The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides instructions and specific file paths for uploading agent session traces to Hugging Face, including directories such as ~/.claude/projects, ~/.codex/sessions, and ~/.pi/agent/sessions. These locations contain sensitive data, including private prompts and potentially hardcoded secrets. The skill includes explicit warnings regarding the privacy risks and the need for redaction before transmission.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx -y @huggingface/hub to download and execute the Hugging Face Hub CLI for dataset management. These resources are provided by a well-known technology service.
[COMMAND_EXECUTION]: The skill instructions involve executing shell commands via curl for API interactions and npx for repository operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external Hugging Face datasets which are user-contributed and may contain malicious instructions designed to influence agent behavior.
Ingestion points: External data is ingested through the Hugging Face Dataset Viewer API via endpoints like /rows, /search, and /filter (as documented in SKILL.md).
Boundary markers: No explicit boundary markers or delimiters for untrusted content are defined in the instructional text.
Capability inventory: The skill has access to command execution (curl, npx) and file upload capabilities.
Sanitization: The instructions do not specify sanitization or validation protocols for the data retrieved from the Dataset Viewer API.
[DYNAMIC_EXECUTION]: The skill uses npx to dynamically load and execute the @huggingface/hub package at runtime for data uploading tasks.

hugging-face-dataset-viewer