The Agent Skills Directory

[SAFE]: The skill serves as a documentation-based architecture companion, providing a 10-point checklist for developers to align with project blueprints, FSM boundaries, and component modularity before writing code.
[COMMAND_EXECUTION]: The instructions recommend using standard development tools such as git grep to identify code consumers and uv run tach check to verify repository dependency boundaries. These are routine operations for maintaining codebase architecture and do not exhibit malicious intent.
[EXTERNAL_DOWNLOADS]: The skill mentions an optional installation of a provider-level architecture skill (claude-api) from the anthropics/skills repository. This is a reference to a trusted source and is presented as a configuration recommendation rather than an automated or hidden download.
[PROMPT_INJECTION]: While the skill contains strong instructions to enforce architectural invariants (e.g., "Never let code and BLUEPRINT drift" or "Never invert an existing must-block regression test"), these are project-specific safety and quality constraints rather than attempts to bypass the agent's internal safety guidelines.

architecture-design