The Agent Skills Directory

[SAFE]: The skill consists of markdown documentation and architectural rules. It does not include any executable scripts, obfuscated payloads, or suspicious network operations.
[SAFE]: The instructions actively promote security best practices, including the use of T3 Env for strict environment variable validation, the prevention of sensitive data leakage in logs, and the requirement for signature verification on webhooks.
[INDIRECT_PROMPT_INJECTION]: The skill functions by analyzing and reorganizing project source code, which serves as an untrusted data ingestion surface.
- Ingestion points: The agent reads the directory structure and the contents of source files (routes, schemas, and services) within a user's repository.
- Boundary markers: The skill includes a priority rule stating that local repository documentation (e.g., AGENTS.md) overrides the skill's instructions unless a migration is explicitly requested.
- Capability inventory: The agent has the capability to read, write, and move files to scaffold new features or reorganize the project layout.
- Sanitization: There are no explicit instructions for the agent to sanitize or escape project content it reads before processing it for architectural review.

elysia-architecture