The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/fetch_explosive_covers.py executes a PowerShell command via subprocess.run on Windows platforms.
Evidence: subprocess.run(["powershell", "-NoProfile", "-command", "[Environment]::GetEnvironmentVariable('REDFOX_API_KEY', 'User')"], ...)
Context: This is used specifically to retrieve the REDFOX_API_KEY from the user's environment variables if not already present in the current process environment.
[DATA_EXFILTRATION]: The script scripts/fetch_explosive_covers.py attempts to read local shell configuration files.
Evidence: The script iterates through .zshrc, .bashrc, .bash_profile, and .profile in the user's home directory.
Context: The script searches for the string export REDFOX_API_KEY= to automatically configure the required API key for the redfox.hk service. It does not exfiltrate the full content of these files.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external data.
Ingestion points: Data is fetched from the vendor's API at redfox.hk, including social media note titles and descriptions.
Boundary markers: None explicitly implemented; the agent processes the retrieved text and images directly.
Capability inventory: The skill uses a vision tool to analyze remote images and the agent's reasoning to generate design schemes and image prompts.
Sanitization: No specific sanitization or filtering of the retrieved API data is documented.

xiaohongshu-cover