The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Node.js script (scripts/preview.mjs) to perform project inspection, harness initialization, and rendering. This is the primary mechanism for the skill's functionality.
[DATA_EXPOSURE]: The skill scans the local project directory to identify React/Next.js components and CSS styles. It includes security check functions (safeProjectPath and safeHermaiPath) to ensure that file operations are restricted to the repository and the .hermai/ directory, specifically excluding sensitive paths like .env, .git, and node_modules.
[INDIRECT_PROMPT_INJECTION]: The skill represents an indirect prompt injection surface due to its core function of reading and processing untrusted local project files.
Ingestion points: The scripts/preview.mjs script reads content from project files located in identified route directories (e.g., src/app, pages) and style files (e.g., globals.css, *.module.css).
Boundary markers: The skill instructions define a clear 'Privacy boundary' and 'Safety' rules that explicitly forbid reading sensitive files or sending code/data to external servers.
Capability inventory: The skill can read local files, write to the .hermai/ directory, and start a local-only HTTP server (127.0.0.1) for viewing results.
Sanitization: The script implements an escapeHtml function to sanitize dynamic content (such as company names and descriptions) before it is rendered into the preview HTML files, and it applies a restrictive Content Security Policy (CSP) to the local preview server.

brand-preview