The Agent Skills Directory

[SAFE]: The skill is explicitly designed for architectural scaffolding in a non-executing mode. The SKILL.md file defines mandatory safety boundaries that forbid importing provider SDKs, reading credentials, or creating live transactions.
[INDIRECT_PROMPT_INJECTION]: The skill manages an attack surface where untrusted data (order IDs, user emails) is ingested into database operations. This is mitigated effectively: 1) Ingestion points: orderId and user_email in API routes and database helpers. 2) Boundary markers: SKILL.md requires validation and idempotency patterns. 3) Capability inventory: Database writes via pg, Prisma, and Supabase clients. 4) Sanitization: The provided implementations use parameterized queries and a strict column whitelist (ALLOWED_UPDATE_COLUMNS) in references/database-postgresql.md to prevent SQL injection.
[CREDENTIALS_UNSAFE]: Static analysis flagged database connection strings; however, these are identified as standard documentation placeholders (e.g., postgresql://user:password@host:5432/dbname) rather than hardcoded secrets.
[EXTERNAL_DOWNLOADS]: The skill references standard, well-known packages (pg, prisma, @supabase/supabase-js) from official registries for development purposes, which are considered safe.

mercadopago-integration