The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js scripts (scripts/detect-token-drift.mjs and scripts/check-contrast.mjs) to perform technical audits on the target codebase. These scripts are read-only, non-networked, and used solely for deterministic design-system and accessibility verification.
[DATA_EXPOSURE]: As part of its gap-analysis workflow, the skill identifies and audits sensitive configuration patterns, such as hard-coded credentials or environment variable exposure in files like .env. This behavior is limited to documenting technical debt and does not involve data exfiltration.
[PROMPT_INJECTION]: Instructions include clear directives for documentation quality and workflow (e.g., "Check that root DESIGN.md, when present, was evaluated") but do not attempt to override AI safety filters or system-level constraints.
[DATA_EXPOSURE]: Indirect prompt injection attack surface evaluation:
Ingestion points: Ingests untrusted project files including TSX/JSX code, CSS files, and configuration/documentation files from the target repository.
Boundary markers: The skill lacks explicit instructions to use boundary delimiters when processing external code, though it instructs the agent to cite concrete evidence with file paths instead of large code blocks.
Capability inventory: The skill executes local Node.js utility scripts via node for color and token analysis. No network access or file-writing capabilities are requested or used by these scripts.
Sanitization: No specific sanitization logic is implemented for the data processed from the codebase before it is incorporated into the generated Markdown documentation.

frontend-docs