Claude Code Multi-Agent Development Sample workflow
A Claude Code sample: a lead agent, pools of coders, DevOps and reviewers, a spec and four test tiers
- 75 ★ on the repository
- Last commit
- MIT-0
Broad access
Why this level
- The configuration changes `~/.claude`, wires hooks and lets agents spawn other agents
- The DevOps role works with AWS infrastructure and the sample is explicitly marked as not fit for production
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/claude-code-multi-agent-development-sampleDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The configuration changes `~/.claude`, wires hooks and lets agents spawn other agents
- The DevOps role works with AWS infrastructure and the sample is explicitly marked as not fit for production
This is third-party code. Review the repository files before installing.
What it does
The repository holds a sample `.claude` directory for team development with several agents. The main agent takes the lead role: `fullstack-agent` studies the task, writes `spec.md` and `design.md`, builds a `tasks.md` plan of file-disjoint tasks and hands over a launch package. It then spawns pools of background agents: up to six coders, up to two DevOps and up to four reviewers that claim free tasks from a shared queue. Review is chosen by change size: mechanical edits go to `review-lite-agent` while anything touching access, auth, resources or deploys goes to `review-agent`. There are three review cycles per scope and a final end-to-end review of the assembled system. `sa-agent` can be added for a Well-Architected assessment. Every stage is gated by four test tiers from `rules/testing-standards.md`: unit, integration, verification of deployed resources under the application's real identity and end-to-end journeys. The set includes Python hooks, AWS security rules, four skills and the `brainstorm` and `optimize-my-claude` commands.
Who it is for. Teams building an AWS application with Claude Code who want to take a ready scheme of roles, checks and tests as the basis for their own.
Good fit when
- A feature should be carried from spec to a deployed application and you need a way to split it into parallel tasks
- You must decide which changes need a quick review and which need a full one
- You want agents to verify resources under the application's real identity rather than against mocks
- A team needs a starting point to take apart and adapt to its own rules
Not a fit when
- You need a production-ready set: the authors state the sample is not approved for production
- The project is not on AWS: the security rules and the DevOps agent are tuned to Amazon cloud
- Your own setup already sits in `~/.claude`: copying overwrites files with the same names
Example request
Take the lead role: study the task, write a spec and a plan of independent tasks, then start the pool of coders and reviewersLimitations
The authors note that the repository is a sample and is not approved for production: settings must be reviewed and adapted to the team and security policy, and an organization needs legal review and the LLM Legal Approval table filled in before use. Installation copies files into `~/.claude` and overwrites same-named ones, so back up first or merge selectively. The agent-pool scheme spends many tokens and review cycles are limited to three per scope. Models and effort set in frontmatter apply only when a role is spawned as a subagent.
How to disable. Restore the `~/.claude.bak` backup or delete the copied `agents/`, `rules/`, `skills/` and `hooks/` folders and the hooks block in `settings.json`. After a selective merge delete only the added files.
FAQ
Which roles are in the sample?
The lead `fullstack-agent`, then `coding-agent`, `devops-agent`, `review-agent` and `review-lite-agent`, and on demand `sa-agent` as an AWS architect.
How do I install without losing my own setup?
Copy `~/.claude` to `~/.claude.bak`, then move files with `cp -rn agents/ rules/ skills/ hooks/ ~/.claude/`. The `-n` flag does not overwrite existing files.
What are the four test tiers?
T1 unit, T2 integration and contract, T3 deployed-resource verification under the application's identity with no mocks and T4 end-to-end journeys against the spec acceptance criteria.
Why does the lead work in the main agent?
A subagent cannot spawn its own subagents and the harness allows one nesting level, so only the main agent can drive the pools.
Is the sample fit for production?
No, the authors state it is not approved for production and serves as a starting point. An organization needs legal review before use.
Discussion
No questions or reviews yet.
Related
WorkflowsA GitHub Action that runs Claude Code in pull requests and issues: answers mentions, reviews and makes changes
Claude Watch draws a localhost map of project logic, shows Claude Code tool calls live and answers questions
Claude Code plugins by shokai: Codex second opinions, pull request review, repository security patrol and a factory mode.
Continuous Claude
continuous-claude
Claude Code in a continuous loop: iteration, pull request, waiting for checks and merge. Stops on several completion signals. jq and gh are needed
A Rust CLI and TUI: agents in containers, parallel tabs, TOML workflows from issue to PR and a scheduled agent squad
A container for Claude Code, Codex, Copilot, OpenCode and Pi: the agent runs without prompts while a gateway limits network.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Claude Code Multi-Agent Development Sample (this card) | yes | 75 | |
| AWS Agent Skills | no | 1,175 | skills for AWS services with no pool and review scheme |
| AWS DevOps Agent Tools | yes | 141 | forty-five skills and CloudFormation templates for AWS DevOps Agent |
| Gem Team | no | 229 | 12 roles and a plan, TDD, verification and learning process |
README in short
The description presents the sample as a setup for multi-agent, spec-driven development: a lead and three specialists backed by skills, rules and plugins. Agent tables with models, a five-step process, four test tiers, installation and a legal notice follow.