The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded credentials for production infrastructure found in SKILL.md.
Evidence: The lftp command in the 'Upload Single File' section includes plaintext credentials for alfonso@csrdevelopment.com with the password @#s;v1#%1M$+.
Evidence: Local development credentials admin / admin123 are exposed in the 'Available Sites' section for local2.hustletogether.com.
[COMMAND_EXECUTION]: The Python scripts use subprocess.run(shell=True) with unsanitized user inputs, creating a high risk of command injection.
Evidence: In scripts/wp-page.py, the create_page function takes arguments like --title, --slug, and --content and interpolates them directly into a docker exec string. While it attempts basic character escaping for quotes, it does not prevent shell metacharacters (e.g., $(...), ` , |, ;) from being executed by the underlying shell.
Evidence: In scripts/wp-seo.py, the set_seo_field function similarly interpolates the value argument into a shell string after only escaping single and double quotes, allowing for arbitrary command execution on the host or within the container.
[DATA_EXFILTRATION]: The skill instructs the agent to store sensitive API keys in predictable file paths on the local system.
Evidence: Instructions in SKILL.md specify storing the Pexels API key in /root/.pexels-api-key and the Unsplash API key in /root/.unsplash-api-key. These paths are outside the skill's own directory and are easily targetable for exfiltration by other malicious scripts or processes.

wordpress-admin