The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install the @inferencesh/sdk package from the npm registry, which is the official library for the vendor's platform.
[DATA_EXFILTRATION]: The SDK documentation describes features for uploading local files, buffers, and data URIs to the inference.sh platform via client.uploadFile and client.run. This is a standard functional requirement for processing multi-modal data with AI models.
[COMMAND_EXECUTION]: The reference material provides patterns for enabling agent-side code execution and includes implementation examples using eval() within tool handlers to process mathematical expressions.
[PROMPT_INJECTION]: The skill documents an agent integration architecture that ingests untrusted data from user messages and file uploads, creating a surface for indirect prompt injection.
Ingestion points: Messages passed to agent.sendMessage and files provided to client.uploadFile or as agent attachments.
Boundary markers: Not explicitly required in agent instructions, though SDK examples demonstrate approval workflows.
Capability inventory: The environment includes restricted shell access (via allowed-tools), file system interactions, and network operations.
Sanitization: Documentation highlights the use of human-in-the-loop patterns via the requireApproval() method.
[NO_CODE]: The skill is comprised of markdown documentation and reference guides; it does not distribute standalone executable scripts, binaries, or automated shell commands.
[SAFE]: All identified external domains, repositories, and package dependencies originate from the vendor's official infrastructure (inferen-sh), with no signs of malicious patterns or deceptive content.

javascript-sdk