@code-yeongyu/senpi-codemode 2026.10.8 → 2026.10.10-2

package/src/tool/eval-request.ts

@@ -17,9 +17,19 @@ export function normalizeEvalSummary(value: unknown): string | undefined {
 	return normalized.length === 0 ? undefined : normalized;
 }
 
+export class EvalIsolateInvalidError extends TypeError {
+	readonly name = "EvalIsolateInvalidError";
+	readonly code = "eval_isolate_invalid";
+
+	constructor(reason: string) {
+		super(`eval_isolate_invalid: ${reason}`);
+	}
+}
+
 export function parseEvalRequest(
 	params: unknown,
 	enabledLanguages: readonly EvalLanguage[] = evalLanguageOrder,
+	options: { readonly sandbox?: boolean } = {},
 ): EvalToolRequest {
 	if (!isRecord(params)) throw new TypeError("eval parameters must be an object");
 	if (params.action === "list") return { action: "list" };
@@ -30,7 +40,9 @@ export function parseEvalRequest(
 	}
 	if (params.action !== undefined && params.action !== "run")
 		throw new TypeError(`Unknown eval action "${String(params.action)}"`);
-	if (!isEvalLanguage(params.language)) throw new TypeError(evalRunRequiresLanguageMessage(enabledLanguages));
+	if (params.language === undefined) throw new TypeError(evalRunRequiresLanguageMessage(enabledLanguages));
+	if (!isEvalLanguage(params.language))
+		throw new TypeError(`eval run language must be one of: ${enabledLanguages.join(", ")}`);
 	if (typeof params.code !== "string")
 		throw new TypeError("eval run requires code — the cell body to execute, verbatim");
 	const summary = normalizeEvalSummary(params.summary);
@@ -40,6 +52,15 @@ export function parseEvalRequest(
 		);
 	if (params.on_timeout !== undefined && params.on_timeout !== "detach" && params.on_timeout !== "error")
 		throw new TypeError(`Unknown eval on_timeout value "${String(params.on_timeout)}"`);
+	if (params.isolate !== undefined && params.isolate !== false) {
+		if (params.isolate !== true) throw new EvalIsolateInvalidError("isolate must be true or false");
+		if (options.sandbox !== true) {
+			throw new EvalIsolateInvalidError("sandbox cells are turned off (set sandbox.enabled in codemode settings)");
+		}
+		if (params.language !== "js")
+			throw new EvalIsolateInvalidError('sandbox cells run JavaScript only (language: "js")');
+		if (params.reset === true) throw new EvalIsolateInvalidError("a sandbox cell starts fresh; reset does not apply");
+	}
 	return {
 		language: params.language,
 		code: params.code,
@@ -48,6 +69,7 @@ export function parseEvalRequest(
 		...(typeof params.timeout === "number" ? { timeout: params.timeout } : {}),
 		...(params.on_timeout === "detach" || params.on_timeout === "error" ? { on_timeout: params.on_timeout } : {}),
 		...(typeof params.reset === "boolean" ? { reset: params.reset } : {}),
+		...(params.isolate === true ? { isolate: true } : {}),
 	};
 }
 

package/src/tool/eval-tool-options.ts

@@ -8,6 +8,8 @@ import type {
 import type { EvalSchemaToolInfo } from "../bridges/schema-bridge.ts";
 import type { CompletionRequest, CompletionResult } from "../completion/handler.ts";
 import type { ResolvedCodemodeSettings } from "../config/settings.ts";
+import type { JsEnvironments } from "../environments/js-environments.ts";
+import type { PythonEnvironments } from "../environments/python-environments.ts";
 import type { EvalExecutionTracker } from "../extension/session-manager.ts";
 import type { HandleRegistry } from "../handles/handle-registry.ts";
 import type { EvalTimeoutFactory } from "./cell-execution.ts";
@@ -69,6 +71,10 @@ export interface CreateEvalToolOptions {
 	readonly bunSkillPath?: string;
 	/** Kernel globals of the tools active when a cell is submitted; read once per cell. */
 	readonly kernelPreludes?: () => readonly KernelPreludeContribution[];
+	/** The session's Python environments: `%pip` / `%environment` cells and the import root of every Python cell. */
+	readonly pythonEnvironments?: PythonEnvironments;
+	/** The session's managed JavaScript packages: `%bun add` / `%npm add` cells and the bare-import fallback root. */
+	readonly jsEnvironments?: JsEnvironments;
 	/** Contributions whose documentation lines the description lists; snapshot taken when the tool is (re)registered. */
 	readonly promptKernelPreludes?: readonly KernelPreludeContribution[];
 }

package/src/tool/eval-tool.ts

@@ -1,5 +1,6 @@
 import type { AgentToolResult, ToolDefinition } from "@code-yeongyu/senpi";
 import type { TUnsafe } from "typebox";
+import { resolveAdvertiseHelpers, resolveSandbox } from "../config/feature-settings.ts";
 import {
 	DEFAULT_FOREGROUND_WINDOW_SECONDS,
 	defaultCodemodeSettings,
@@ -8,7 +9,12 @@ import {
 import { buildEvalPrompt } from "../prompt/eval-prompt.ts";
 import { EvalDetachedCellManager } from "./detached-cell-manager.ts";
 import { executeEvalControl } from "./detached-eval-result.ts";
-import { isEvalControlRequest, normalizeEvalSummary, parseEvalRequest } from "./eval-request.ts";
+import {
+	EvalIsolateInvalidError,
+	isEvalControlRequest,
+	normalizeEvalSummary,
+	parseEvalRequest,
+} from "./eval-request.ts";
 import type { CreateEvalToolOptions } from "./eval-tool-options.ts";
 import { runEvalCell } from "./run-eval-cell.ts";
 import {
@@ -43,9 +49,11 @@ export function createEvalTool(options: CreateEvalToolOptions) {
 		foregroundWindowSeconds,
 		hardLimitSeconds: options.hardLimitSeconds ?? defaultEvalDeadlineSeconds.hardLimitSeconds,
 	};
-	const parameters = createEvalInputSchema(options.enabledLanguages, deadlines);
+	const sandbox = resolveSandbox(options.settings ?? defaultCodemodeSettings).enabled;
+	const parameters = createEvalInputSchema(options.enabledLanguages, deadlines, { sandbox });
 	const prompt = buildEvalPrompt(options.enabledLanguages, {
 		spawns: options.spawns ?? false,
+		...(resolveAdvertiseHelpers(options.settings ?? defaultCodemodeSettings) ? { advertiseHelpers: true } : {}),
 		monitor: options.monitor,
 		maxDetachedCells,
 		runBudgetSeconds: deadlines.runBudgetSeconds,
@@ -74,9 +82,14 @@ export function createEvalTool(options: CreateEvalToolOptions) {
 	async function execute(
 		...[toolCallId, params, signal, onUpdate, ctx]: EvalExecuteArgs<EvalToolRequest>
 	): Promise<AgentToolResult<EvalResultDetails>> {
-		const request = parseEvalRequest(params, languages);
+		const request = parseEvalRequest(params, languages, { sandbox });
 		if (isEvalControlRequest(request)) return await executeEvalControl(cellManager, request);
-		if (options.proxyExecutor) return await options.proxyExecutor(request, signal);
+		if (options.proxyExecutor) {
+			// A proxy runs cells elsewhere and knows nothing of sandbox cells: an isolated cell must never run unisolated.
+			if ("isolate" in request && request.isolate === true)
+				throw new EvalIsolateInvalidError("isolate: true is not available through this eval proxy");
+			return await options.proxyExecutor(request, signal);
+		}
 		if (!languages.includes(request.language))
 			throw new RangeError(
 				`Unsupported eval language "${request.language}". Enabled languages: ${languages.join(", ")}`,

package/src/tool/load-cell.ts

@@ -0,0 +1,88 @@
+import { closeSync, constants, fstatSync, openSync, readSync } from "node:fs";
+import { isAbsolute, join, relative, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+
+export type LoadedCell =
+	| { readonly ok: true; readonly code: string; readonly sourceFile: string }
+	| { readonly ok: false; readonly message: string };
+
+export interface LoadCellOptions {
+	readonly cwd: string;
+	readonly artifactsDir: string | undefined;
+}
+
+const MAX_LOAD_BYTES = 8 * 1024 * 1024;
+const URL_SCHEME = /^([a-z][a-z0-9+.-]*):\/\//i;
+
+function errorCode(error: unknown): string | undefined {
+	return typeof error === "object" && error !== null && "code" in error && typeof error.code === "string"
+		? error.code
+		: undefined;
+}
+
+function resolveTarget(target: string, options: LoadCellOptions): { path: string } | { message: string } {
+	const scheme = URL_SCHEME.exec(target)?.[1]?.toLowerCase();
+	try {
+		if (scheme === "file") return { path: fileURLToPath(target) };
+		if (scheme === "local") {
+			if (options.artifactsDir === undefined)
+				return { message: "%load local:// needs a session artifacts directory" };
+			const root = join(options.artifactsDir, "local");
+			const path = resolve(root, decodeURIComponent(target.slice("local://".length)));
+			const inside = relative(root, path);
+			if (inside.startsWith("..") || isAbsolute(inside))
+				return { message: `%load path escapes local://: ${target}` };
+			return { path };
+		}
+	} catch {
+		// fileURLToPath and decodeURIComponent throw on a malformed URL; the message names the user's own text.
+		return { message: `%load could not read the path in ${target}` };
+	}
+	if (scheme !== undefined) {
+		return {
+			message: `%load reads local files only (a path, local:// or file://); it does not fetch ${scheme}:// URLs`,
+		};
+	}
+	return { path: resolve(options.cwd, target) };
+}
+
+/**
+ * Reads a `%load` target when its cell's turn comes in the kernel's queue, so a file the previous cell wrote is
+ * there. Every failure is a refusal that names the target as the user wrote it, never an absolute path.
+ */
+/** Reads from `fd` until EOF or `limit` bytes, whichever comes first; a file still growing cannot exceed it. */
+export function readAtMost(fd: number, limit: number): Buffer {
+	const buffer = Buffer.alloc(limit);
+	let filled = 0;
+	while (filled < limit) {
+		const read = readSync(fd, buffer, filled, limit - filled, null);
+		if (read === 0) break;
+		filled += read;
+	}
+	return buffer.subarray(0, filled);
+}
+
+export function loadCell(target: string, options: LoadCellOptions): LoadedCell {
+	const resolved = resolveTarget(target, options);
+	if ("message" in resolved) return { ok: false, message: resolved.message };
+	let fd: number | undefined;
+	try {
+		// One non-blocking open, then checks on that same handle: a FIFO or device swapped in after a check can
+		// neither block the host nor be read, and the size cap keeps a huge file from stalling the event loop.
+		fd = openSync(resolved.path, constants.O_RDONLY | constants.O_NONBLOCK);
+		const info = fstatSync(fd);
+		if (!info.isFile()) return { ok: false, message: `not a file: ${target}` };
+		const tooLarge = { ok: false as const, message: `%load reads files up to 8 MiB: ${target}` };
+		if (info.size > MAX_LOAD_BYTES) return tooLarge;
+		const bytes = readAtMost(fd, MAX_LOAD_BYTES + 1);
+		if (bytes.length > MAX_LOAD_BYTES) return tooLarge;
+		return { ok: true, code: bytes.toString("utf8"), sourceFile: resolved.path };
+	} catch (error) {
+		const code = errorCode(error);
+		if (code === "ENOENT" || code === "ENOTDIR") return { ok: false, message: `file not found: ${target}` };
+		if (code === "EACCES" || code === "EPERM") return { ok: false, message: `permission denied: ${target}` };
+		return { ok: false, message: `%load could not read ${target}${code === undefined ? "" : ` (${code})`}` };
+	} finally {
+		if (fd !== undefined) closeSync(fd);
+	}
+}

package/src/tool/magic-cell-host.ts

@@ -0,0 +1,123 @@
+import type { JsEnvironments } from "../environments/js-environments.ts";
+import type { PythonEnvironments } from "../environments/python-environments.ts";
+import { TIMEOUT_PAUSE_OP, TIMEOUT_RESUME_OP } from "../timeouts/bridge-timeout.ts";
+import { MagicCellError, parseMagicCell } from "./magic-cells.ts";
+import type { EvalLanguage, HostCellExecutor } from "./types.ts";
+
+export type MagicCellPlan =
+	| { readonly kind: "ordinary" }
+	| { readonly kind: "host"; readonly executor: HostCellExecutor }
+	| { readonly kind: "refused"; readonly message: string }
+	| { readonly kind: "load"; readonly target: string };
+
+export function planMagicCell(
+	language: EvalLanguage,
+	code: string,
+	environments: PythonEnvironments | undefined,
+	jsEnvironments?: JsEnvironments,
+): MagicCellPlan {
+	let magic: ReturnType<typeof parseMagicCell>;
+	try {
+		magic = parseMagicCell(language, code);
+	} catch (error) {
+		if (error instanceof MagicCellError) return { kind: "refused", message: error.message };
+		throw error;
+	}
+	if (magic === undefined) return { kind: "ordinary" };
+	if (magic.kind === "load") return { kind: "load", target: magic.target };
+	if (language === "js") return planJsMagic(magic, jsEnvironments);
+	if (environments === undefined) {
+		return {
+			kind: "refused",
+			message: "environment_installer_unavailable: this session has no Python interpreter to install packages for",
+		};
+	}
+	if (magic.kind === "environment") {
+		const mode = magic.mode;
+		return {
+			kind: "host",
+			executor: async () => {
+				await environments.setMode(mode);
+				const where = mode === "project" ? "the session directory" : "this session's managed revisions";
+				return { ok: true, valueRepr: `environment: ${mode} (${where})` };
+			},
+		};
+	}
+	const requirements = magic.args;
+	return {
+		kind: "host",
+		// Install time is parked time for the run budget, the same accounting as a bridge call.
+		executor: async ({ signal, emit }) => {
+			emit({ type: "status", event: { op: TIMEOUT_PAUSE_OP } });
+			try {
+				const receipt = await environments.install(requirements, signal, (stream, data) =>
+					emit({ type: "text", stream, data }),
+				);
+				const packages = receipt.resolved.length > 0 ? receipt.resolved.join(", ") : "nothing new";
+				return {
+					ok: true,
+					valueRepr: `installed ${packages} into ${receipt.mode} (revision ${receipt.revision}); already-imported modules stay cached until reset`,
+				};
+			} catch (error) {
+				const message = error instanceof Error ? error.message : String(error);
+				return { ok: false, error: { message } };
+			} finally {
+				emit({ type: "status", event: { op: TIMEOUT_RESUME_OP } });
+			}
+		},
+	};
+}
+
+function planJsMagic(
+	magic: Exclude<NonNullable<ReturnType<typeof parseMagicCell>>, { kind: "load" }>,
+	environments: JsEnvironments | undefined,
+): MagicCellPlan {
+	if (environments === undefined) {
+		return {
+			kind: "refused",
+			message: "environment_installer_unavailable: this session has no JavaScript package environment",
+		};
+	}
+	if (magic.kind === "environment") {
+		const mode = magic.mode;
+		return {
+			kind: "host",
+			executor: async () => {
+				await environments.setMode(mode);
+				const where = mode === "project" ? "the session directory" : "this session's managed revisions";
+				return { ok: true, valueRepr: `environment: ${mode} (${where})` };
+			},
+		};
+	}
+	if (magic.kind !== "js-add") return { kind: "refused", message: "this magic is not available in JavaScript" };
+	const requested = magic.args;
+	return {
+		kind: "host",
+		executor: async ({ signal, emit }) => {
+			emit({ type: "status", event: { op: TIMEOUT_PAUSE_OP } });
+			try {
+				const receipt = await environments.install(
+					requested,
+					signal,
+					(stream, data) => emit({ type: "text", stream, data }),
+					magic.installer,
+				);
+				const added = receipt.added.length > 0 ? receipt.added.join(", ") : "nothing new";
+				const where =
+					receipt.revision === undefined ? receipt.mode : `${receipt.mode} (revision ${receipt.revision})`;
+				const shadow =
+					receipt.shadowed.length > 0
+						? `; environment_resolution_conflict: ${receipt.shadowed.join(", ")} still ${receipt.shadowed.length === 1 ? "resolves" : "resolve"} from the project's node_modules first`
+						: "";
+				return {
+					ok: true,
+					valueRepr: `added ${added} with ${receipt.installer} into ${where}${shadow}; already-imported modules stay cached until reset`,
+				};
+			} catch (error) {
+				return { ok: false, error: { message: error instanceof Error ? error.message : String(error) } };
+			} finally {
+				emit({ type: "status", event: { op: TIMEOUT_RESUME_OP } });
+			}
+		},
+	};
+}

package/src/tool/magic-cells.ts

@@ -0,0 +1,75 @@
+import type { EvalLanguage } from "./types.ts";
+
+export type MagicCell =
+	| { readonly kind: "pip"; readonly args: string }
+	| { readonly kind: "environment"; readonly mode: "managed" | "project" }
+	| { readonly kind: "load"; readonly target: string }
+	| { readonly kind: "js-add"; readonly installer: "bun" | "npm"; readonly args: string };
+
+const HOST_MAGICS = ["pip", "environment", "load", "bun", "npm"] as const;
+const LOAD_LANGUAGES: ReadonlySet<EvalLanguage> = new Set(["py", "js"]);
+const COMMENT_PREFIX: Partial<Record<EvalLanguage, string>> = { py: "#", js: "//" };
+type HostMagic = (typeof HOST_MAGICS)[number];
+
+export class MagicCellError extends Error {
+	readonly name = "MagicCellError";
+}
+
+function joinContinuations(lines: readonly string[]): string[] {
+	const joined: string[] = [];
+	let pending: string | undefined;
+	for (const line of lines) {
+		const current = pending === undefined ? line : `${pending} ${line.trim()}`;
+		if (current.trimEnd().endsWith("\\")) pending = current.trimEnd().slice(0, -1).trimEnd();
+		else {
+			joined.push(current);
+			pending = undefined;
+		}
+	}
+	if (pending !== undefined) joined.push(pending);
+	return joined;
+}
+
+function hostMagicOf(language: EvalLanguage, line: string): HostMagic | undefined {
+	const match = /^%([A-Za-z]+)(?:\s|$)/.exec(line.trim());
+	const name = match?.[1];
+	const magic = HOST_MAGICS.find((candidate) => candidate === name);
+	if (magic === "load") return LOAD_LANGUAGES.has(language) ? magic : undefined;
+	if (magic === "bun" || magic === "npm") return language === "js" ? magic : undefined;
+	if (magic === "environment") return language === "py" || language === "js" ? magic : undefined;
+	return language === "py" ? magic : undefined;
+}
+
+/**
+ * A cell whose first code line (blank and comment lines skipped) is a host magic runs on the host instead of
+ * the interpreter: `%pip` and `%environment` in Python, `%load` in Python and JavaScript. In Python a trailing
+ * backslash continues the line. Any other cell is ordinary code, so a magic-looking line later in the cell
+ * (say, inside a string) is left alone. A magic followed by more code is refused, because the host step must
+ * finish before the code that depends on it runs.
+ */
+export function parseMagicCell(language: EvalLanguage, code: string): MagicCell | undefined {
+	const comment = COMMENT_PREFIX[language];
+	// Comment lines go before Python's backslash continuations are joined: Python never continues a comment line.
+	const raw = code.split("\n").filter((line) => comment === undefined || !line.trim().startsWith(comment));
+	const lines = (language === "py" ? joinContinuations(raw) : raw).filter((line) => line.trim() !== "");
+	const first = lines[0] ?? "";
+	const magic = hostMagicOf(language, first);
+	if (magic === undefined) return undefined;
+	if (lines.length > 1)
+		throw new MagicCellError(`put %${magic} on its own cell, then run the code that uses it in the next cell`);
+	const args = first.trim().slice(`%${magic}`.length).trim();
+	if (magic === "pip") return { kind: "pip", args };
+	if (magic === "bun" || magic === "npm") {
+		const [verb, ...rest] = args.split(/\s+/);
+		if (verb !== "add" && verb !== "install") {
+			throw new MagicCellError(`%${magic} supports only add: %${magic} add <package ...>`);
+		}
+		return { kind: "js-add", installer: magic, args: rest.join(" ") };
+	}
+	if (magic === "load") {
+		if (args === "") throw new MagicCellError("%load takes one argument: the path of the file to run");
+		return { kind: "load", target: args };
+	}
+	if (args === "managed" || args === "project") return { kind: "environment", mode: args };
+	throw new MagicCellError("%environment takes one argument: managed or project");
+}

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

@@ -7,12 +7,14 @@ 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 {
 	KERNEL_TOOLS_CAPABILITIES,
 	type KernelToolsCapability,
 	type KernelToolsDescribeResult,
 } from "../kernels/js/kernel-tools-types.ts";
+import { sandboxCellExecutor } 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";
@@ -23,6 +25,8 @@ import { EvalKernelResetRefusedError } from "./eval-kernel-reset-refused-error.t
 import { evalTimeoutBehavior } from "./eval-request.ts";
 import type { CreateEvalToolOptions, EvalCellInvocation } from "./eval-tool-options.ts";
 import { describeTimeoutState } from "./interrupt-note.ts";
+import { loadCell } from "./load-cell.ts";
+import { planMagicCell } from "./magic-cell-host.ts";
 import type { EvalKernel, EvalToolDetails } from "./types.ts";
 
 export async function runEvalCell(
@@ -205,11 +209,21 @@ async function executeCell(
 		const pending = handler.handle(message);
 		void pending.catch((error: unknown) => execution.cancel(error));
 	};
+	let releaseCellKernelTools: (() => void) | undefined;
 	try {
 		const kernel = await execution.wait(options.kernelManager.getKernel(invocation.input.language, onMessage));
 		// Computed before the handler so its construction-time snapshot captures this cell's capability.
 		// Worker messages later restore that snapshot before calling host tools (#1754, #2512).
 		const kernelTools = kernelToolsFor(kernel);
+		// Subprocess kernels call the host over the bridge, outside this async context. Their calls carry a fresh secret
+		// minted for this run and sent only to this cell's kernel, never the model-visible cell id, so code in another
+		// kernel cannot name this cell into its tools. JS calls already run inside this cell's scope (#1754).
+		const bridgeCellToken =
+			kernelTools === undefined || invocation.input.language === "js" ? undefined : randomUUID();
+		releaseCellKernelTools =
+			kernelTools === undefined || bridgeCellToken === undefined
+				? undefined
+				: options.kernelManager.bindCellKernelTools?.(bridgeCellToken, kernelTools);
 		const runBound = async (): Promise<AgentToolResult<EvalToolDetails>> => {
 			const queue = kernel.queueSnapshot();
 			state.queuedBehind = [...(queue.activeCellId === null ? [] : [queue.activeCellId]), ...queue.queuedCellIds];
@@ -249,10 +263,51 @@ 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,
+				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 };
+			// 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,
 					onStarted: () => {
@@ -285,6 +340,8 @@ async function executeCell(
 		if (error instanceof Error && error.name === "TimeoutError") throw await describeTimeoutState(error, execution);
 		throw error;
 	} finally {
+		// First, so nothing below can skip it: from here on a host call carrying this run's secret gets no kernel tools.
+		releaseCellKernelTools?.();
 		state.active = false;
 		bridgeAbortController.abort();
 		execution.finish();

package/src/tool/types.ts

@@ -8,6 +8,7 @@ import {
 	DEFAULT_RUN_BUDGET_SECONDS,
 	defaultCodemodeSettings,
 } from "../config/settings.ts";
+import type { KernelToolsCapability } from "../kernels/js/kernel-tools-types.ts";
 import type { TruncationMeta } from "../output/output-meta.ts";
 
 export const evalLanguageOrder = ["js", "py", "rb", "jl"] as const;
@@ -51,6 +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 {
@@ -116,14 +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).
@@ -167,8 +182,35 @@ export interface EvalKernelRunInput {
 	readonly onMessage?: (message: KernelToHostMessage) => void;
 	/** Globals of the tools active when the cell was submitted; kernels without preludes ignore them. */
 	readonly kernelPreludes?: readonly KernelPreludeContribution[];
+	/**
+	 * Python only: read when the cell starts running (not when it was queued), so a cell queued behind `%pip`
+	 * sees the revision it published. `""` means no environment: the previous revision leaves the import path.
+	 */
+	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.
+	 */
+	readonly resolveAtStart?: () => CellSourceAtStart;
+	/**
+	 * Subprocess kernels attach this to the cell's host calls, and the host gives those calls the cell's kernel tools.
+	 * A fresh secret per run, sent only to the kernel that runs the cell; never a model-visible id.
+	 */
+	readonly bridgeCellToken?: string;
 }
 
+export type CellSourceAtStart =
+	| { readonly ok: true; readonly code: string; readonly sourceFile?: string }
+	| { readonly ok: false; readonly message: string };
+
 export interface KernelInterruptHandle {
 	/** Resolves once the kernel knows whether user state survived the interrupt. */
 	readonly stateRetained: Promise<boolean>;
@@ -205,6 +247,11 @@ export interface EvalKernelManager {
 	 * Identity-checked, so releasing a superseded listener never unbinds a newer cell's listener.
 	 */
 	releaseKernelListener?(language: EvalLanguage, onMessage: (message: KernelToHostMessage) => void): void;
+	/**
+	 * Binds a running cell's kernel-tools capability to the host calls that carry `token` (subprocess kernels reach
+	 * the host over the bridge, outside the cell's async context). The returned release runs when the cell settles.
+	 */
+	bindCellKernelTools?(token: string, capability: KernelToolsCapability): () => void;
 }
 
 export type ExecuteTool = (