Claude Code Multi-Agent workflow
A Claude Code workspace folder: hooks with a local Ollama detect the project type, pick skills and remind you to update the docs
- 305 ★ on the repository
- Last commit
Extended access
Why this level
- Python hooks run on every session start and after code edits
- The local Ollama model reads project files to detect its type and framework
- The hooks create and change documentation files in the project's `project_document/`
Install
- Works in
Manual install
git clone https://github.com/Prorise-cool/claude-code-multi-agent.gitThen `cd claude-code-multi-agent`, put the project in this folder and open it in Claude Code. Ollama and `uv` are installed separately.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Python hooks run on every session start and after code edits
- The local Ollama model reads project files to detect its type and framework
- The hooks create and change documentation files in the project's `project_document/`
This is third-party code. Review the repository files before installing.
What it does
The repository is meant as a ready workspace for Claude Code rather than a plugin: you clone it, put your project inside the folder and open that folder in Claude Code. Python hooks run through `uv run` and do the rest. At session start (the `SessionStart` hook) a local Ollama model (`gemma3:1b` by default) detects the project type and framework, for example Python with FastAPI or Node.js with React, and Claude Code gets that context without a manual description. The hook then finds all skills and loads them; the README claims more than 300. The `.claude/agents` folder holds agent groups: business analysis, UX research, documentation and specialists (CSS, Python, TypeScript, plus the chain spec-analyst, spec-architect, spec-developer, spec-planner, spec-reviewer, spec-tester and spec-validator). `.claude/commands` holds commands, for example a business analysis set: capability map, journey map, business model canvas. Skills are called as `/backend-specialist`, `/frontend-specialist`, `/testing-specialist`, `/security-specialist`, `/architecture-specialist` and `/devops-specialist`. Intent analysis tells a simple question from a complex task and recommends MCP tools (Sequential Thinking, Task Manager) and fitting skills. The hooks create three files in `project_document/`: `DEVELOPMENT.md`, `KNOWLEDGE.md` and `CHANGELOG.md`. After a code edit Claude Code gets a reminder to update them, and at session start the files go into context instead of Memory MCP. The hooks also read git settings and name a branch strategy (github-flow or git-flow). Prompt texts live in `prompts.json`, and voice output (`HOOKS_TTS_ENABLED`) is off by default.
Who it is for. People who work in Claude Code across several projects and want it to recognize the stack, pick skills and keep documentation without manual setup.
Good fit when
- You keep explaining to Claude Code what stack a project is written in
- You want ready specialists (`/backend-specialist`, `/testing-specialist`) instead of long prompts
- Docs drift from the code and you need a forced reminder after each edit
- A task is vague and you want hints on which MCP tools and skills fit it
- You need a chain of roles from requirements analysis to review: the agents spec-analyst, spec-architect, spec-developer and spec-tester
Not a fit when
- You want a plugin that installs into an existing project with one command: here the project moves into the cloned folder
- Ollama and `uv` cannot be installed on the machine: without Ollama only a basic mode remains, where skills run by hand
- Plain Claude Code without hooks is enough and extra Python runs on every session get in the way
Example request
What project type is this and which skills did you load? Then use `/backend-specialist` to design a REST API for authenticationLimitations
- The README and docs are written in Chinese. There is no English or Russian translation.
- The repository is not a plugin: you put the project inside the cloned folder or open that folder in Claude Code.
- Project detection and intent analysis rely on Ollama. Without it a basic mode with manual skills remains. The `gemma3:1b` model is small, and the quality of detection depends on it.
- The skill count (over 300) comes from the README and was not verified in the file tree separately.
How to disable. The hooks are wired to settings inside the repository folder itself, so opening the project in Claude Code outside that folder or deleting the clone is enough. The `project_document/` directory with the three docs stays in the project and can be removed by hand.
FAQ
How do I install Claude Code Multi-Agent?
Clone the repository with `git clone https://github.com/Prorise-cool/claude-code-multi-agent.git`, install Ollama and `uv`, pull the model with `ollama pull gemma3:1b`. Then put the project inside the folder and open it in Claude Code.
How do I know the install worked?
After the first message in Claude Code three signs appear: a detected project type (for example Python), a count of loaded skills and a `project_document/` directory with three doc files.
Is Ollama required?
It is recommended: it detects the project type, analyzes intent and picks skills. Without it a basic mode remains, where skills run only by hand.
Which skills can I call?
For example `/backend-specialist`, `/frontend-specialist`, `/testing-specialist`, `/security-specialist`, `/architecture-specialist` and `/devops-specialist`. Claude Code gives the full list when you ask it to list the available skills.
Which doc files are kept automatically?
`DEVELOPMENT.md` with tasks and progress, `KNOWLEDGE.md` with decisions and code patterns, `CHANGELOG.md` with the change history. They live in `project_document/`, and after a code edit Claude Code gets a reminder to update them.
Discussion
No questions or reviews yet.
Related
WorkflowsA template for Claude Code projects: parallel agent swarms, workflow skills, safety hooks and rules in one install.
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
A Russian-language Claude Code course of 12 modules and 8 labs with a ready kit of skills, subagents and hooks for a project
A learning kit for Claude Code: feature walkthroughs plus copy-paste templates for commands, skills, memory, hooks and subagents
An environment for Claude Code that keeps context across sessions via hooks, ledgers and handoffs, orchestrates agents and saves tokens
Multi-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
README in short
The Chinese README opens with a definition of the project and the four pains it removes: no project awareness, manual setup, no skill recommendations and forgotten docs. Then come the benefits, a five-minute quick start installing Ollama and `uv`, detailed instructions, usage examples and a contributors section.