The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local data by reading project source code, database migrations, and configuration files to ensure code consistency. This ingestion point lacks explicit boundary markers or sanitization for embedded instructions. The agent's capabilities include writing files and executing local build tools (make, sqlc, go mod), creating a potential surface for instructions within source files to influence behavior, though this is inherent to its role as a coding assistant.
[COMMAND_EXECUTION]: The agent is instructed to run standard developer tooling including 'sqlc generate', 'go mod tidy', and 'make build'. These commands are restricted to the local project environment for the purpose of code generation and verification, which is consistent with the skill's primary purpose.
[SAFE]: The skill explicitly mandates multiple security best practices, such as requiring 'HttpOnly', 'Secure', and 'SameSite' flags for session cookies, using bcrypt for password hashing, and ensuring session tokens are stored hashed and deleted upon logout. It also contains rules to prevent sensitive data from being logged or returned in error responses.

genesis-auth