The Agent Skills Directory

[SAFE]: The skill implements a structured adversarial code review process using model diversity. The configuration is managed through local files (e.g., ~/.claude/rules/pstack-models.md) which is a standard practice for agent settings.
[SAFE]: All external references target local files within the skill package (references/reviewer-prompt.md, references/rubric.md, etc.), ensuring execution context remains controlled.
[SAFE]: The use of subagent tools (Task, Agent, spawn_agent) is consistent with the intended purpose of spawning reviewer models. The instructions explicitly warn against auto-applying changes, maintaining a human-in-the-loop safety boundary.
[SAFE]: Analysis of the rubric and lead judgment framework shows a focus on identifying legitimate software engineering concerns (correctness, security, maintainability) rather than malicious behavior.

interrogate