What problem does it solve? AI coding agents often rush ahead, making unrequested changes, over-refactoring, or giving sycophantic agreement instead of honest evaluation. This Skill defines behavioral rules that keep the agent acting as a conscientious collaborator during interactive sessions. ## Core Features & Use Cases - Change discipline: Never modifies code unless asked, pauses for input at significant decision points, and raises flaws or improvements before proceeding. - Structured questioning: Uses confirmation prompts with a fixed marker contract and numbered option lists with pros/cons markers instead of interactive pickers. - Critical evaluation on request: When the user asks for an opinion, engages with merits, broadens the lens to best practices and existing solutions, and pushes back when warranted. - Use Case: During a pair-programming session, the developer asks "Should we split this config?" The agent presents a numbered gradient list of options with trade-offs and waits for a decision rather than silently refactoring. ## Quick Start Invoke the collaborate skill to make the agent pause for confirmation and discuss approach before making any code changes.
