The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML content from external sources via the seo_checker.py tool, creating a vulnerability to indirect prompt injection.
Ingestion points: scripts/seo_checker.py accepts URLs and file paths for analysis via command-line arguments.
Boundary markers: The skill does not define specific delimiters or instructions for the agent to treat the tool output as untrusted content.
Capability inventory: The script seo_checker.py uses urllib.request to perform network operations and can read local files.
Sanitization: No sanitization is performed on the extracted HTML data (titles, headings, meta descriptions) before it is presented to the agent.
[DATA_EXFILTRATION]: The tool seo_checker.py performs network requests to user-supplied URLs using urllib.request.urlopen. This functionality, while necessary for auditing, allows the agent to access arbitrary external or internal network resources.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute Python scripts (seo_checker.py and seo_health_scorer.py) via the shell to perform its core functions.

seo-audit