The Agent Skills Directory

Structured Process for Prompt Refactoring: The skill provides a comprehensive methodology for auditing and merging agent instructions. While this involves processing untrusted or third-party prompts (an indirect prompt injection surface), the skill includes mandatory evaluation steps and human-in-the-loop check-ins to ensure system integrity. \n- Path Traversal Mitigations: The provided skill_loader.py template includes specific checks to ensure that dynamically loaded skill files reside within the intended directory, preventing potential directory traversal attempts. \n- Secure Tooling Integration: The instructions reference official platform tools and plugins for prompt auditing and cost optimization, ensuring that the extension of capabilities relies on trusted services. \n- Data Sanitization in Reporting: The included scripts (e.g., coupling_map.py) contain utility functions to sanitize data before it is included in generated reports, reducing the risk of formatting-based injection in analysis outputs.

single-agent-loop