Open Steps: 7 Skills Making AI Agents Report in Human Language

Pavlo, a non-engineer with 20 years of product experience managing 50+ developers, created Open Steps to solve the problem of AI agents reporting in technical jargon. The system includes 7 Skills that force agents to communicate in human-readable language. After completing work, agents produce a one-screen report starting with a clear completion status, then answer three questions: what action is needed from the user, whether new issues were created, and whether the task can be closed. Bad news is displayed on a separate line, not hidden in paragraphs. Steps requiring human action are numbered, with agents completing their part first. For irreversible decisions like contracts or migrations, a separate uninvolved agent is tasked with reverse-engineering reasons for assumed failure. The writing style borrows from aviation manual simplified English: short sentences, active voice, one meaning per sentence. Testing showed 100% trigger rate on Opus 5 and 95% on Sonnet 5 across 21 daily queries repeated three times. The system installs as a Claude Code plugin and is compatible with Codex, Cursor, and Gemini CLI.