBeTruthy()
+		expect(screen.getByText("Where the time went")).toBeTruthy()
 
 		fireEvent.click(screen.getByRole("tab", { name: /Traces/ }))
 
 		expect(screen.getByText("Model / target")).toBeTruthy()
-		expect(screen.queryByText(/Completed, with/)).toBeNull()
+		expect(screen.queryByText("Where the time went")).toBeNull()
 	})
 
 	// The tab choice lives beside the other cross-view state in SessionViews: