Claude Code PM is a set of workflows that directly converts product ideas into GitHub Issues and code through “spec document-driven steps”+”parallel AI agents”. It supports instructions such as/pmepic-oneshot splitting tasks and/pm:issue-start executing tasks. It can completely preserve the context, achieve team collaboration through GitHub, and ensure that every line of code can be traced back to the corresponding product specification.
Benefits of using it:
- 3x faster delivery
- Defect rate reduced by 75%
- 89% reduction in context loss
Multiple agents can be used simultaneously to further improve R & D efficiency
Recently, there is a more interesting project in the AI programming circle: CCPM (Claude Code Project Manager). It is not a new AI model, nor is it a code generation tool, but a set of Development workflow around AI programming and design。It attempts to solve a problem that many people already encounter: When AI is involved in writing code, project development can easily become confusing.
The way many people write code now is actually like this: open Claude Code or other AI assistant, describe a requirement, let it generate code, and then continue the conversation to modify it. It seems efficient in a short period of time, but as the project gets bigger, problems arise-the context begins to be lost, there is no clear correspondence between requirements and code, and codes generated by different AI are prone to conflicts.
This is exactly what CCPM wants to solve. It proposes an idea:Integrate AI programming into standard software engineering processes, rather than relying on chat-based development.
The core of the entire workflow is a very simple chain:
Product Specification (Spec / PRD)
→ Epic
→ GitHub Issue
→ CodeWhen a new product idea emerges, developers first write a specification document rather than having AI write code directly. Next, through CCPM commands, this specification can be automatically disassembled into multiple tasks. Each task becomes a GitHub Issue and can be handed over to an AI agent to implement it.
CCPM provides some commands to drive this process. for example /pm:epic-oneshot Task structures can be automatically generated based on specification documents, and /pm:issue-start An AI agent will be launched to perform the corresponding task. This way, each piece of code can be traced back to an Issue, and each issue comes from the original product specification.
The entire system runs around GitHub. Requirements, tasks, and submission records are all kept in the warehouse, so the development process is naturally traceable. This is actually very close to the traditional software engineering process, but now it is no longer entirely human developers who perform tasks, but AI.
Another key design is parallel development. CCPM uses Git’s worktree mechanism to create a separate development environment for each task. In this way, multiple AI agents can develop different modules at the same time without interfering with each other. For example, one agent is responsible for the API, one for the front-end components, and another for testing the code. Each task has its own branch and context, which is finally merged back into the main branch through Pull Request.
This method is in sharp contrast to the “AI chat and writing code” that many people now commonly use. The author even specifically proposed a concept called “No Vibe Coding”。This means that instead of relying on feelings and chat to drive development, drive AI through clear specifications and task structures.
From a technical perspective, this project is not complicated. The content in the repository is mainly scripts, commands, and templates to help the team combine requirements documents, GitHub Issues, and AI agents. it is more like a Implementation tools for development methodology, rather than a huge framework.
But the direction it represents is actually quite interesting. As AI programming tools become stronger, the development process itself is changing. In the past, software engineering was “people wrote code and assisted by tools”, but now it has gradually become “AI writes code and people designs the process.” In this model, what really matters may no longer be the code itself, but rather How to organize tasks, how to define specifications, and how to manage agent collaboration。
CCPM is trying to answer this question: what should the development process look like if the software of the future is done by multiple AI agents?
From this perspective, this project is not just a tool, but more like an exploration of future AI software engineering.
Github：https://github.com/automazeio/ccpm
Oil tubing: