The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for architectural auditing. Its instructions explicitly enforce a read-only policy for existing codebase files, focusing only on design documents.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes untrusted documentation content.
Ingestion points: Reads external content from LLD files (.maestro/<slug>/lld/*.md) and OpenAPI specifications.
Boundary markers: Absent; the skill does not wrap ingested content in delimiters or ignore-embedded-instruction warnings.
Capability inventory: The skill is granted Bash, Write, and Task tools. While the instructions emphasize read-only behavior, the presence of these tools combined with untrusted data ingestion represents a potential surface.
Sanitization: None; the skill processes design documentation as raw text without validation or filtering.

architecture-review