The Agent Skills Directory

[DYNAMIC_EXECUTION]: The hf_stats.sh script dynamically constructs a Python command by interpolating shell variables into a string passed to python3 -c.
Evidence: The script uses the pattern repo_id = '$REPO_ID' inside a block of Python code executed via the shell. Because $REPO_ID (derived from agent/user input) is not sanitized, a value containing a single quote could be used to break out of the string literal and execute arbitrary Python code (e.g., ' + __import__('os').system(...) + ').
[COMMAND_EXECUTION]: The skill relies on executing system binaries to fulfill its primary purpose.
Evidence: The script hf_stats.sh invokes curl for network requests and python3 for JSON processing.
[EXTERNAL_DOWNLOADS]: The skill retrieves data from a well-known external service.
Evidence: It uses curl to fetch metadata from official Hugging Face API endpoints such as https://huggingface.co/api/models/. This is documented as functional behavior for interacting with a recognized service.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a conduit for untrusted data from an external API, which could potentially contain malicious instructions targeting the agent.
Ingestion points: Data is fetched from the Hugging Face API and saved to a temporary file in hf_stats.sh.
Boundary markers: Absent. The script outputs formatted text directly to the agent's context without clear delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the ability to execute shell commands (curl, python3) and perform file operations.
Sanitization: While the script parses data as JSON, the resulting string values are printed to standard output without further sanitization before being processed by the AI agent.

hugging-face-stat