Open Steps 🧩 Skill Free

What it does

Open Steps is a pack of Agent Skills that sits between your coding agent and the humans who need to understand what it did. Instead of scrollback full of diffs, tool calls and jargon, it produces short, plain-language reports: a one-screen "done or not" verdict, numbered steps a non-technical person can follow, and a premortem of what could go wrong before a risky change ships.

What is inside

Eight skills, each a self-contained SKILL.md with reference examples:

• os-done-or-not: a one-screen completion verdict
• os-step-by-step: numbered instructions for non-developers
• os-ask-simple: rephrases the agent's clarifying questions
• os-what-could-go-wrong: a premortem before risky decisions
• os-whats-next: suggests the next task with reasoning
• os-check-work: verifies claims made in earlier sessions
• os-say-simple: rewrites technical text in plain language
• os-big-picture: a living map of feature status and backlog

The repo also ships session-start/stop hooks and an evals folder used to measure the skills' output quality.

Works with

Full support, including the plugin and routing hook, is for Claude Code. Codex, Cursor and Gemini CLI are supported by copying the skill folders directly, without the hook automation.

How to install

For Claude Code:

git clone https://github.com/kharmanskyi/open-steps.git
claude plugin marketplace add ./open-steps && claude plugin install open-steps@open-steps
grep -q 'os-done-or-not' ~/.claude/CLAUDE.md 2>/dev/null || cat open-steps/docs/routing-block.md >> ~/.claude/CLAUDE.md

For Codex, Cursor or Gemini CLI:

mkdir -p ~/.agents/skills && cp -R open-steps/skills/os-* ~/.agents/skills/

Maintenance and safety

MIT licensed with roughly 950 GitHub stars, 91 forks and commits as recent as mid-September 2026. The skills only reformat and summarize output your agent already produced; they don't add network calls or credentials.

Who should use it

Useful for anyone shipping AI-agent work to non-technical stakeholders: freelancers reporting to clients, teams with a mixed technical/non-technical audience, or solo builders who want an honest, jargon-free check on whether a task is actually done.