The Agent Skills Directory

[SAFE]: The skill implements strong security boundaries, including a 'Schema Isolation' rule that prevents unauthorized database access across project boundaries.
[SAFE]: Cross-Site Scripting (XSS) protections are enforced throughout the UI layer using a dedicated out() helper that applies htmlspecialchars to all dynamic content rendered in PHP templates.
[SAFE]: Database security is handled via Supabase Row-Level Security (RLS) policies, ensuring that anonymous users can only read non-deleted records and perform write-only lead submissions.
[SAFE]: Sensitive environment variables (Supabase keys) are managed via .env files which are correctly excluded from version control via .gitignore rules established in the project genesis step.
[COMMAND_EXECUTION]: The workflow utilizes standard system utilities such as psql for database interaction, curl for smoke testing endpoints, and composer for dependency management. These are used in a controlled manner following hardcoded developer instructions.
[EXTERNAL_DOWNLOADS]: Dependencies (e.g., vlucas/phpdotenv) are pulled from trusted registries (Packagist/Composer), following standard development practices.
[SAFE]: The skill involves a 'Brain Hardening' step where an agent records project metadata into a markdown snapshot for future AI sessions.
Ingestion points: Metadata and aggregate record counts from the Supabase database.
Boundary markers: Structured markdown headers and sections in the snapshot file.
Capability inventory: Reading/writing local files and executing database queries via psql.
Sanitization: Content is limited to schema definitions and numerical counts, which naturally mitigates the risk of malicious user-supplied data being reflected in the snapshot.

claude-website