@code-yeongyu/senpi-codemode 2026.10.9 → 2026.10.10-10

package/src/tool/render.ts

@@ -7,6 +7,7 @@ import {
 	type ToolDefinition,
 	type ToolRenderResultOptions,
 	truncateToVisualLines,
+	visibleWidth,
 } from "@code-yeongyu/senpi";
 import type { TruncationMeta } from "../output/output-meta.ts";
 import { highlightedCode } from "./code-preview.ts";
@@ -21,6 +22,7 @@ import {
 	JSON_TREE_SCALAR_LEN_EXPANDED,
 	renderJsonTreeLines,
 } from "./json-tree.ts";
+import { knownLanguage, leadsWithHeadline, liveHeadline } from "./live-headline.ts";
 import { formatRuntimeBadge } from "./runtime-label.ts";
 import { codePointPrefix, formatDuration, renderToolCallWidget } from "./tool-widgets.ts";
 import type {
@@ -341,18 +343,37 @@ function cellHeader(cell: EvalCellResult, environment: RenderEnvironment, badges
 		environment.spinnerFrame ?? Math.floor((cellElapsedMs(cell, environment) ?? 0) / LIVE_RENDER_TICK_MS),
 	);
 	const runtimeBadge = cell.runtime === undefined ? "" : ` (${formatRuntimeBadge(cell.language, cell.runtime)})`;
-	let header = `eval ${cell.language}${runtimeBadge} ${presentation.label} ${presentation.icon}`;
+	let header = leadsWithHeadline(cell.status)
+		? `eval ${cell.language}${runtimeBadge} ${presentation.label}`
+		: `eval ${cell.language}${runtimeBadge} ${presentation.label} ${presentation.icon}`;
 	if (cell.queuedBehind !== undefined && cell.queuedBehind.length > 0)
 		header += ` · queued behind ${cell.queuedBehind.map(sanitizeTerminalLabel).join(", ")}`;
+	else if (cell.queuedBehind !== undefined && cell.status === "queued")
+		header += ` · waiting for the ${cell.language} kernel to be ready`;
 	const throughputBadge = badges.throughput === undefined ? undefined : formatThroughputBadge(badges.throughput);
 	if (throughputBadge !== undefined) header += ` · ${throughputBadge}`;
 	const elapsedMs = badges.throughput?.wallDurationMs ?? cellElapsedMs(cell, environment);
 	if (elapsedMs !== undefined) header += ` · ${formatDuration(elapsedMs)}`;
 	if (badges.reset) header += " · reset";
 	if (badges.timeout !== undefined) header += ` · timeout ${badges.timeout}s`;
+	if (leadsWithHeadline(cell.status))
+		header = headlined(presentation.icon, cell.summary, cell.code, header, environment);
 	return style(environment.theme, presentation.color, header);
 }
 
+// An in-progress row leads with what the cell is doing (senpi#2802); collapsed, the headline is cut so the whole
+// header stays on one line, and the code moves behind expand.
+function headlined(
+	icon: string,
+	summary: string | undefined,
+	code: string | undefined,
+	rest: string,
+	environment: RenderEnvironment,
+): string {
+	const budget = environment.expanded ? undefined : environment.width - 3 - visibleWidth(`${icon}  · ${rest}`);
+	return `${icon} ${liveHeadline(summary, code, budget)} · ${rest}`;
+}
+
 function previewText(
 	text: string,
 	maxLines: number,
@@ -612,12 +633,23 @@ function renderAgentProgressEvents(events: readonly EvalStatusEvent[], environme
 }
 
 function renderCell(cell: EvalCellResult, environment: RenderEnvironment, badges: CellBadges): string[] {
+	if (leadsWithHeadline(cell.status) && !environment.expanded && cell.output.trim().length === 0) {
+		const statusEvents = (cell.statusEvents ?? []).filter((event) => event.op !== "agent");
+		if (statusEvents.length === 0) {
+			// Nothing to frame yet: the collapsed live row is its headline alone, one line, no empty box.
+			const lines = renderPrefixed(cellHeader(cell, environment, badges), environment, LIVE_LINE_PREFIX);
+			const agentEvents = (cell.statusEvents ?? []).filter((event) => event.op === "agent");
+			if (agentEvents.length > 0) appendLines(lines, renderAgentProgressEvents(agentEvents, environment));
+			return lines;
+		}
+	}
 	const lines = renderPrefixed(cellHeader(cell, environment, badges), environment, {
 		prefix: "╭─ ",
 		continuation: "│  ",
 		color: "borderAccent",
 	});
-	if (cell.summary !== undefined) {
+	const headlined = leadsWithHeadline(cell.status);
+	if (cell.summary !== undefined && !headlined) {
 		appendLines(
 			lines,
 			summaryVisualLines(cell.summary, Math.max(1, environment.width - 2), environment.expanded).map(
@@ -626,11 +658,14 @@ function renderCell(cell: EvalCellResult, environment: RenderEnvironment, badges
 		);
 	}
 	const innerWidth = Math.max(1, environment.width - 2);
-	const codePreview = previewText(
-		highlightedCode(cell.code, cell.language, environment.theme, environment.repaint),
-		environment.expanded ? Number.POSITIVE_INFINITY : CODE_PREVIEW_LINES,
-		innerWidth,
-	);
+	const codePreview =
+		headlined && !environment.expanded
+			? { lines: [], skipped: 0 }
+			: previewText(
+					highlightedCode(cell.code, cell.language, environment.theme, environment.repaint),
+					environment.expanded ? Number.POSITIVE_INFINITY : CODE_PREVIEW_LINES,
+					innerWidth,
+				);
 	if (codePreview.skipped > 0) {
 		appendLines(
 			lines,
@@ -879,6 +914,10 @@ function displaySummary(summary: string | undefined): string | undefined {
 	return normalizeEvalSummary(summary);
 }
 
+function cellIdSuffix(cellId: unknown): string {
+	return typeof cellId === "string" && cellId !== "" ? ` ${sanitizeTerminalLabel(cellId)}` : "";
+}
+
 export function renderEvalCall(
 	args: EvalToolRequest,
 	theme: Theme | undefined,
@@ -892,15 +931,23 @@ export function renderEvalCall(
 		return component;
 	}
 	if (!isEvalRunInput(args)) {
-		const title = args.action === "list" ? "eval list" : `eval ${args.action} ${args.cell_id}`;
+		// While a peek/stop call streams in, `cell_id` can still be missing; the title is the action alone until it arrives.
+		const title = args.action === "list" ? "eval list" : `eval ${args.action}${cellIdSuffix(args.cell_id)}`;
 		component.setBlocks([{ kind: "text", text: style(theme, "toolTitle", title) }]);
 		return component;
 	}
+	// While the model streams the call, `code` usually arrives before `language` and `summary`; a partial call still
+	// renders (the host would otherwise fall back to a raw key=value row).
+	const code = typeof args.code === "string" ? args.code : "";
+	const language = knownLanguage(args.language);
 	if (theme === undefined && context.spinnerFrame === undefined) {
 		const reset = args.reset === true ? " reset" : "";
 		const timeout = args.timeout === undefined ? "" : ` timeout ${args.timeout}s`;
 		component.setBlocks([
-			{ kind: "text", text: style(theme, "toolTitle", `eval ${args.language}${reset}${timeout}`) },
+			{
+				kind: "text",
+				text: style(theme, "toolTitle", `eval${language === undefined ? "" : ` ${language}`}${reset}${timeout}`),
+			},
 			...(displaySummary(args.summary) === undefined
 				? []
 				: [summaryBlock(displaySummary(args.summary) ?? "", theme, context.expanded)]),
@@ -909,9 +956,11 @@ export function renderEvalCall(
 				text: style(
 					theme,
 					"mdCodeBlock",
-					args.code.trim().length > 0
-						? displayCode(args.code, args.language, context.argsComplete ? context.invalidate : undefined)
-						: "...",
+					code.trim().length === 0
+						? "..."
+						: language === undefined
+							? code
+							: displayCode(code, language, context.argsComplete ? context.invalidate : undefined),
 				),
 				maxVisualLines: context.expanded ? undefined : CODE_PREVIEW_LINES,
 				collapseKind: "code",
@@ -933,11 +982,13 @@ export function renderEvalCall(
 					now: renderNow(context),
 					...(context.argsComplete ? { repaint: context.invalidate } : {}),
 				};
+				const summary = displaySummary(args.summary);
+				if (language === undefined) return streamingCallLines(summary, code, environment);
 				const cell: EvalCellResult = {
 					index: 0,
-					...(displaySummary(args.summary) === undefined ? {} : { summary: displaySummary(args.summary) ?? "" }),
-					code: args.code,
-					language: args.language,
+					...(summary === undefined ? {} : { summary }),
+					code,
+					language,
 					output: "",
 					status: context.spinnerFrame === undefined ? "pending" : "running",
 				};
@@ -952,6 +1003,14 @@ export function renderEvalCall(
 	return component;
 }
 
+/** A call whose language has not streamed in yet: the headline row alone, until the full frame can render. */
+function streamingCallLines(summary: string | undefined, code: string, environment: RenderEnvironment): string[] {
+	const icon = environment.spinnerFrame === undefined ? "○" : spinner(environment.spinnerFrame);
+	return renderPrefixed(headlined(icon, summary, code, "eval", environment), environment, LIVE_LINE_PREFIX);
+}
+
+const LIVE_LINE_PREFIX: PrefixStyle = { prefix: "╶─ ", continuation: "   ", color: "borderAccent" };
+
 export function renderEvalResult(
 	result: AgentToolResult<EvalResultDetails>,
 	options: ToolRenderResultOptions,

package/src/tool/run-eval-cell.ts

@@ -7,12 +7,16 @@ import {
 	kernelToolsStorage,
 } from "@code-yeongyu/senpi";
 import type { KernelToHostMessage } from "../bridge/protocol.ts";
+import { resolveSandbox } from "../config/feature-settings.ts";
 import { DEFAULT_FOREGROUND_WINDOW_SECONDS, defaultCodemodeSettings } from "../config/settings.ts";
+import type { JsEnvironments } from "../environments/js-environments.ts";
+import type { PackagesInstallEnvironments } from "../environments/packages-install.ts";
 import {
 	KERNEL_TOOLS_CAPABILITIES,
 	type KernelToolsCapability,
 	type KernelToolsDescribeResult,
 } from "../kernels/js/kernel-tools-types.ts";
+import { sandboxCellExecutor, sandboxRuntimeInfo } from "../kernels/sandbox/sandbox-cell.ts";
 import { TIMEOUT_PAUSE_OP, TIMEOUT_RESUME_OP } from "../timeouts/bridge-timeout.ts";
 import { abortError, CellExecution, defaultTimeoutFactory } from "./cell-execution.ts";
 import { CellHandler, type CellState } from "./cell-handler.ts";
@@ -42,7 +46,8 @@ export async function runEvalCell(
 	const bridgeAbortController = new AbortController();
 	const cellSignal = AbortSignal.any([invocation.signal, bridgeAbortController.signal]);
 	const bridgeContext: ExtensionContext = { ...invocation.ctx, signal: cellSignal };
-	const runtime = options.runtimes?.[invocation.input.language];
+	const runtime =
+		invocation.input.isolate === true ? sandboxRuntimeInfo() : options.runtimes?.[invocation.input.language];
 	const state: CellState = {
 		input: invocation.input,
 		...(runtime === undefined ? {} : { runtime }),
@@ -238,6 +243,7 @@ async function executeCell(
 				...(kernelTools === undefined ? {} : { kernelTools }),
 				...(options.handles === undefined ? {} : { handles: options.handles }),
 				hardDeadlineMs: cell.startedAtMs + cell.hardLimitSeconds * 1_000,
+				...cellEnvironments(invocation.input.language, options),
 			});
 			handler = activeHandler;
 			cellManager.bindKernel(
@@ -261,25 +267,50 @@ async function executeCell(
 			}
 			if (invocation.input.reset) await execution.wait(kernel.reset());
 			execution.setKernel(kernel);
-			const magic = planMagicCell(invocation.input.language, invocation.input.code, options.pythonEnvironments);
+			const magic = planMagicCell(
+				invocation.input.language,
+				invocation.input.code,
+				options.pythonEnvironments,
+				options.jsEnvironments,
+			);
 			// Resolved when the cell's turn comes in the kernel's queue: a %load reads the file the cells ahead of it
 			// wrote, and a refusal settles in queue order like any cell.
 			const loadOptions = { cwd: invocation.ctx.cwd, artifactsDir: options.artifactsDir };
-			const resolveAtStart =
-				magic.kind === "load"
+			// An isolated cell never runs a file outside its sandbox: %load there is refused.
+			const isolatedLoad = invocation.input.isolate === true && magic.kind === "load";
+			const resolveAtStart = isolatedLoad
+				? () => ({
+						ok: false as const,
+						message:
+							"isolate: true cannot run a %load cell: an isolated cell sees no host files, and loading one would break that isolation. Run %load in a normal cell, or paste the code into the isolated cell.",
+					})
+				: magic.kind === "load"
 					? () => loadCell(magic.target, loadOptions)
 					: magic.kind === "refused"
 						? () => ({ ok: false as const, message: magic.message })
 						: undefined;
+			const isolated =
+				invocation.input.isolate === true && resolveAtStart === undefined
+					? sandboxCellExecutor(invocation.input.code, {
+							sandbox: resolveSandbox(options.settings ?? {}),
+							executeTool: options.executeTool,
+							toolNames: () => (options.listTools?.() ?? []).map((tool) => tool.name),
+							describeTool: (name) => options.listTools?.().find((tool) => tool.name === name)?.description,
+						})
+					: undefined;
 			const environments = invocation.input.language === "py" ? options.pythonEnvironments : undefined;
 			const envRoot = environments === undefined ? undefined : () => environments.activeRoot ?? "";
+			const jsEnvironments = invocation.input.language === "js" ? options.jsEnvironments : undefined;
+			const packageRoot = jsEnvironments === undefined ? undefined : () => jsEnvironments.packageRoot;
 			const result = await execution.wait(
 				kernel.run({
 					cellId: invocation.cellId,
 					code: invocation.input.code,
 					...(resolveAtStart === undefined ? {} : { resolveAtStart }),
 					...(magic.kind === "host" ? { host: magic.executor } : {}),
+					...(isolated === undefined ? {} : { host: isolated }),
 					...(envRoot === undefined ? {} : { envRoot }),
+					...(packageRoot === undefined ? {} : { packageRoot }),
 					...(bridgeCellToken === undefined ? {} : { bridgeCellToken }),
 					kernelPreludes: options.kernelPreludes?.(),
 					onMessage,
@@ -336,3 +367,14 @@ function kernelToolsFor(kernel: EvalKernel): KernelToolsCapability | undefined {
 		invoke: (request, options) => withTools.invokeKernelTool(request, options),
 	} satisfies ExtensionKernelTools;
 }
+
+// Only JS cells call tools in-process; a Python cell reaches packages.install() over the bridge, whose session manager
+// holds the same Python environment its magics use.
+function cellEnvironments(
+	language: string,
+	options: { readonly jsEnvironments?: JsEnvironments },
+): { readonly environments?: PackagesInstallEnvironments } {
+	if (language === "js" && options.jsEnvironments !== undefined)
+		return { environments: { js: options.jsEnvironments } };
+	return {};
+}

package/src/tool/runtime-label.ts

@@ -10,7 +10,8 @@ const ELLIPSIS = "\u2026";
  * runtime name because node and bun are otherwise indistinguishable.
  */
 export function formatRuntimeBadge(language: EvalLanguage, runtime: EvalRuntimeInfo, home: string = homedir()): string {
-	const label = language === "js" ? `${runtime.name} ${runtime.version}` : runtime.version;
+	const base = language === "js" ? `${runtime.name} ${runtime.version}` : runtime.version;
+	const label = runtime.isolation === undefined ? base : `${base}, ${runtime.isolation}`;
 	if (runtime.path === undefined || runtime.path.length === 0) return label;
 	return `${label}, ${minifyPath(runtime.path, home)}`;
 }

package/src/tool/types.ts

@@ -52,6 +52,8 @@ export interface EvalToolInput {
 	readonly timeout?: number;
 	readonly on_timeout?: "detach" | "error";
 	readonly reset?: boolean;
+	/** Run in a fresh sandbox VM instead of the persistent kernel; accepted only while `sandbox.enabled` is on. */
+	readonly isolate?: boolean;
 }
 
 export interface EvalListInput {
@@ -117,14 +119,26 @@ const fullEvalInputSchema = Type.Object(
 /** Runtime accepts a discriminated run/control union. */
 export type EvalInputSchema = TUnsafe<EvalToolRequest> & Pick<typeof fullEvalInputSchema, "properties">;
 
+const ISOLATE_FIELD_DESCRIPTION =
+	"js only. Fresh sandbox per call: no persistence, no fs/net/process/timers; only tool.* plus print/display.";
+
 export function createEvalInputSchema(
 	enabled: EnabledEvalLanguages,
 	deadlines: EvalDeadlineSeconds = defaultEvalDeadlineSeconds,
+	options: { readonly sandbox?: boolean } = {},
 ): EvalInputSchema {
 	const languages = enabledLanguageList(enabled);
 	if (languages.length === 0) throw new Error("eval requires at least one enabled language");
 	const languageSchema = evalLanguageUnion(languages);
-	const properties = evalInputProperties(languageSchema, deadlines);
+	// The sandbox field exists in the schema only while sandbox cells are turned on, so the default schema
+	// (and the prompt built from it) stays exactly what it was.
+	const properties =
+		options.sandbox === true && languages.includes("js")
+			? {
+					...evalInputProperties(languageSchema, deadlines),
+					isolate: Type.Optional(Type.Boolean({ description: ISOLATE_FIELD_DESCRIPTION })),
+				}
+			: evalInputProperties(languageSchema, deadlines);
 	return Type.Unsafe<EvalToolRequest>(
 		Type.Object(properties, {
 			// Keep branches self-contained for Mistral-hosted GLM (#2240).
@@ -175,6 +189,12 @@ export interface EvalKernelRunInput {
 	readonly envRoot?: () => string;
 	/** The file a `%load` cell runs: tracebacks name it and its relative imports resolve from its directory. */
 	readonly sourceFile?: string;
+	/**
+	 * JavaScript only: the session's managed package revision; bare imports that do not resolve from cwd fall back to
+	 * it. Read when the cell starts running (not when it was queued), so a cell queued behind `%bun add` resolves the
+	 * revision that install published.
+	 */
+	readonly packageRoot?: () => string | undefined;
 	/**
 	 * Called when the cell's turn comes in the kernel's queue, never earlier: what to run instead of `code` (a `%load`
 	 * cell's file), or a refusal that settles the cell as failed in queue order.
@@ -259,6 +279,8 @@ export interface EvalRuntimeInfo {
 	readonly name: string;
 	readonly version: string;
 	readonly path?: string;
+	/** JS only: the kernel runs in a child process (`process`), or the cell ran in an isolated QuickJS VM (`sandbox`). */
+	readonly isolation?: "process" | "sandbox";
 }
 
 export type EvalRuntimes = Readonly<Partial<Record<EvalLanguage, EvalRuntimeInfo>>>;