The Agent Skills Directory

[SAFE]: The skill implements a rigorous 'blind panel' architecture for software assessments. It spawns independent sub-agents with restricted, read-only permissions to verify evidence. The instructions explicitly forbid these agents from editing tracked files, accessing secrets without permission, or interacting with external systems without a human-in-the-loop checkpoint.
[SAFE]: All external downloads and references point to the vendor's own local repository structure (references/*.md) or use standard platform tools. There are no calls to untrusted remote sites or obfuscated code execution patterns.
[SAFE]: The skill includes defensive instructions to prevent 'self-scoring' or intent-based reporting, requiring reproducible commands and artifacts for every readiness claim. It mandates that any 'outward-facing' action—such as deployments, DNS changes, or paid resource allocation—requires explicit user consent via a 'Checkpoint Question' workflow.

trl