The Agent Skills Directory

[COMMAND_EXECUTION]: Documentation in SKILL.md advises using sudo apt-get install ffmpeg to satisfy prerequisites for video ingestion. If an agent attempts to execute this command autonomously, it would involve a privilege escalation attempt.
[EXTERNAL_DOWNLOADS]: The skill requires installing multiple third-party packages from PyPI and NPM (e.g., deeplake, pymupdf, sharp). It also facilitates downloading data from HuggingFace via the _huggingface parameter in the ingestion method.
[PROMPT_INJECTION]: The skill ingests untrusted external data, including local files (PDF, Video, Text) and HuggingFace datasets, creating a surface for indirect prompt injection where instructions embedded in the data could influence agent behavior.
Ingestion points: Ingestion of local file system paths and HuggingFace datasets through the client.ingest method as documented in SKILL.md and examples.md.
Boundary markers: No specific delimiters or boundary markers are described to isolate data from potential instructions during ingestion.
Capability inventory: The skill allows the agent to read/write files and execute shell commands via the Bash and Read tools specified in the frontmatter.
Sanitization: No evidence of content-level sanitization or validation is present in the provided documentation or format classes.

deeplake-managed