The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation suggests the use of sudo to install the system utility ffmpeg to support video ingestion tasks.
[EXTERNAL_DOWNLOADS]: The skill relies on multiple third-party libraries including requests, aiohttp, pymupdf, Pillow, pycocotools, numpy, pandas, pdfjs-dist, and sharp. It also supports fetching datasets from HuggingFace.
[REMOTE_CODE_EXECUTION]: The Node.js implementation utilizes a WebAssembly (WASM) module compiled from C++ for data processing, involving the execution of compiled binary code.
[PROMPT_INJECTION]: The skill processes untrusted external data (such as PDF text, video segments, and image metadata) which creates a surface for indirect prompt injection. A malicious file could contain instructions that influence the agent's behavior when the processed data is later queried or summarized. \n
Ingestion points: client.ingest method in SKILL.md used for processing local files and external datasets. \n
Boundary markers: No specific boundary markers or instructions to ignore embedded commands are mentioned in the documentation. \n
Capability inventory: The skill is capable of file reading, making network requests to the DeepLake API, and executing SQL queries as described in SKILL.md and reference.md. \n
Sanitization: No explicit sanitization or validation processes for the content of ingested files are described.

deeplake-managed