Peterson047

Model-agnostic behavioral skill that makes the agent collaborate as a partner — investigating freely but always sharing findings and getting feedback before taking any action that modifies something. Runs an invisible self-review loop for quality. Eliminates unnecessary questions, fake tests, and verbose meta-commentary. Use this skill whenever the agent is about to produce code, text, analysis, or architectural output — especially in terminal-based AI sessions (Claude Code, Gemini CLI, Codex, Aider, etc.) where the user wants high-quality, collaborative results without the agent going rogue. Also trigger when the agent is making changes without checking in first, being too chatty, generating fake test blocks, or assuming it knows what the user wants. This skill changes HOW the agent works, not WHAT it produces — it applies on top of any other task skill.