The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local Node.js script (extract-example.mjs) to perform its primary function. This script accepts command-line arguments derived from user input or environment context (source path and line number).
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted source code provided via the <source-path> argument. If a processed file contains malicious code within a JSDoc **Example** block, that code is extracted and written to a new file in the ./scratchpad directory. If the agent or user is later prompted to execute this extracted code, it could lead to execution of the malicious content.
Ingestion points: The extract-example.mjs script reads external source files provided in the source-path argument.
Boundary markers: None identified; the script extracts content based on simple string and regex matches for JSDoc syntax.
Capability inventory: The script has the capability to read files (readFileSync), write files (writeFileSync), and create directories (mkdirSync).
Sanitization: The script validates the --runner identifier using a strict regex (/^[A-Za-z_$][\w$]*$/) and ensures the line number is a safe integer. However, there is no validation to ensure the source-path or out-dir remain within the project's scope.
[DATA_EXPOSURE]: The script uses path.resolve() on the user-provided sourcePath without checking if the resolved path is within the workspace. This allows the script to read any file on the system that the current process has permissions for, provided the file contains the expected JSDoc markers.

scratchpad