The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/runner.py file executes shell commands defined in setup_commands using subprocess.run. These commands are dynamically generated by an LLM in scripts/scenario_generator.py based on the content of the file being analyzed. While shlex.split is used to prevent shell-based injection, the commands themselves are unvalidated and derived from external input.- [REMOTE_CODE_EXECUTION]: The skill facilitates remote code execution by automating the generation and execution of setup scripts. An attacker providing a malicious rule or skill file could influence the LLM to output dangerous command sequences (e.g., network exfiltration or system modification) that the skill then executes.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests untrusted files to generate 'compliance specs' and 'scenarios' without sufficient sanitization or instruction boundaries. A malicious file could include instructions that subvert the generation logic to create harmful test cases or setup steps.- [DATA_EXFILTRATION]: While not directly exfiltrating data, the combination of command execution and the ability to read arbitrary files for 'compliance testing' provides a high-surface area for data harvesting if the LLM generation process is compromised by a malicious input file.

skill-comply