The Agent Skills Directory

[COMMAND_EXECUTION]: The script hf_stats.sh contains a severe command injection vulnerability. User-controlled input $REPO_ID is used inside double-quoted strings in the shell script (e.g., URL="...${REPO_ID}..."), allowing for shell command substitution at the point of variable assignment. Furthermore, $REPO_ID is interpolated into a Python string executed via python3 -c (e.g., repo_id = '$REPO_ID'), which allows for arbitrary Python code execution if the input contains a single quote.
[EXTERNAL_DOWNLOADS]: The skill fetches model and dataset metadata from Hugging Face's official API using curl.
[PROMPT_INJECTION]: The skill processes external content from the Hugging Face API, creating a surface for indirect prompt injection.
Ingestion points: JSON data is fetched from the Hugging Face API in hf_stats.sh and parsed for display.
Boundary markers: Absent. The raw formatted output of the script is presented directly to the agent.
Capability inventory: The skill has the ability to execute shell commands and Python code, which could be leveraged if the agent is manipulated by injected instructions.
Sanitization: Absent. The script prints values directly from the API response without escaping or validation.

hugging-face-stat