Sandbox Environment logo

Sandbox Environment

Community
pssah4
sandbox-environment

Sandbox API reference, proven patterns, and library recommendations for evaluate_expression

Overview

Publisherpssah4
Repositoryvault-operator
Skill namesandbox-environment
Stars
262
Forks
30
Bundled files
Instructions only
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by pssah4 on GitHub. Read the source before you install it.

Installation

Install the Sandbox Environment AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/pssah4/vault-operator/tree/main/bundled-skills/sandbox-environment
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/pssah4/vault-operator.git /tmp/vault-operator
mkdir -p .claude/skills
cp -r /tmp/vault-operator/bundled-skills/sandbox-environment .claude/skills/sandbox-environment
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Sandbox Environment in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Sandbox Environment on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Sandbox Environment is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Sandbox Environment Reference

Available APIs

ctx.vault

  • ctx.vault.read(path: string): Promise<string> -- Read text file
  • ctx.vault.readBinary(path: string): Promise<ArrayBuffer> -- Read binary file
  • ctx.vault.write(path: string, content: string): Promise<void> -- Write text (max 10MB, no writes to .obsidian/)
  • ctx.vault.writeBinary(path: string, content: ArrayBuffer): Promise<void> -- Write binary (max 10MB)
  • ctx.vault.list(path: string): Promise<string[]> -- List folder contents

ctx.requestUrl

  • ctx.requestUrl(url: string, options?: {method?: string, body?: string}): Promise<{status: number, text: string}>
  • HTTPS only. Allowed domains: esm.sh, cdn.jsdelivr.net, unpkg.com, registry.npmjs.org
  • Rate limit: 5 requests/minute

Standard Globals

Promise, JSON, Math, Date, Object (full), Array, Map, Set, RegExp, Number, String, Boolean, Symbol, setTimeout, clearTimeout, TextEncoder, TextDecoder, parseInt, parseFloat, isNaN, isFinite, encodeURIComponent, decodeURIComponent, Error, TypeError, RangeError

TypedArrays (for binary data)

Uint8Array, Int8Array, Uint16Array, Int16Array, Uint32Array, Int32Array, Float32Array, Float64Array, ArrayBuffer, DataView

NOT Available (will cause errors)

Buffer (Node.js), require(), dynamic import(), process, fs, path, __dirname, __filename, child_process, WebAssembly, eval(), new Function()

require, import(), process, globalThis, eval, new Function and WebAssembly are not merely absent: the AstValidator rejects the script BEFORE it compiles if the source so much as contains them, and its comment stripper deliberately keeps string literals. A script that searches for these tokens must therefore assemble them at runtime ('pro' + 'cess'), or it rejects itself. new RegExp is allowed.

The one rule that has no error message

A skill script is a single, self-contained file. No import, no require.

A static import x from './y.js' passes every check: the AstValidator only blocks require( and import(, and the repo has no other gate. But esbuild.transform (no bundling) rewrites it to require(), and there is no require here. The script dies on the user's first call, with an error that names none of this. Put everything in one file.

What the environment actually is

A Chromium iframe (sandbox="allow-scripts", no allow-same-origin, CSP default-src 'none'), NOT a Node vm. The vm worker was removed after a confirmed RCE; sandbox-worker.js in the repo root is a dead build artefact. Browser globals therefore exist in the usual browser sense, but the CSP gives them nothing to talk to: there is no network for fetch, and document belongs to a blank, origin-less frame. Do not build on them. ctx.vault and ctx.requestUrl are the entire surface you are given, and they are enough.

The return value is the only transport

The agent sees exactly one thing: the value your execute returns, JSON-stringified. A thrown error becomes Script execution error: {message}, so put the recovery step in the message. Everything the agent must know goes in the return value.

console.log is not a no-op, despite what this file used to claim. It works, and the output is visible in the Electron DevTools console. It simply is not a transport: nothing can read a child realm's console from the parent, and across an opaque origin the parent cannot even patch it. Use it to debug with DevTools open; never to return a result.

There is no network, and this one is absolute

fetch, XMLHttpRequest, WebSocket, EventSource and navigator.sendBeacon all EXIST as globals. Every call fails. The CSP is default-src 'none' with no connect-src, connect-src falls back to default-src by spec, and 'none' matches no URL at all: not a remote one, not a relative one, not a same-origin one. fetch() rejects with TypeError: Failed to fetch before a packet leaves the process. There is no URL, no retry and no fallback that changes this.

Dynamic import('https://esm.sh/...') is dead for the same reason: a script fetch falls to script-src, which grants 'unsafe-inline' 'unsafe-eval' and no URL sources. 'unsafe-eval' permits new Function; it does not permit loading code from a URL.

ctx.requestUrl is the only way out, and it works because the PARENT makes the request, not you.

Storage throws, it does not fail softly

localStorage, sessionStorage, document.cookie and indexedDB are blocked by the opaque origin. Reading the property THROWS a SecurityError; even typeof localStorage throws. An unguarded reference kills the script at the point of access, which is harsher than the network case. Persist through ctx.vault.

One fragility that is not ours to fix

A srcdoc iframe inherits the embedding document's CSP, and multiple policies are enforced as a conjunction: the meta CSP is a floor, never a ceiling. So new Function works only for as long as Obsidian's own renderer policy tolerates unsafe-eval. If Obsidian ever tightens it, this sandbox stops working and no change to the plugin can repair it.

Proven Patterns

Binary File Generation (PPTX, XLSX, DOCX, PDF)

IMPORTANT: Use the dedicated built-in tools instead of the sandbox for Office documents:

  • PPTX: Use create_pptx tool (not evaluate_expression)
  • XLSX: Use create_xlsx tool (not evaluate_expression)
  • DOCX: Use create_docx tool (not evaluate_expression)
  • PDF: Use workspace:export-pdf (Tier 1) or pandoc-pdf recipe (Tier 2)

These built-in tools run in the plugin context with full Node.js access and produce professional output. The sandbox should NOT be used for binary file generation -- it lacks Buffer/Blob support.

If you need simple PDF generation from scratch (not conversion), pdf-lib remains available in the sandbox:

typescript
import { PDFDocument, StandardFonts } from 'pdf-lib';
const doc = await PDFDocument.create();
const page = doc.addPage([595, 842]); // A4
const font = await doc.embedFont(StandardFonts.Helvetica);
page.drawText('Hello World', { x: 50, y: 750, size: 24, font });
const buf = await doc.save();
await ctx.vault.writeBinary('output.pdf', buf);
return 'Created output.pdf';

Data Transformation

typescript
const content = await ctx.vault.read('data.csv');
const rows = content.split('\n').map(r => r.split(','));
const markdown = '| ' + rows[0].join(' | ') + ' |\n| ' + rows[0].map(() => '---').join(' | ') + ' |\n' + rows.slice(1).map(r => '| ' + r.join(' | ') + ' |').join('\n');
await ctx.vault.write('data.md', markdown);
return 'Converted CSV to Markdown table';

HTTP Request (CDN only)

typescript
const resp = await ctx.requestUrl('https://esm.sh/lodash@4.17.21/package.json');
const pkg = JSON.parse(resp.text);
return pkg.version;

Anti-Patterns (DO NOT USE)

  • new Blob([data]) -- Blob not available. Use new Uint8Array(data) or ArrayBuffer
  • Buffer.from(str) -- Buffer not available. Use new TextEncoder().encode(str)
  • require('fs') -- require not available. Use ctx.vault
  • fetch(url) -- fetch not available. Use ctx.requestUrl
  • import('module') -- dynamic import not available. Use static import + dependencies param
  • outputType: 'blob' -- Blob not available. Use outputType: 'arraybuffer'
  • document.createElement() -- DOM not available
  • window.crypto.getRandomValues() -- crypto not available. Use Math.random()

Library Recommendations

TaskRecommendedAvoidReason
Excelexceljssheetjs/xlsxExcelJS writeBuffer() returns ArrayBuffer natively
PDFpdf-libjspdfpdf-lib pure JS, no DOM dependency
PPTXpptxgenjs (with caution)officegenpptxgenjs supports arraybuffer output
Imagessharp (if available)canvascanvas requires DOM
JSON/CSVbuilt-inpapaparsepapaparse works but unnecessary overhead
Datesbuilt-in Datemomentmoment too heavy for sandbox

Resource Limits

  • Heap: 128 MB, sampled every 500 ms, then killed
  • Execution timeout: 30 seconds
  • Write size: max 10 MB per operation
  • Write rate: max 10 writes/minute
  • Request rate: max 5 HTTP requests/minute
  • Writes to .obsidian/ blocked (security)

Frequently asked questions

What does the Sandbox Environment AI skill do?

Sandbox API reference, proven patterns, and library recommendations for evaluate_expression

Why use Sandbox Environment on TypingMind?

Because you install it once and use it with any model. Sandbox Environment is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Sandbox Environment in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/pssah4/vault-operator/tree/main/bundled-skills/sandbox-environment. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Sandbox Environment?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Sandbox Environment?

As many as you like. As long as a model supports skills, you can use Sandbox Environment with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Sandbox Environment AI skill free?

Yes. It is published on GitHub by pssah4 under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇