The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The script introduces an indirect prompt injection vulnerability (Category 8). It ingests untrusted data (source code files) and prints findings to the console without boundary markers. An attacker could embed instructions in the target code to manipulate the agent's reasoning.
Ingestion points: All .ts files within the target directory scanned by rg.
Boundary markers: None; findings and code snippets are printed directly to stdout.
Capability inventory: Reading local file content and outputting to the agent's context.
Sanitization: Only function name extraction uses a strict \w+ regex; other checks print line content directly without escaping.
[COMMAND_EXECUTION] (LOW): The script relies on the external ripgrep (rg) utility to perform file system analysis. While the directory variable is properly quoted to mitigate simple command injection, the utility provides the agent with a mechanism for broad visibility into the local file system.

security-convex