The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates application code that handles external data (URL parameters and database records), creating a potential attack surface for indirect prompt injection if those values were to be used in LLM prompts later.
Ingestion points: URL parameters such as :account_id are processed in the LoadAccount plug and on_mount hooks within references/auth.md.
Boundary markers: The generated application code focuses on standard web logic and does not implement LLM-specific boundary markers.
Capability inventory: The generated code handles database interactions via Ecto and manages user navigation through Phoenix.LiveView redirects.
Sanitization: The skill incorporates strong sanitization practices, including the use of Ecto.UUID.cast for identifier validation and strict scoping of account access to the authenticated user via AccountContext.get_account_for_user/2.
[CREDENTIALS_UNSAFE]: The priv/repo/seeds.exs template in references/router-seeds.md contains a hardcoded password (password1234password1234) for a test user. This is a common development-time placeholder for local testing and does not represent a leak of production secrets.

bootstrap-accounts