The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive shell operations including the dynamic generation of scripts (e.g., /tmp/audit_routes.sh, /tmp/audit_files.sh) in the temporary directory, modifying file permissions with 'chmod +x', and executing these scripts to perform path enumeration and vulnerability tests.\n- [REMOTE_CODE_EXECUTION]: Automated security scans identified a high-risk pattern where data from an external source (crt.sh) is piped into a Python interpreter. Although the provided code in the instructions uses the '-c' flag for hardcoded processing logic, the pattern of interpreting network responses via code execution is a significant security concern.\n- [EXTERNAL_DOWNLOADS]: The skill downloads numerous JavaScript chunks and remote content from the audit target and third-party services like crt.sh to perform its analysis and reconnaissance phases.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its ingestion of untrusted remote content.\n
Ingestion points: The skill fetches the full content of the target website into the 'PAGE' variable during reconnaissance in SKILL.md.\n
Boundary markers: No delimiters or 'ignore embedded instructions' warnings are present when processing the fetched HTML content.\n
Capability inventory: Includes extensive subprocess calls (curl, bash, python3), file-write capabilities, and permission modification commands.\n
Sanitization: No escaping or validation is performed on the remote 'PAGE' content before it is passed to shell utilities for grep-based detection.

web-security-audit