The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the run_command tool to execute local TypeScript scripts (scripts/audit-ui.ts and scripts/check-contrast.ts) via the Bun runtime. These scripts perform deterministic static analysis of frontend code to identify UI anti-patterns and calculate WCAG contrast ratios.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted frontend source code (e.g., JSX, TSX, CSS) to perform audits and refactoring tasks. This creates a potential surface where malicious instructions embedded in the analyzed code could attempt to influence the agent's output.
Ingestion points: User-provided file paths for UI components and templates.
Boundary markers: The instructions mandate a specific report format using a severity table and require "proof" (binding rule + runtime path + deterministic fix) for each finding to ensure the agent ignores non-compliant data instructions.
Capability inventory: view_file, replace_file_content, write_to_file, and run_command (limited to internal scripts).
Sanitization: Findings are derived from regex-based patterns in the audit script and mathematical contrast calculations, rather than dynamic execution of the ingested data.

refactor-ui