The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user project files (Dart, YAML, XML, Plist) to detect security vulnerabilities, creating a surface for indirect prompt injection.
Ingestion points: Scripts in the scripts/ directory read files from the target project root (e.g., lib/**/*.dart, pubspec.yaml, AndroidManifest.xml, Info.plist).
Boundary markers: The scripts generate structured JSON reports containing findings. While structured, the code and pattern fields contain raw snippets from the analyzed files.
Capability inventory: The skill allows the agent to execute Python scripts that perform file system reads and writes within the target project directory.
Sanitization: The extracted code snippets are not sanitized or escaped before being placed in the JSON output, which could allow malicious instructions embedded in the analyzed code to influence the agent's subsequent analysis or reporting.
[COMMAND_EXECUTION]: The check_dependencies.py script executes external commands to analyze project dependencies.
Evidence: The script uses subprocess.run to invoke flutter pub outdated --json (or fvm flutter ...).
Context: This is a legitimate operation for a dependency checker. The implementation follows security best practices by using a list of arguments and avoiding shell=True.

owasp-mobile-security-checker