The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to automate codebase analysis. Key operations include using git to enumerate project files and diff commits, and invoking bundled Node.js and Python scripts to process project data. These operations are restricted to the project root and the skill's data directory.
[EXTERNAL_DOWNLOADS]: To ensure functionality, the skill may run pnpm install in the plugin's root directory. This is used to build the necessary @understand-anything/core module if its build artifacts are missing. This is a standard initialization procedure for this class of tool.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the project being analyzed, including README files and package manifests. It mitigates this attack surface by including explicit guardrail instructions in the sub-agent prompts to ignore any instructions, policy text, or prompt-like directives found within those files.
[DYNAMIC_EXECUTION]: Several scripts dynamically resolve and import the core logic from the plugin's packages using path-based resolution. This is done to ensure compatibility across different installation environments and follows a deterministic search path through local plugin directories.

understand