The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill reads and processes untrusted source code files, which serves as an ingestion point for potentially malicious instructions. While no explicit sanitization is performed on the file content, the risk is reduced by the use of strict, predefined regex patterns for comment identification and a mandatory human review step before any file-write operations occur. * Evidence Chain: 1. Ingestion points: Source code files read via rg and sed (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: File-write access via sed -i and file searching via rg (SKILL.md). 4. Sanitization: Uses fixed regex templates and requires user confirmation before execution.- Command Execution (SAFE): The skill utilizes standard utilities (sed, rg, git) to perform its tasks. These commands are executed locally for the primary purpose of the skill and do not involve downloading or executing remote scripts.

remove-comments