kasparovabi/claude-mythos-scaffold — reverse-engineered prompt

Build me a small Claude Code scaffold that makes the agent work in a more disciplined, systematic way, without pretending it can do magic. I want it to package reusable prompt skills for planning, breaking a task down, choosing tools, verifying results, reflecting on mistakes, and recovering when a step fails. It should feel inspired by the more persistent Claude Mythos style, but be honest that this only narrows part of the gap and does not replace real reasoning.

Please make it easy to drop into a normal Claude Code setup, with a simple slash command like /mythos-mode that activates the workflow. Include two optional modes, one for long form research synthesis across multiple sources, and one for codebase migration or framework upgrade work. If possible, add a lightweight memory or vault friendly option and a sync hook for notes.

Keep the docs practical and clear, including what is tested, what is newer, and what the limits are. You can look up current Claude Code conventions if needed.