The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include executing various system auditing commands. Notably, it contains a bash-specific TCP connection pattern (exec 3<>/dev/tcp/127.0.0.1/22) in references/manual-checks.md for performing banner grabs. While intended for local auditing, this specific redirection syntax is commonly associated with network-based exploitation and reverse shells.\n- [DATA_EXFILTRATION]: The skill requires the agent to search for and audit sensitive file paths, including SSH private keys (~/.ssh/id_*), environment files (.env), and process environment variables (/proc/*/environ). Although the methodology mandates recording only file metadata and variable names rather than their values, the capability to read and process these high-value assets represents a significant data exposure risk.\n- [INDIRECT_PROMPT_INJECTION]: The skill performs auditing of local network services using curl. Malicious or compromised local services could return responses (e.g., headers or body content) specifically crafted to manipulate the agent's reasoning or influence subsequent audit steps.\n
Ingestion points: Responses from local HTTP services captured via curl, and metadata from discovery tools like find and apt.\n
Boundary markers: No specific delimiters or safety warnings are included for handling tool outputs within the instructions.\n
Capability inventory: Significant local capabilities including shell access, file system interaction, and local script execution.\n
Sanitization: No explicit sanitization or validation of data retrieved from local services is defined in the methodology.

host-pentest