The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external websites to evaluate 'answer-readiness' and 'extractable knowledge'. This creates a surface where an attacker-controlled website could include instructions aimed at manipulating the agent's analysis or subsequent implementation steps.
Ingestion points: The scripts/audit_url.py script fetches and returns the full HTML body, robots.txt, and sitemap content from user-provided URLs to the agent.
Boundary markers: The skill includes instructions to 'Avoid... instructions aimed at manipulating an AI model', but it does not specify explicit delimiters or 'ignore' instructions for the raw HTML data processed during the audit.
Capability inventory: The agent is authorized to write new content to the repository, modify routing/metadata, and execute the audit script.
Sanitization: While the script uses a basic HTML parser to extract specific tags, the full body content is provided to the agent for evaluation of 'answer-readiness' and 'intent coverage', which is where malicious instructions would reside.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a local Python script (scripts/audit_url.py) to perform network-based audits of URLs. This is the intended primary functionality of the tool.

geo-optimizer