The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill instructs the agent to access local directories containing sensitive session traces from various agents (Claude Code, Codex, Pi Agent) and upload them to Hugging Face.
Evidence: Mentions paths such as ~/.claude/projects, ~/.codex/sessions, and ~/.pi/agent/sessions as sources for raw JSONL session traces.
Exfiltration Vector: Provides commands to upload these local directories to remote Hugging Face dataset repositories using the hf CLI and npx @huggingface/hub.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the Hugging Face Dataset Viewer API, which serves as an ingestion point for external data into the agent's context.
Ingestion points: Data is retrieved via /rows, /search, and /filter endpoints from arbitrary Hugging Face datasets (SKILL.md).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when handling the retrieved dataset content.
Capability inventory: The agent has capabilities to execute shell commands (curl, npx, hf CLI) and perform network operations.
Sanitization: No explicit sanitization or validation of the fetched dataset content is defined in the skill.
[EXTERNAL_DOWNLOADS]: The skill uses npx to download and execute the @huggingface/hub package at runtime.
Evidence: npx -y @huggingface/hub upload datasets/<namespace>/<repo> ./local/parquet-folder data in SKILL.md. This downloads and runs the official package from the well-known registry.

huggingface-datasets