The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes sudo apt-get install libgtk-3-dev for Linux environment setup, which requires administrative privileges and can be used to modify system-level configurations.
[REMOTE_CODE_EXECUTION]: The skill implements a 'Test Generation' phase where it dynamically creates Dart test files (integration_test/*.dart) based on its analysis of the project's source code and then executes them using flutter test. This script generation and execution pattern can be exploited if the analyzed source code contains malicious patterns designed to influence the generated test logic.
[EXTERNAL_DOWNLOADS]: The skill performs several external operations including flutter pub get, npx prisma migrate deploy, and apt-get install, which fetch and install third-party dependencies and system packages during the setup phase.
[PERSISTENCE]: In the 'Self-Evolution Telemetry' section, the skill attempts to write execution metadata to ~/.claude/projects/skill-telemetry.md. Writing to hidden directories within the user's home folder outside of the active project directory can be used as a persistence or tracking mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes untrusted project data (e.g., pubspec.yaml, route configurations, and screen widgets) to build interaction maps and user flows. This content is then used to generate executable tests, creating a surface for indirect prompt injection where malicious project data could alter the agent's testing behavior.

walkthrough