The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script in scripts/extract.mjs fetches JavaScript files (data/document.js and data.js) from arbitrary, user-provided remote URLs and executes them using vm.runInContext. Since the vm module is not intended to be a secure sandbox, this creates a vector for Remote Code Execution (RCE) if a user is tricking into providing a malicious Axure URL.
[DYNAMIC_EXECUTION]: The function evaluateAxureCode dynamically evaluates external JavaScript code at runtime to extract prototype data, using string-to-code execution patterns that bypass static analysis.
[EXTERNAL_DOWNLOADS]: The skill automatically installs the playwright package and chromium browser binaries on the first run using npm and npx. By default, it uses a well-known mirror (https://npmmirror.com/mirrors/playwright) for downloads.
[COMMAND_EXECUTION]: The skill uses child_process.execSync to run dependency installation commands. It also launches Chromium with the --disable-web-security and --allow-file-access-from-files flags, which weakens the browser's security boundaries and could be exploited by a malicious site.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from remote prototypes, creating an attack surface where embedded instructions in the prototype data could influence agent behavior.
Ingestion points: Prototype metadata, annotations, and sitemap data are fetched from remote URLs in scripts/extract.mjs.
Boundary markers: Absent; the data is parsed and evaluated directly.
Capability inventory: The script has access to shell execution (execSync), file system writes (fs.writeFileSync), and network fetching.
Sanitization: Absent; remote JavaScript content is executed directly in a VM context.

extract-axure-data