Your AI coding agent is great at the work and bad at explaining it. It finishes the job, then reports back in commit hashes, branch names and words you've never heard. A response, technically. Useless unless you wrote the code yourself. That's what Open Steps fixes. A free, open-source skills pack for Claude Code (it also installs on Codex, Cursor and Gemini CLI) that turns your agent's report from engineer-speak into plain language. Instead of "Session TTL misconfig in auth middleware caused 401 cascades", you get "People can sign in again. The fix is live for everyone." Plus a straight verdict: done or not, anything needed from you, any new debt left behind. The pack has 8 skills now: ▪️os-done-or-not: an honest wrap-up. What got done, is it really done, is anything needed from you ▪️os-step-by-step: the agent does everything it can itself first, then gives exact steps for the rest ▪️os-check-work: doesn't trust another session's report, re-checks the actual work ▪️os-what-could-go-wrong: before anything hard to undo, hands it to a fresh agent whose only job is to find what could break ▪️os-ask-simple: when the agent needs a decision, one plain question and one recommendation ▪️os-whats-next: the next step, based on the real state of your project ▪️os-say-simple: any text in plain words, not just agent reports, nothing left out ▪️os-big-picture: where the whole project stands, written down in one file Find it here: https://t.co/m1sjopBSxr MIT licensed. No account, no key, nothing phones home. Built for people who ship a product with an agent but don't read the code. https://t.co/IIcWBu99Fx