The Agent Skills Directory

[SAFE]: The skill defines a legitimate architectural workflow consisting of grounding, sketching, agreement, and implementation phases. It focuses on software design principles such as interface depth, information hiding, and invariant encoding.
[DATA_EXPOSURE]: The skill references local configuration files (e.g., ~/.cursor/rules/pstack-models.mdc) to identify preferred AI models for its 'arena' phase. This is a standard local configuration lookup for developer tools and does not involve the exposure of credentials or sensitive personal information.
[COMMAND_EXECUTION]: While the skill orchestrates other agent skills (like arena, how, and why), it does not define or execute any arbitrary or dangerous shell commands. The environment is further restricted by the disable-model-invocation: true setting in the frontmatter.

architect