The Agent Skills Directory

[SAFE]: The skill operates as a read-only design workflow. It is explicitly instructed not to edit files, commit changes, or perform network operations, which significantly limits the attack surface.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as repository instructions, architecture decisions, and code. While this presents an ingestion surface for untrusted data, the risk is minimal because the skill lacks capabilities to execute commands, write to the filesystem, or perform network requests. It also includes specific instructions to ignore non-normative content and respect authority boundaries.

codebase-design