The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external source code files which could contain malicious instructions designed to influence the agent when it reviews the scan results. * Ingestion points: The script scripts/scan_cmdi.py reads content from a user-provided directory using pathlib.Path.read_text. * Boundary markers: Findings are encapsulated in structured Finding objects before being presented to the agent. * Capability inventory: The skill utilizes Bash(python3:*) to execute the scanner script, and the Read, Glob, and Grep tools for file exploration. * Sanitization: Code snippets matching command-injection patterns are extracted and reported as findings without specific sanitization against prompt injection strings that might be present in the source files.

detecting-command-injection-patterns