The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external web applications and infrastructure by processing untrusted data from target responses, such as HTTP headers, CSP policies, and cloud storage metadata. This creates an attack surface where an attacker-controlled target could provide malicious instructions within these fields to influence the agent.
Ingestion points: Ingests data from target response headers via curl_headers, securityheaders_check, and mozilla_observatory_scan. It also processes CSP strings in csp_evaluator and cloud storage metadata through s3scanner_scan (found in references/tools.md).
Boundary markers: No specific boundary markers or "ignore instructions" are defined to delimit external data from agent instructions.
Capability inventory: The skill utilizes network scanning, HTTP method testing, and cloud storage enumeration across several utility scripts.
Sanitization: There is no evidence of sanitization or strict schema validation for the content received from external targets.
[COMMAND_EXECUTION]: The skill uses various network diagnostic and security tools to perform its hardening verification tasks.
Evidence: Documentation and workflows in references/tools.md and references/workflows.md describe the use of nmap (ssl-enum-ciphers, http-methods), curl, testssl.sh, nuclei, s3scanner, subjack, and nikto to interact with external network services.

pentest-config-hardening