@selesai/code 0.13.40 → 0.14.3

package/dist/extensions/codemode/execute.js

@@ -0,0 +1,645 @@
+/**
+ * Runs one codemode script in the sandbox. Split from tool.ts and loaded through
+ * execute.lazy.ts so the sandbox runtime only loads when a script runs.
+ */
+import { CodemodeSandbox, loadQuickJSWasm, parseCodemodeSource, renderToolSample, toCodemodeIdentifier, } from "@earendil-works/pi-codemode";
+import { getCodemodeWorkerSpecifier, getQuickJSWasmPath } from "../../config.js";
+import { formatSize } from "../../core/tools/truncate.js";
+import { combineUsage } from "../../core/usage-totals.js";
+import { writeOutputFile } from "../../utils/output-files.js";
+import { createToolSearchDocument, DEFAULT_TOOL_SEARCH_LIMIT, getToolRanker } from "../tool-search/tool.js";
+import { CODEMODE_DOCS_PATH, CODEMODE_STORE_ENTRY_TYPE, getCodemodeCallableTools, toCodemodeDeclaration, } from "./tool.js";
+const ARGS_PREVIEW_CHARS = 200;
+const ERROR_PREVIEW_CHARS = 500;
+/** `models.classify()` and `models.generateImages()` calls one script may have in flight; `Promise.all` over many items queues the rest. */
+const MAX_CONCURRENT_MODEL_CALLS = 4;
+/** `searchTools()` calls per script that may use a model-backed ranker; later calls rank locally. */
+const MAX_SESSION_SEARCHES_PER_SCRIPT = 8;
+/**
+ * Heap limit for the QuickJS VM. The worker shares pi's process, so without a limit a runaway
+ * script can grow to wasm32's 4 GiB and take the session down. Overruns throw
+ * `InternalError: out of memory` inside the script.
+ */
+const CODEMODE_MEMORY_LIMIT_BYTES = 256 * 1024 * 1024;
+const MODEL_TYPES = new Set(["chat", "image", "classifier"]);
+function truncateText(text, maxChars) {
+    return text.length > maxChars ? `${text.slice(0, maxChars - 3)}...` : text;
+}
+function previewArgs(args) {
+    if (args === undefined)
+        return "";
+    try {
+        return truncateText(JSON.stringify(args) ?? "", ARGS_PREVIEW_CHARS);
+    }
+    catch {
+        return "";
+    }
+}
+function textOf(result) {
+    return (result.content ?? [])
+        .filter((block) => block.type === "text")
+        .map((block) => block.text)
+        .join("\n");
+}
+function toModelType(value) {
+    if (typeof value === "string" && MODEL_TYPES.has(value))
+        return value;
+    throw new Error(`Unknown model type ${JSON.stringify(value)}. Use "chat", "image", or "classifier".`);
+}
+function toProvider(value) {
+    if (value === undefined || value === null)
+        return undefined;
+    if (typeof value !== "string")
+        throw new Error("provider must be a string");
+    return value;
+}
+/** Catalog entry for scripts. `headers` is dropped because models.json headers can carry credentials. */
+function toModelInfo(model) {
+    const info = { ...model };
+    delete info.headers;
+    return info;
+}
+/** `an image`, `a classifier`. */
+function withArticle(word) {
+    return `${/^[aeiou]/.test(word) ? "an" : "a"} ${word}`;
+}
+function isRecord(value) {
+    return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+/** A script value in an error message: `undefined`, `a string`, `an array`, or its keys (`{ prompt }`). */
+function describeValue(value) {
+    if (value === undefined || value === null)
+        return String(value);
+    if (Array.isArray(value))
+        return value.length === 0 ? "an empty array" : "an array";
+    if (typeof value === "object") {
+        const keys = Object.keys(value);
+        if (keys.length === 0)
+            return "{}";
+        return `{ ${keys.slice(0, 6).join(", ")}${keys.length > 6 ? ", ..." : ""} }`;
+    }
+    return typeof value === "string" ? "a string" : `a ${typeof value}`;
+}
+const CLASSIFIER_CONTEXT_SHAPE = '{ state: { ... }, images?: [{ type: "image", data: <base64>, mimeType }], questions: { <id>: { type: "choice", instructions, criteria: { <label>: <meaning> } } | { type: "score", instructions, criteria: [<lowest level>, ..., <highest level>] } | { type: "bool", instructions, criteria: { true: <meaning>, false: <meaning> } } } }';
+/** Check a script's classifier context, so mistakes fail with the expected shape instead of a provider error. */
+function checkClassifierContext(context) {
+    const fail = (problem) => new Error(`models.classify() ${problem}. Expected context: ${CLASSIFIER_CONTEXT_SHAPE}. See "Classify" in ${CODEMODE_DOCS_PATH}.`);
+    if (!isRecord(context))
+        throw fail(`expects a context object as its second argument, got ${describeValue(context)}`);
+    if (!isRecord(context.state))
+        throw fail(`context.state must be an object, got ${describeValue(context.state)}`);
+    const { images } = context;
+    if (images !== undefined) {
+        if (!Array.isArray(images))
+            throw fail(`context.images must be an array, got ${describeValue(images)}`);
+        images.forEach((image, index) => {
+            if (!isRecord(image) ||
+                image.type !== "image" ||
+                typeof image.data !== "string" ||
+                typeof image.mimeType !== "string") {
+                throw fail(`context.images[${index}] must be an image block, got ${describeValue(image)}`);
+            }
+        });
+    }
+    const { questions } = context;
+    if (!isRecord(questions) || Object.keys(questions).length === 0) {
+        throw fail(`context.questions must map question IDs to questions, got ${describeValue(questions)}`);
+    }
+    const isStrings = (values) => values.length > 0 && values.every((value) => typeof value === "string");
+    for (const [id, question] of Object.entries(questions)) {
+        const at = `context.questions.${id}`;
+        if (!isRecord(question))
+            throw fail(`${at} must be a question object, got ${describeValue(question)}`);
+        if (typeof question.instructions !== "string")
+            throw fail(`${at}.instructions must be a string`);
+        const { criteria } = question;
+        if (question.type === "choice") {
+            if (!isRecord(criteria) || !isStrings(Object.values(criteria))) {
+                throw fail(`${at} is a "choice" question, so criteria must map each label to its meaning`);
+            }
+        }
+        else if (question.type === "score") {
+            if (!Array.isArray(criteria) || !isStrings(criteria)) {
+                throw fail(`${at} is a "score" question, so criteria must list the levels as strings, lowest first`);
+            }
+        }
+        else if (question.type === "bool") {
+            if (!isRecord(criteria) || typeof criteria.true !== "string" || typeof criteria.false !== "string") {
+                throw fail(`${at} is a "bool" question, so criteria must be { true: string, false: string }`);
+            }
+        }
+        else {
+            throw fail(`${at}.type must be "choice", "score", or "bool", got ${JSON.stringify(question.type)}`);
+        }
+    }
+    return context;
+}
+/** Check a script's image context, so mistakes such as `{ prompt }` fail with the expected shape. */
+function checkImagesContext(context) {
+    const fail = (problem) => new Error(`models.generateImages() ${problem}. Expected context: { input: [{ type: "text", text: <prompt> }, ...optional { type: "image", data: <base64>, mimeType } references] }. See "Generate images" in ${CODEMODE_DOCS_PATH}.`);
+    if (!isRecord(context))
+        throw fail(`expects a context object as its second argument, got ${describeValue(context)}`);
+    const { input } = context;
+    if (!Array.isArray(input) || input.length === 0) {
+        throw fail(`context.input must be a non-empty array of blocks, got ${describeValue(input)}`);
+    }
+    input.forEach((block, index) => {
+        if (isRecord(block) && block.type === "text" && typeof block.text === "string")
+            return;
+        if (isRecord(block) &&
+            block.type === "image" &&
+            typeof block.data === "string" &&
+            typeof block.mimeType === "string") {
+            return;
+        }
+        throw fail(`context.input[${index}] must be a text or image block, got ${describeValue(block)}`);
+    });
+    return context;
+}
+/** Runs at most `limit` calls at once, in call order. */
+function createLimiter(limit) {
+    let active = 0;
+    const waiting = [];
+    return async (run) => {
+        if (active >= limit)
+            await new Promise((resolve) => waiting.push(resolve));
+        active++;
+        try {
+            return await run();
+        }
+        finally {
+            active--;
+            waiting.shift()?.();
+        }
+    };
+}
+function isStoreEntryData(data) {
+    if (typeof data !== "object" || data === null)
+        return false;
+    const { set, delete: deleted } = data;
+    return (typeof set === "object" &&
+        set !== null &&
+        Array.isArray(deleted) &&
+        deleted.every((key) => typeof key === "string"));
+}
+/** Values of `load()`: the `codemode-store` entries on the branch, applied from the root. */
+export function readCodemodeStore(branch) {
+    const store = new Map();
+    for (const entry of branch) {
+        if (entry.type !== "custom" || entry.customType !== CODEMODE_STORE_ENTRY_TYPE || !isStoreEntryData(entry.data)) {
+            continue;
+        }
+        for (const key of entry.data.delete)
+            store.delete(key);
+        for (const [key, value] of Object.entries(entry.data.set))
+            store.set(key, value);
+    }
+    return Object.fromEntries(store);
+}
+/** Default token budget for script output. */
+const DEFAULT_MAX_OUTPUT_TOKENS = 10_000;
+/** Characters per token when estimating. */
+const CHARS_PER_TOKEN = 4;
+/** Like the script's `text()`: strings as is, other values as compact JSON. */
+function valueText(value) {
+    if (typeof value === "string")
+        return value;
+    return JSON.stringify(value) ?? String(value);
+}
+/**
+ * Lay out the script's output so the model can tell items apart: providers join adjacent text
+ * blocks with a newline or with nothing. With more than one text item (`text()` or the returned
+ * value), each starts with a `==> text N/M <==` line. `console.*` lines follow all other output in
+ * one `<console_output>` block.
+ */
+function formatOutput(output) {
+    const total = output.filter((item) => item.type === "text" && !item.console).length;
+    const items = [];
+    const consoleLines = [];
+    let index = 0;
+    for (const item of output) {
+        if (item.type === "image") {
+            items.push(item);
+        }
+        else if (item.console) {
+            consoleLines.push(item.text);
+        }
+        else {
+            index++;
+            items.push({ type: "text", text: total > 1 ? `==> text ${index}/${total} <==\n${item.text}` : item.text });
+        }
+    }
+    if (consoleLines.length > 0) {
+        items.push({ type: "text", text: `<console_output>\n${consoleLines.join("\n")}\n</console_output>` });
+    }
+    return items;
+}
+/** Join adjacent text items into one, each part starting on its own line. */
+function joinAdjacentText(items) {
+    const joined = [];
+    for (const item of items) {
+        const last = joined.at(-1);
+        if (item.type === "text" && last?.type === "text") {
+            const separator = last.text === "" || last.text.endsWith("\n") ? "" : "\n";
+            joined[joined.length - 1] = { type: "text", text: `${last.text}${separator}${item.text}` };
+        }
+        else {
+            joined.push(item);
+        }
+    }
+    return joined;
+}
+function formatCallSummary(calls) {
+    if (calls.length === 0)
+        return "No tool calls were made.";
+    return `Tool calls made before the failure (they are not undone): ${calls.map((call) => `${call.name} (${call.status})`).join(", ")}`;
+}
+function formatError(result, calls) {
+    const { error } = result;
+    const head = error.kind === "script"
+        ? (error.stack ?? `${error.name ?? "Error"}: ${error.message}`)
+        : error.kind === "timeout"
+            ? `Script timed out: ${error.message}`
+            : error.kind === "aborted"
+                ? `Script aborted: ${error.message}`
+                : `Script sandbox failed: ${error.message}`;
+    return `${head}\n\n${formatCallSummary(calls)}`;
+}
+/** Write the full text output to a temp file, like bash does for truncated output. */
+async function spillOutput(text) {
+    try {
+        return { path: await writeOutputFile("pi-codemode", ".txt", text) };
+    }
+    catch (error) {
+        return { error: error instanceof Error ? error.message : String(error) };
+    }
+}
+/** File extensions of the image types `image()` accepts. Must list every type the sandbox's `image()` detects. */
+const IMAGE_EXTENSIONS = {
+    "image/png": ".png",
+    "image/jpeg": ".jpg",
+    "image/gif": ".gif",
+    "image/webp": ".webp",
+};
+/**
+ * Save each image to a temp file and put a text item with its path before it. The model sees the
+ * image but has no other way to reach its bytes: scripts cannot write files, and `write` only takes
+ * text. Images shown more than once are saved once.
+ */
+async function saveImages(items) {
+    const labels = new Map();
+    const label = async ({ data, mimeType }) => {
+        const bytes = Buffer.from(data, "base64");
+        const kind = `${mimeType}, ${formatSize(bytes.length)}`;
+        const extension = IMAGE_EXTENSIONS[mimeType];
+        if (!extension)
+            throw new Error(`No file extension for image type ${mimeType}`);
+        // A failed write (disk full, unwritable temp dir) must not discard the result of a script whose
+        // tool calls already ran, so it becomes part of the label.
+        try {
+            const path = await writeOutputFile("pi-codemode", extension, bytes);
+            return `[Image saved to ${path} (${kind})]`;
+        }
+        catch (error) {
+            return `[Image (${kind}) could not be saved: ${error instanceof Error ? error.message : String(error)}]`;
+        }
+    };
+    const result = await Promise.all(items.map(async (item) => {
+        if (item.type !== "image")
+            return [item];
+        let pending = labels.get(item.data);
+        if (!pending) {
+            pending = label(item);
+            labels.set(item.data, pending);
+        }
+        return [{ type: "text", text: await pending }, item];
+    }));
+    return result.flat();
+}
+/**
+ * Apply the token budget: when the combined text exceeds it, the text items become one
+ * item that keeps the start and end of the text, and images follow it. The full text is written to
+ * a temp file.
+ */
+async function truncateOutput(items, maxTokens) {
+    const texts = items.filter((item) => item.type === "text").map((item) => item.text);
+    const combined = texts.join("\n");
+    const budget = maxTokens * CHARS_PER_TOKEN;
+    if (texts.length === 0 || combined.length <= budget)
+        return { items };
+    const headChars = Math.floor(budget / 2);
+    const tailChars = budget - headChars;
+    const removed = combined.length - headChars - tailChars;
+    const head = combined.slice(0, headChars);
+    const tail = tailChars > 0 ? combined.slice(-tailChars) : "";
+    let text = `Warning: truncated output (original token count: ${Math.ceil(combined.length / CHARS_PER_TOKEN)})\nTotal output lines: ${combined.split("\n").length}\n\n${head}…${Math.ceil(removed / CHARS_PER_TOKEN)} tokens truncated…${tail}`;
+    const spilled = await spillOutput(combined);
+    text +=
+        "path" in spilled
+            ? `\n\n[Full output: ${spilled.path} (read with offset/limit)]`
+            : `\n\n[Could not save the full output: ${spilled.error}]`;
+    return {
+        items: [{ type: "text", text }, ...items.filter((item) => item.type === "image")],
+        ...("path" in spilled ? { fullOutputPath: spilled.path } : {}),
+    };
+}
+/**
+ * The value a script receives for a nested call: a tool that declares
+ * `outputSchema` resolves to its `structuredContent`, also for error results that carry one (such
+ * as MCP results with `isError`); any other tool resolves to its text content. Other failures
+ * reject with the tool's error text.
+ */
+function toScriptValue(tool, outcome) {
+    const { result } = outcome;
+    if (tool.outputSchema && result.structuredContent !== undefined)
+        return result.structuredContent;
+    const text = textOf(result);
+    if (outcome.isError)
+        throw new Error(text || `Tool "${tool.name}" failed`);
+    return text;
+}
+/**
+ * Run one script. Without a session context (a plain Agent or a direct call) scripts cannot call
+ * tools, `store()` starts empty, and writes are dropped.
+ */
+export async function executeCodemode(toolCallId, input, signal, onUpdate, ctx, options = {}) {
+    const startedAt = performance.now();
+    const { code, options: sourceOptions } = parseCodemodeSource(input.code);
+    const calls = [];
+    // Usage of the script's `models.*` calls. Nested tool calls report theirs through the session.
+    let modelUsage;
+    // Images returned by `models.generateImages()`, to notice a script that never shows them.
+    let generatedImages = 0;
+    const addGeneratedImages = (count) => {
+        generatedImages += count;
+    };
+    const addModelUsage = (usage) => {
+        modelUsage = modelUsage ? combineUsage(modelUsage, usage) : usage;
+    };
+    const snapshot = () => ({ calls: calls.map((call) => ({ ...call })) });
+    const publish = () => onUpdate?.({ content: [], details: snapshot() });
+    const callable = ctx ? getCodemodeCallableTools(ctx.tools) : [];
+    // ALL_TOOLS entries carry the declaration.
+    const guidelines = options.getToolGuidelines?.();
+    const samples = new Map(callable.map((tool) => [tool.name, renderToolSample(toCodemodeDeclaration(tool, guidelines?.get(tool.name)))]));
+    const sandboxTools = callable.map((tool) => ({
+        name: tool.name,
+        description: samples.get(tool.name),
+        execute: async (args, { signal: callSignal }) => {
+            const record = {
+                id: `${toolCallId}/?`,
+                name: tool.name,
+                args: previewArgs(args),
+                status: "running",
+            };
+            calls.push(record);
+            publish();
+            const callStartedAt = performance.now();
+            // Only tools from ctx.tools are callable, so ctx is set here.
+            if (!ctx)
+                throw new Error("Tool calls need a session");
+            const outcome = await ctx.executeTool(tool.name, args, { signal: callSignal });
+            record.id = outcome.toolCall.id;
+            record.durationMs = performance.now() - callStartedAt;
+            if (outcome.isError) {
+                record.status = callSignal.aborted ? "cancelled" : "error";
+                record.error = truncateText(textOf(outcome.result) || `Tool "${tool.name}" failed`, ERROR_PREVIEW_CHARS);
+            }
+            else {
+                record.status = "ok";
+            }
+            publish();
+            return toScriptValue(tool, outcome);
+        },
+    }));
+    const sandbox = new CodemodeSandbox({
+        tools: sandboxTools,
+        globals: [
+            ...createDiscoveryGlobals(callable, samples, options, ctx),
+            ...(options.models && ctx
+                ? createModelGlobals(ctx.modelRegistry, toolCallId, calls, publish, addModelUsage, addGeneratedImages)
+                : []),
+        ],
+        timeoutMs: sourceOptions.timeoutMs ?? Number.POSITIVE_INFINITY,
+        memoryLimitBytes: CODEMODE_MEMORY_LIMIT_BYTES,
+        wasm: loadQuickJSWasm(getQuickJSWasmPath()),
+        workerUrl: getCodemodeWorkerSpecifier(),
+    });
+    let result;
+    try {
+        const store = ctx ? readCodemodeStore(ctx.sessionManager.getBranch()) : {};
+        result = await sandbox.execute(code, { signal, store });
+    }
+    finally {
+        await sandbox.close();
+    }
+    // Calls still marked running were cut off by the script ending, a timeout, or an abort.
+    for (const call of calls) {
+        if (call.status === "running")
+            call.status = "cancelled";
+    }
+    const scriptOutput = [...result.output];
+    if (result.ok) {
+        const { set, delete: deleted } = result.storeWrites;
+        if (Object.keys(set).length > 0 || deleted.length > 0) {
+            options.appendEntry?.(CODEMODE_STORE_ENTRY_TYPE, { set, delete: deleted });
+        }
+        // pi extension: a returned value is appended like text().
+        if (result.value !== undefined)
+            scriptOutput.push({ type: "text", text: valueText(result.value) });
+    }
+    const items = formatOutput(scriptOutput);
+    if (!result.ok)
+        items.push({ type: "text", text: `Script error:\n${formatError(result, calls)}` });
+    if (generatedImages > 0 && !items.some((item) => item.type === "image")) {
+        items.push({
+            type: "text",
+            text: `Note: models.generateImages() returned ${generatedImages} image${generatedImages === 1 ? "" : "s"} that the script did not show. Show each image block of result.output with image(block).`,
+        });
+    }
+    const truncated = await truncateOutput(joinAdjacentText(items), sourceOptions.maxOutputTokens ?? DEFAULT_MAX_OUTPUT_TOKENS);
+    // After truncation, which joins the text items and moves images after them, so each path stays
+    // next to its image and is never cut.
+    const output = joinAdjacentText(await saveImages(truncated.items));
+    const wallTime = ((performance.now() - startedAt) / 1000).toFixed(1);
+    const header = `${result.ok ? "Script completed" : "Script failed"}\nWall time ${wallTime} seconds\nOutput:\n`;
+    const details = snapshot();
+    if (truncated.fullOutputPath)
+        details.fullOutputPath = truncated.fullOutputPath;
+    return {
+        content: [{ type: "text", text: header }, ...output],
+        details,
+        ...(modelUsage ? { usage: modelUsage } : {}),
+        ...(result.ok ? {} : { isError: true }),
+    };
+}
+/**
+ * Whether `query` names the namespace: its name, its script identifier (`mcp__dev-radius` is
+ * `mcp__dev_radius`), or the part after its last `__` in either form (`dev-radius`, `dev_radius`).
+ */
+function isNamespaceName(namespace, query) {
+    const id = toCodemodeIdentifier(namespace);
+    const queryId = toCodemodeIdentifier(query);
+    const suffix = (name) => (name.includes("__") ? name.slice(name.lastIndexOf("__") + 2) : undefined);
+    return namespace === query || id === queryId || suffix(namespace) === query || suffix(id) === queryId;
+}
+/**
+ * `searchTools()`, `describeTool()`, and `describeNamespace()`: ranked search and lookup over the
+ * script's nested tools and their namespaces.
+ */
+function createDiscoveryGlobals(tools, samples, options, ctx) {
+    // A script may loop over searchTools(). Past this many searches the session is withheld from the
+    // ranker, so a model-backed ranker falls back to its local ranking instead of billing per iteration.
+    let sessionSearches = 0;
+    const entry = (name) => ({ name: toCodemodeIdentifier(name), description: samples.get(name) ?? "" });
+    return [
+        {
+            name: "searchTools",
+            spread: true,
+            execute: async (args, { signal }) => {
+                const [query, searchOptions] = args;
+                if (typeof query !== "string")
+                    throw new Error("searchTools() expects a query string");
+                const limit = searchOptions?.limit ?? DEFAULT_TOOL_SEARCH_LIMIT;
+                if (typeof limit !== "number" || !Number.isInteger(limit) || limit <= 0) {
+                    throw new Error("searchTools() limit must be a positive integer");
+                }
+                const namespace = searchOptions?.namespace;
+                if (namespace !== undefined && namespace !== null && typeof namespace !== "string") {
+                    throw new Error("searchTools() namespace must be a string");
+                }
+                const documents = tools.flatMap((tool) => {
+                    const toolNamespace = options.getToolNamespace?.(tool.name);
+                    if (namespace && (!toolNamespace || !isNamespaceName(toolNamespace.name, namespace)))
+                        return [];
+                    return [createToolSearchDocument(tool, toolNamespace)];
+                });
+                const session = ctx && sessionSearches < MAX_SESSION_SEARCHES_PER_SCRIPT ? ctx : undefined;
+                if (session)
+                    sessionSearches += 1;
+                const ranked = await getToolRanker().rank(query, documents, limit, { signal, ctx: session });
+                const known = new Set(documents.map((document) => document.name));
+                return ranked
+                    .filter((match) => known.has(match.name))
+                    .slice(0, limit)
+                    .map((match) => entry(match.name));
+            },
+        },
+        {
+            name: "describeTool",
+            spread: true,
+            execute: (args) => {
+                const [name] = args;
+                if (typeof name !== "string")
+                    throw new Error("describeTool() expects a tool name");
+                const tool = tools.find((candidate) => candidate.name === name || toCodemodeIdentifier(candidate.name) === name);
+                return tool ? samples.get(tool.name) : undefined;
+            },
+        },
+        {
+            name: "describeNamespace",
+            spread: true,
+            execute: (args) => {
+                const [name] = args;
+                if (typeof name !== "string")
+                    throw new Error("describeNamespace() expects a namespace name");
+                let namespace;
+                const names = [];
+                for (const tool of tools) {
+                    const toolNamespace = options.getToolNamespace?.(tool.name);
+                    if (!toolNamespace || !isNamespaceName(toolNamespace.name, name))
+                        continue;
+                    namespace ??= toolNamespace;
+                    names.push(toCodemodeIdentifier(tool.name));
+                }
+                if (!namespace)
+                    return undefined;
+                return {
+                    name: namespace.name,
+                    ...(namespace.description ? { description: namespace.description } : {}),
+                    ...(namespace.instructions ? { instructions: namespace.instructions } : {}),
+                    tools: names,
+                };
+            },
+        },
+    ];
+}
+/**
+ * `models.*` for scripts: the model registry methods documented in docs/codemode.md.
+ * Classifier and image calls appear as nested call rows so the renderer shows them, and their usage
+ * goes to `addUsage`. Rows show only the model, never prompts or image data.
+ */
+function createModelGlobals(models, toolCallId, calls, publish, addUsage, addGeneratedImages) {
+    const limit = createLimiter(MAX_CONCURRENT_MODEL_CALLS);
+    let callCount = 0;
+    /**
+     * Resolve the script's model by provider and id only, check the context, then run the call as a
+     * nested call row. A script-supplied baseUrl or headers must never receive the credentials.
+     */
+    const runModelCall = async (name, type, [model, context], checkContext, run) => {
+        const listHint = `List the ${type} models you can use with models.getAvailableOfType("${type}").`;
+        if (!isRecord(model) || typeof model.provider !== "string" || typeof model.id !== "string") {
+            // undefined arrives as null: spread arguments cross the sandbox as a JSON array.
+            const undefinedHint = model === undefined || model === null
+                ? " models.getModelOfType() returns undefined for an unknown provider or id."
+                : "";
+            throw new Error(`${name}() expects ${withArticle(type)} model as its first argument, got ${describeValue(model)}.${undefinedHint} ${listHint}`);
+        }
+        const { provider, id } = model;
+        const ref = `${provider}/${id}`;
+        const resolved = models.getModelOfType(type, provider, id);
+        if (!resolved) {
+            const actualType = [...MODEL_TYPES].find((other) => other !== type && models.getModelOfType(other, provider, id) !== undefined);
+            throw new Error(actualType
+                ? `"${ref}" is ${withArticle(actualType)} model, not ${withArticle(type)} model. ${listHint}`
+                : `Unknown ${type} model "${ref}". ${listHint}`);
+        }
+        const checked = checkContext(context);
+        const record = {
+            id: `${toolCallId}/${name}/${++callCount}`,
+            name,
+            args: `${resolved.provider}/${resolved.id}`,
+            status: "running",
+        };
+        calls.push(record);
+        publish();
+        const startedAt = performance.now();
+        const result = await limit(() => run(resolved, checked));
+        record.durationMs = performance.now() - startedAt;
+        record.status = result.stopReason === "stop" ? "ok" : result.stopReason === "aborted" ? "cancelled" : "error";
+        if (result.errorMessage)
+            record.error = truncateText(result.errorMessage, ERROR_PREVIEW_CHARS);
+        if (result.usage) {
+            record.cost = result.usage.cost.total;
+            addUsage(result.usage);
+        }
+        publish();
+        return result;
+    };
+    const implementations = {
+        "models.getModelsOfType": (args) => {
+            const [type, provider] = args;
+            return models.getModelsOfType(toModelType(type), toProvider(provider)).map(toModelInfo);
+        },
+        "models.getAvailableOfType": async (args, { signal }) => {
+            const [type, provider] = args;
+            const available = await models.getAvailableOfType(toModelType(type), toProvider(provider), { signal });
+            return available.map(toModelInfo);
+        },
+        "models.getModelOfType": (args) => {
+            const [type, provider, id] = args;
+            if (typeof provider !== "string" || typeof id !== "string") {
+                throw new Error(`models.getModelOfType(type, provider, id) expects three strings, got (${args.map(describeValue).join(", ")}). The provider and the id are separate arguments, for example models.getModelOfType("classifier", "typesafe", "jev-latest").`);
+            }
+            const model = models.getModelOfType(toModelType(type), provider, id);
+            return model === undefined ? undefined : toModelInfo(model);
+        },
+        "models.classify": (args, { signal }) => runModelCall("models.classify", "classifier", args, checkClassifierContext, (resolved, context) => models.classify(resolved, context, { signal })),
+        "models.generateImages": (args, { signal }) => runModelCall("models.generateImages", "image", args, checkImagesContext, async (resolved, context) => {
+            const result = await models.generateImages(resolved, context, { signal });
+            addGeneratedImages(result.output.filter((block) => block.type === "image").length);
+            return result;
+        }),
+    };
+    return Object.entries(implementations).map(([name, execute]) => ({ name, spread: true, execute }));
+}

package/dist/extensions/codemode/execute.lazy.d.ts

@@ -0,0 +1,2 @@
+/** Loads the codemode sandbox runtime on the first script (see execute.ts). */
+export declare const loadCodemodeExecutor: () => Promise<typeof import("./execute.ts")>;

package/dist/extensions/codemode/execute.lazy.js

@@ -0,0 +1,2 @@
+/** Loads the codemode sandbox runtime on the first script (see execute.ts). */
+export const loadCodemodeExecutor = () => import("./execute.js");

package/dist/extensions/codemode/index.d.ts

@@ -0,0 +1,20 @@
+/**
+ * The `codemode` tool as an extension. The CLI loads it as a built-in extension; SDK users
+ * add `createCodemodeExtension()` to their extension factories.
+ *
+ * `codemode` is registered inactive. Activate it with `--tools`, the `defaultTools` setting, or
+ * `setActiveTools()`; the MCP extension activates it when MCP tools are only reachable from scripts.
+ */
+import type { ExtensionFactory } from "../../core/extensions/types.ts";
+import type { CodemodeMode } from "../../core/settings-manager.ts";
+export interface CodemodeExtensionOptions {
+    /** Overrides the `codemode.mode` setting. */
+    mode?: CodemodeMode;
+    /** Overrides the `codemode.inlineBudget` setting. */
+    inlineBudget?: number;
+    /** Expose the model catalog and classifiers to scripts as `models`. Default: `true`. */
+    models?: boolean;
+}
+export declare function createCodemodeExtension(options?: CodemodeExtensionOptions): ExtensionFactory;
+declare const _default: ExtensionFactory;
+export default _default;