The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bash script scripts/recon-scan.sh to execute network tools like nmap, whatweb, and curl against user-defined targets. It also provides instructions for manually running these tools in the terminal during exploitation phases.
[INDIRECT_PROMPT_INJECTION]: The skill presents a significant attack surface for indirect prompt injection. Untrusted data is ingested when scanning external web targets (via headers, robots.txt, and HTML crawling in Phase 2). The skill lacks explicit boundary markers or content sanitization to prevent these external inputs from influencing the agent's actions, which are backed by powerful capabilities including terminal and file system access.
[DATA_EXFILTRATION]: The skill processes sensitive information, including captured credentials and vulnerability evidence. It acknowledges specific platform-level data exposure risks ("Aux-client leakage") and provides procedures for redacting secrets in chat and storing them in local files.
[DYNAMIC_EXECUTION]: The scripts/recon-scan.sh utility uses embedded Python snippets to perform runtime hostname resolution and CIDR scope validation.

web-pentest