The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines functions in references/tools.md that facilitate the execution of external CLI tools, including curl, nmap, nghttp, and specialized smuggling scanners like smuggler.py and h2csmuggler. The workflows in references/workflows.md detail the use of these tools for network reconnaissance and vulnerability validation.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it is designed to ingest and analyze raw HTTP traffic from external, untrusted servers.
Ingestion points: Raw HTTP responses and headers are ingested through python_socket_request, curl_h2_request, and automated scanning tools (defined in references/tools.md and utilized in references/workflows.md).
Boundary markers: The instructions do not specify the use of delimiters or defensive prompts to mitigate the risk of the agent interpreting malicious instructions embedded in server responses.
Capability inventory: The skill is granted extensive network capabilities, including low-level socket operations, custom HTTP/2 framing, and CDN cache management (file: references/tools.md).
Sanitization: No explicit sanitization or filtering of external server responses is detailed in the workflows or tool definitions across references/tools.md and references/workflows.md.

pentest-http-smuggling