mfarzanansari/praxis-workflow-os

Praxis Workflow OS is a personal productivity system built on top of AI coding agents like Claude Code and Codex. Instead of handing you a generic set of prompts, it interviews you about your actual work, builds a custom workflow plan, and then helps you improve that plan over time based on real evidence. It is designed for a wide range of professionals, including writers, designers, researchers, and founders. The system starts with a simple model of work: a trigger leads to evidence, which informs decisions and actions, producing deliverables that get verified and handed off, leaving behind durable lessons. Praxis uses an AI agent to conduct an adaptive interview, asking one question at a time to understand how you work. It then uses your answers to create a blueprint, which is a small map of skills, a knowledge model, and a set of approval gates for your tasks. The repository contains several specialized skills. The main skill routes your request to the right tool. Other skills handle the interview, blueprint creation, setup, and a process for converting useful session residue into curated notes. There is also a retrospective skill that reviews your work history to find friction and suggest improvements. You can install the whole system or just a few pieces, and it includes tools to safely build an Obsidian vault without overwriting your existing notes. Praxis separates five concerns that often get tangled in personal AI systems: intent, orchestration, execution, knowledge, and runtime memory. By keeping these distinct, it avoids the common failure modes of generic prompt packs that never learn your habits, or sprawling agent collections that become too complicated to manage. It treats your workflow as an operating model with explicit contracts and human approval gates. You can install Praxis through a few different paths, including a cross-client CLI, a local Python installer, or native plugin marketplaces. A companion project called Prompt Polish can optionally help refine the specific prompts your agent uses. The full README is longer than what was shown.