The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a local config.json file, creating a potential surface for indirect injection.
Ingestion points: Reads configuration data from config.json using fs.readFileSync in Step 2.
Boundary markers: The skill employs JSON.parse and a dedicated validateSchema function to ensure input adheres to a predefined schema before processing.
Capability inventory: The skill has the ability to write to the local file system (fs.writeFileSync for saving configuration and fs.appendFileSync for updating .gitignore).
Sanitization: Input is validated against a formal JSON schema (config.schema.json) to prevent malformed or unexpected data from being processed.
[DATA_EXPOSURE]: The skill includes logic to prevent the exposure of sensitive data. In Step 8, it explicitly excludes potentially sensitive fields (like GitHub tokens) when saving the configuration to disk and automatically adds the configuration file to .gitignore to prevent accidental commits of local settings.

configuration