The Agent Skills Directory

Indirect Prompt Injection (MEDIUM): The skill reads local source code and passes it directly into AI prompts. It lacks boundary markers or sanitization, which could allow malicious code comments to influence the agent's logic.
Ingestion points: Local source code files via cat and search results via rg.
Boundary markers: None identified.
Capability inventory: Shell command execution (ripgrep, git, cat) and network access via the Gemini CLI.
Sanitization: None provided.
Data Exfiltration (LOW): The skill transmits local source code to an external AI service. While this is the intended functionality, it constitutes an exposure risk for sensitive intellectual property.
External Downloads (LOW): Recommends installing ripgrep via Homebrew and google-generativeai via pip. These are considered trusted sources per [TRUST-SCOPE-RULE].
Unverifiable Dependencies (MEDIUM): The skill uses a gemini command-line tool that is not a standard component of the recommended google-generativeai Python library, indicating a dependency on an unverified or custom execution environment.

investigate