Fix Messed-Up Git State with Safe Commands

The biggest risk when asking an AI for Git help is saying "undo this" without providing your actual state. When lacking context, the AI typically assumes the most generic path—often recommending hard resets or force pushes. Executing those blindly can silently erase uncommitted changes or overwrite a teammate's commits.

This is why context is split into two distinct parts. A user's written description can be inaccurate or based on a false premise (e.g., "I thought I checked out a branch"). However, git status output is ground truth. Providing both and requiring the AI to restate the state based on the status in format item ① ensures any misunderstandings are caught immediately.

The clarifying questions requirement forms the second safety net. Safe commands depend entirely on whether commits are already pushed or shared, which cannot be determined from local git status alone. Forcing questions first prevents destructive suggestions.

The three constraints prevent separate failure modes: requiring explicit warnings stops history loss, forbidding command chaining prevents subsequent commands from running if a prior step fails, and disallowing alternative options stops users from accidentally mixing incompatible recovery methods. Wrapping user data in """ prevents commit messages from being interpreted as prompt instructions.

Without knowing the repository state, the AI will provide a generic solution like git reset --hard and a force push. Uncommitted changes might be wiped out permanently, or remote history might be overwritten, breaking sync with teammates. Without explanations for each command, you won't know what changed, leading to running more random commands from Google and worsening the state.