Jesse Vincent writes the plan for an executor with no memory
A hand-off habit worth borrowing from superpowers.
Coding agents lose context. Jesse Vincent's open-source superpowers wraps them in a development methodology built from small composable skills, and the agent checks for a relevant skill before starting any task: brainstorm first, then design, then plan, then implement.
The specific choice sits inside writing-plans. Work is broken into tasks of two to five minutes, and every task carries exact file paths, complete code, and verification steps. The plan is written to be followed by someone with no project context at all, because that is exactly who executes it: the subagent-driven-development skill dispatches a fresh subagent per task, reviews each task for spec compliance and code quality, then runs one broad review of the whole branch at the end.
There is a trade-off. This front-loads design and planning effort, and a very detailed plan can go stale as the code moves. The repository itself ships a cheaper path, executing-plans, which runs everything inline in one session with a single final review. What is worth borrowing is the direction of the hand-off: instructions written to survive a context boundary, and review between tasks instead of only at the end.
Repository review, not a hands-on benchmark. Superpowers was not installed or run here; its workflow claims are the project's own.
Sources & credit: Repository & documentation · @obra · GitHub Trending · Observed 2026-11-20