The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/queue.py executes a sibling skill's shell script (start.sh) using subprocess.run. The command is invoked with a list of arguments rather than a shell string, which is a security best practice that prevents shell injection attacks.
[COMMAND_EXECUTION]: The skill interacts with the local operating system's process management via os.kill to monitor and terminate specific upload processes based on PIDs stored in a tracking file. This functionality is restricted to the intended purpose of managing the upload queue.
[DATA_EXFILTRATION]: The skill manages state using JSON files located in the /tmp/ directory. While it reads from and writes to these files, no evidence was found of sensitive data being transmitted to external or unauthorized domains. The skill checks for environment variables like HF_TOKEN and R2_SECRET_ACCESS_KEY but does not exfiltrate them.
[PROMPT_INJECTION]: The skill mitigates potential injection attacks by validating all model identifiers against a restrictive regular expression (^[A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+$) before any processing or execution occurs. This ensures that only valid HuggingFace repository names are accepted and prevents malicious characters from being passed to the shell.

acervo-ship-queue