The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation provides example Python code for custom tools using subprocess.run(shell=True) with string interpolation for the target and payload variables. This pattern is susceptible to command injection if the inputs originate from untrusted sources.
[COMMAND_EXECUTION]: The skill is centered around an agent that autonomously executes shell commands and security tools (like nmap and sqlmap). This design grants the agent high privilege to perform network and system operations.
[INDIRECT_PROMPT_INJECTION]: The agent ingests data from external targets, such as web page content and command outputs, and feeds this data back into the LLM reasoning loop. 1. Ingestion points: Data is collected via HTTPTool and ShellTool components. 2. Boundary markers: The system prompt uses markdown headers, which provide limited protection against malicious content in target responses. 3. Capability inventory: The agent possesses extensive capabilities including network scanning, web exploitation, and arbitrary shell command execution. 4. Sanitization: No evidence of output sanitization or validation is present.
[EXTERNAL_DOWNLOADS]: The installation instructions require cloning a repository from an external source (github.com/oritera/Cairn.git) and installing associated dependencies.

cairn-ai-pentest