The Agent Skills Directory

Prompt Injection (SAFE): No override or bypass instructions were found. Strong instructional language is limited to formatting rules such as the prohibition of diamond shapes due to JSON compatibility issues.\n- Data Exposure & Exfiltration (SAFE): File access is restricted to standard project configuration files (e.g., package.json, Dockerfile) for architectural mapping. No credential harvesting patterns or external network exfiltration logic are present.\n- Indirect Prompt Injection (SAFE): Although the tool ingests untrusted codebase data, the risk is inherent to codebase analysis tools and no specific exploitation patterns were found. Evidence: 1. Ingestion points: Codebase analysis tools (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: Standard agent file read/write tools. 4. Sanitization: Absent.\n- Persistence and Privilege (SAFE): No mechanisms for persistence, service installation, or privilege escalation were detected.\n- No Code (SAFE): The skill contains no executable scripts, binaries, or external dependencies, consisting entirely of documentation, pseudo-code validation logic, and configuration.

excalidraw