The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents the installation of the @inferencesh/sdk package from the NPM registry, along with other standard libraries such as express, hono, zod, and @upstash/ratelimit for server-side proxying and validation.
[COMMAND_EXECUTION]: The skill has permissions to execute npm, npx, node, pnpm, and yarn commands, which are necessary for the development and deployment of JavaScript-based AI applications.
[REMOTE_CODE_EXECUTION]: Documentation within references/tool-builder.md and references/agent-patterns.md highlights features for runtime code execution. Specifically, it demonstrates enabling codeExecution for agents and includes an example using eval() for a calculator tool. While shown in a technical context, these patterns allow for dynamic execution of content generated by or provided to the AI.
[PROMPT_INJECTION]: The skill defines an attack surface for indirect prompt injection by creating agents that process untrusted data.
Ingestion points: Data enters the system via agent.sendMessage and client.uploadFile (e.g., references/files.md).
Boundary markers: The provided patterns do not explicitly demonstrate the use of delimiters or 'ignore' instructions for external data.
Capability inventory: Agents have access to codeExecution, webhookTool for external API calls, and various platform-hosted AI applications.
Sanitization: Code examples do not show explicit sanitization or validation of tool arguments before processing.

javascript-sdk