I am building retcon (https://github.com/WujiLabs/retcon), a Claude Code middleman layer allowing the AI agent to edit its own conversation history or rewind itself to a previous round. More than often you ask the AI to do something. It misunderstands. You correct it by words. But the messed-up turn stays in context, and the AI is fighting both your correction and its own past mistake.
retcon flips the model. The tool is named for the verb recontextualize: introducing new information that reshapes how the past is understood. Instead of you typing a /rewind command, the AI retcons its own past. You just tell it what you want.
retcon flips the model. The tool is named for the verb recontextualize: introducing new information that reshapes how the past is understood. Instead of you typing a /rewind command, the AI retcons its own past. You just tell it what you want.