The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted AWS WAF configuration data provided as JSON files, which represents an indirect prompt injection surface if the agent obeys instructions embedded within rule names or statements.
Ingestion points: The waf-assess.py script reads and normalizes user-provided JSON files containing WAF rule names, statements, and metadata.
Boundary markers: The instructions do not define specific delimiters for the untrusted data, although they encourage the agent to verify the machine's findings against the configuration.
Capability inventory: The skill uses file_read and file_write permissions to process artifacts and generate the final HTML report.
Sanitization: The waf-report.py script applies robust HTML escaping and sanitization to all content before rendering it into the final assessment report, mitigating cross-site scripting risks in the generated document.

ddos-guardian