The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell commands for reconnaissance and payload manipulation.
Evidence: Use of cat, grep, base64, and xmllint to identify endpoints and transform SAML payloads.
[INDIRECT_PROMPT_INJECTION]: The skill instructions involve processing external, untrusted SAML assertions which represents an injection surface.
Ingestion points: The workflow involves decoding and parsing BASE64_SAML strings potentially sourced from a target application.
Boundary markers: The skill implements a 'Mandatory confirmation gate' requiring the agent to ask for target details, verify authorization, and show exact commands before execution.
Capability inventory: The skill utilizes shell execution for file reading, string manipulation, and XML formatting across its attack descriptions.
Sanitization: No explicit sanitization or validation of the input SAML strings is defined before the agent processes them via base64 or xmllint.

hunt-saml