alanhuang168/ai-project-workflow — explained in plain English
Analysis updated 2026-05-18
Force an AI coding assistant to write down requirements, architecture, and design docs before writing code.
Track project progress stage by stage through a state file, with human approval gates between stages.
Install the same structured workflow across multiple AI tools like Cursor, Claude Code, Codex, and TRAE.
Validate that a project's adapter files and workflow state are in sync using the apw CLI.
| alanhuang168/ai-project-workflow | ajaxfnc-yt/d.w.i.f | exportanything/comfyui-ltx-director-motion-brush | |
|---|---|---|---|
| Stars | 17 | 17 | 17 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | easy | moderate | moderate |
| Complexity | 3/5 | 2/5 | 3/5 |
| Audience | developer | developer | vibe coder |
Figures from each repo's GitHub metadata at analysis time.
Installs via npx or a global npm package, requires an AI coding assistant already set up in the project.
AI Project Workflow, also called APW, is a structured software delivery process built for AI coding assistants such as Cursor, Claude Code, Codex, Qoder, CodeBuddy, CatPaw, and TRAE. It exists to solve a specific problem: AI coding assistants often start writing code before requirements, architecture, design, and tests are actually clear, which leads to unwanted features, quietly expanding scope, and tests that get reported as passing without ever really running. The project works by keeping one core set of rules, skills, agent definitions, templates, and schemas as the single source of truth, then generating a lightweight adapter for whichever AI tool is being used. Every project that installs it gets a hidden .ai-workflow directory holding the current stage, completed stages, and version information, alongside visible entry files like AGENTS.md and CLAUDE.md that the AI tools read automatically. Older projects using a root level core or VERSION folder still work, since APW checks the newer location first and falls back to the old layout. The workflow itself has eight stages: init, prd, hld, sdd, impl, review, deploy, and retro. Each stage reads specific upstream documents, produces a specific output file under docs, and requires the human user to review and approve before the AI can move to the next stage. For example, the prd stage produces docs/PRD.md from the user's description, and the hld stage produces docs/ARCH.md from the PRD. Documents are treated as authoritative in order, so if code and the detailed design document disagree, either the code changes or the design document gets formally revised first, never a silent drift between the two. The project ships an apw command line tool for installing the workflow into a project, syncing adapter files, validating structure, checking status, and migrating from older layouts. It never overwrites existing user files, instead writing a .new file alongside any conflict for comparison. Installation and setup have been verified against Cursor, Claude Code, Codex, TRAE, and CodeBuddy, with Qoder still pending verification due to insufficient test account quota. The project is released under the MIT license.
AI Project Workflow is an eight-stage, document-gated software delivery process for AI coding assistants like Cursor and Claude Code, forcing requirements, design, and tests to be written down and approved before code moves forward.
Mainly JavaScript. The stack also includes JavaScript, Node.js, CLI.
MIT license: free to use, modify, and distribute, including commercially, as long as the copyright notice is kept.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.