The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill is designed to wrap the claude CLI via bash, which is its stated and legitimate purpose.
[INDIRECT_PROMPT_INJECTION] (LOW): As a natural language interface, it has an inherent surface for indirect prompt injection. 1. Ingestion points: The prompt passed to the CLI tool in claude -p. 2. Boundary markers: No explicit delimiters are shown in the examples. 3. Capability inventory: The tool has permission to execute the claude binary via bash. 4. Sanitization: No specific sanitization is performed within the skill documentation, relying on the underlying platform's tool-calling safety and the CLI's internal logic.
[SAFE] (SAFE): No malicious obfuscation, credential exfiltration, or persistence mechanisms were detected.

ask-claude