The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation references official GitHub repositories and installation instructions for well-known security tools including Metasploit, Sliver, MITRE Caldera, Pacu, Prowler, ScoutSuite, and BloodHound.
Evidence: Tables in section 5 list installation commands and GitHub URLs for version-pinned reference.
Mitigation: These references are provided for planning only; the agent environment is explicitly configured with forbidden-tools: [Write, Edit, Bash, WebFetch], preventing any automated installation or execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from upstream tools such as reconnaissance scan findings and purple team technique lists which could contain untrusted data.
Ingestion points: Section 6 identifies security-pentest-recon-scan findings and security-response-red-purple-team technique lists as primary context inputs.
Boundary markers: Includes an extensive 11-section 'Pre-check' (Section 3) requiring verification of a signed Rules of Engagement (ROE) document before processing specific tool context.
Capability inventory: The skill is restricted to the Read tool only. Write, Edit, Bash, and WebFetch are explicitly forbidden.
Sanitization: Section 7 specifies the use of an external appsec-sdk to redact sensitive data (credentials, hashes, tokens) before outputting results.

security-pentest-exploitation-planning