The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/rsa_refresh_generator.py uses sys.path.insert(0, COMPLIANCE_PATH) to dynamically load the compliance.ad_copy_validator module from a directory specified by the COMPLIANCE_PATH environment variable. Similarly, it loads review and competitor analysis modules from SERP_API_PATH. This dynamic loading from computed paths is a potential security risk if the environment variables point to untrusted or insecure directories.\n- [PROMPT_INJECTION]: The skill processes untrusted data scraped from external business websites (stored in the website_text field of the context JSON) to generate headlines and descriptions for Google Ads. This creates a surface for indirect prompt injection attacks, where a malicious website could include hidden instructions to manipulate the ad copy generation process. The analysis identified the ingestion point in rsa_refresh_generator.py, the lack of explicit sanitization for prompt injection in the data flow, and the agent's broad capability to generate ad content.\n- [EXTERNAL_DOWNLOADS]: The skill's setup instructions in README.md provide curl commands to download Python scripts and markdown reference files from the author's GitHub repository (fourteenwm/ppc-ai-skills). These are legitimate vendor resources used for the skill's operation.

rsa-refresh