The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building applications that ingest external data (e.g., HTTP requests, file reading), which creates a potential surface for indirect prompt injection attacks if the agent processes this data.
Ingestion points: The skill describes handling HTTP request bodies via req.on('data', ...) and reading files from the disk using fs.readFile and fs.readdir in SKILL.md and AGENTS.md.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are provided for the agent to use when handling content from these sources.
Capability inventory: The skill documentation includes examples for child_process (command execution), fs (file system writes), and net/http (network operations).
Sanitization: The skill includes a dedicated "Security" section in SKILL.md that advises on input validation, sanitization of URLs, and path traversal prevention.
[COMMAND_EXECUTION]: The skill provides instructions and examples for executing system commands using Node.js built-in modules.
Evidence: Detailed documentation and code examples for child_process.spawn, exec, execFile, and fork are included in both SKILL.md and AGENTS.md to demonstrate process management.
[DATA_EXFILTRATION]: The skill details how to perform network operations which, while standard for Node.js, could be used for data exfiltration.
Evidence: Documentation for fetch, http.createServer, net.connect, and http2.createSecureServer demonstrates how to send and receive data over the network.
[DYNAMIC_EXECUTION]: The skill contains instructions for generating and manipulating executable binaries.
Evidence: SKILL.md describes the creation of Single Executable Applications (SEA) using the postject tool to inject blobs into the Node.js binary.
[EXTERNAL_DOWNLOADS]: The skill references and downloads configuration from well-known and trusted sources.
Evidence: Links to official Node.js documentation at nodejs.org and the official GitHub repository at github.com/nodejs/node are provided in metadata.json and SKILL.md.

node