The Agent Skills Directory

[SAFE]: The skill operates entirely on local project documentation (docs/architecture/) and source code/designs. No network access or data exfiltration paths were identified.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from design documents and git diffs to perform semantic analysis. As a semantic checker, it is inherently exposed to data-driven instructions, but its risk is minimal due to lack of sensitive capabilities.
Ingestion points: Reads design.md and executes git diff on the repository content.
Boundary markers: The instructions do not define specific delimiters for separating the subject data from the analysis instructions.
Capability inventory: The skill is limited to generating text-based architectural verdicts; it has no file-writing, network, or command-execution tools.
Sanitization: No explicit sanitization or filtering of the input content is described.

review-invariants