The Agent Skills Directory

Prompt Injection (SAFE): No direct malicious instructions or safety bypass attempts were detected within the skill instructions.
Indirect Prompt Injection (LOW): The skill processes untrusted data by reading files from the local codebase, which could contain instructions designed to manipulate the agent's output.
Ingestion points: Local file system via file search and read operations defined in SKILL.md.
Boundary markers: Not explicitly defined in the instructions for the agent when processing file content.
Capability inventory: Execution of the gh issue create command.
Sanitization: No explicit sanitization of file content is performed, though the workflow requires explicit user approval before the final command execution, providing a strong mitigation.
Command Execution (SAFE): The skill utilizes the gh CLI to create issues. This is consistent with the skill's stated purpose, and the requirement for user approval before execution prevents unauthorized command runs.
Data Exposure & Exfiltration (SAFE): File access is limited to the purpose of researching the codebase for issue drafting. No unauthorized network requests or data exfiltration patterns were identified.

draft-issue