The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from architectural documentation and code changes. Ingestion points include docs/architecture/INDEX.md and code diffs. While it lacks explicit boundary markers or sanitization for these inputs, it also lacks any exploitable capabilities such as network access or shell execution, neutralizing the threat.
[SAFE]: The skill is purely instructional and advisory, focusing on semantic judgment of architectural invariants without performing sensitive operations. No remote dependencies, obfuscation, or persistence mechanisms were detected.

inspect-invariants