kharmanskyi/open-steps

What it solves

Open Steps is designed for non-technical builders who use AI agents (like Claude Code, Cursor, or Gemini CLI) to develop software. It solves the "communication gap" where agents provide overly technical reports filled with jargon and commit hashes, making it difficult for a non-engineer to determine if a task is actually finished or what the next steps are.

How it works

The project provides a "pack" of specialized skills and hooks that can be installed as plugins or added to agent configuration files. These skills guide the AI agent to communicate in plain language, following simplified English principles (inspired by ASD-STE100).

Key components include:

• Skills: Specific instructions (e.g., os-done-or-not, os-big-picture) that the agent invokes to generate plain-language reports, checklists, and project maps.
• Hooks: Shell scripts that manage session starts and ends, ensuring the agent provides a summary report when work concludes.
• Routing Block: A set of rules added to the agent's system instructions (like CLAUDE.md) to ensure the agent consistently uses the correct skills.
• BIG-PICTURE.md: A living document that tracks the overall product state, features, and backlog, serving as a shared memory for both the human and the agent.

Who it’s for

Market-led builders and product-focused individuals who are "vibecoding" or building products without a deep engineering background and need the AI agent to report progress and ask questions in a way that is accessible and actionable.

Highlights

• Plain-Language Reporting: Converts technical jargon into clear verdicts (e.g., "People can sign in again") and structured tables.
• Comprehensive Skill Set: Includes tools for verifying work (os-check-work), predicting failures (os-what-could-go-wrong), and simplifying complex text (os-say-simple).
• Project Mapping: Maintains a BIG-PICTURE.md file to track feature progress and identify "quiet" or rotting code via git analysis.
• Cross-Agent Compatibility: Works with Claude Code, Codex, Cursor, and Gemini CLI.
• Evidence-Based: Includes a testing suite that measures skill activation rates across different LLM models (Haiku, Sonnet, Opus).