The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes potentially untrusted content (the generated output) within review prompts. While this creates a theoretical surface for indirect prompt injection, the framework mitigates this through independent context isolation for reviewers and structured output requirements.
[COMMAND_EXECUTION]: The framework utilizes the platform's native subagent tools to orchestrate parallel reviews, ensuring that each validation step occurs in a fresh, isolated process.
[SAFE]: The architecture includes safety-oriented design patterns such as convergence loops with iteration caps and explicit adversarial role-setting for reviewer agents to prevent bias.

santa-method