@code-yeongyu/senpi-codemode 2026.10.8 → 2026.10.10-10

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 = (
@@ -232,6 +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>>>;