Built from real-world experience

Claude Crew

A software engineering company inside Claude Code — one manager and twelve specialist agents.

The manager plans, dispatches, and verifies. Specialists read, write, test, research, diagnose, and judge—with a hard boundary around every role.

Get started View on GitHub

1 manager plans and verifies

12 specialists with hard role boundaries

No build step drop-in configuration

Why Claude Crew

Discipline where single-agent setups lose focus.

Each capability has one owner, and the amount of process scales with the risk of the change.

Hard role separation

Reading, writing, testing, diagnosis, and judgment each belong to a defined specialist.

Cost-tiered model assignment

Four role aliases put orchestration, file work, analysis, and trivial edits on the right model tier.

Drop-in configuration

A stack-neutral rulebook, agent definitions, and one profile—with no fork, plugin, or build step.

One manager + 12 specialists

The manager scopes, plans, delegates, and verifies while specialists execute within clear boundaries.

One-file customization

Customize the stack in one file.

This is the crew’s sole customization surface. Edit it to adapt Claude Crew to any project.


              
              skills/stack-profile/SKILL.md

Keep the core stack-neutral. CLAUDE.md and the agent definitions must not be edited to change stacks.

Response language and register
Primary stack
Secondary stacks
UI conventions
High-risk areas
Project rules
Test and type-check commands
Bundled default: TypeScript / Next.js App Router / Express / Tailwind / Persian RTL

How it works

Process scales with risk. Models scale with responsibility.

The manager sizes each task, runs the matching chain, and assigns work through four cost-aware role aliases.

Workflow chains

TRIVIAL[explorer if location unknown] → fixer → reviewer → done
NORMALexplorer → English plan → implementer → reviewer ∥ tester → done
HIGHexplorer → guardian PRE → English plan → implementer → tester → reviewer → security* → guardian POST → done
RESEARCHresearcher is inserted before the English plan whenever external facts are required

Model role aliases

Alias	Used for
lead`claude-lead`	Management, review, security, and regression gates
engineer`claude-engineer`	Reconnaissance, implementation, tests, and prose
analyst`claude-analyst`	External research, root-cause analysis, and bug hunting
assistant`claude-assistant`	Low-risk, single-file TRIVIAL edits

The crew

Twelve specialists. One clear owner for every capability.

Each agent stays inside a hard boundary, keeping execution focused and verification independent.

explorerengineer

Maps the codebase and returns a structured, read-only reconnaissance report.

backendengineer

Implements APIs, business logic, data access, authentication, and integrations.

frontendengineer

Implements UI, components, styling, state, forms, and accessibility.

testerengineer

Writes and runs focused automated tests.

writerengineer

Writes documentation, prose, and copy in the profile language.

researcheranalyst

Researches external documentation, APIs, errors, and version differences.

debuggeranalyst

Performs deep root-cause analysis when the cause is not apparent.

qaanalyst

Hunts for unreported edge cases, race conditions, and broken failure paths.

reviewerlead

Reviews completed changes adversarially and reports defects.

securitylead

Audits trust boundaries, authorization, input, data exposure, and integrations.

guardianlead

Records and checks regression baselines around HIGH-risk changes.

fixerassistant

Applies a few obvious, low-risk lines in one file for TRIVIAL work.

Quick start

Install the complete crew in minutes.

Choose your operating system. Each installation copies the manager rules, all twelve agents, the stack profile, and the gateway settings template.

git clone https://github.com/miladjs/claude-crew.git
cd claude-crew
mkdir -p ~/.claude
cp CLAUDE.md ~/.claude/CLAUDE.md
cp -r agents ~/.claude/
cp -r skills ~/.claude/
cp settings.example.json ~/.claude/settings.json

git clone https://github.com/miladjs/claude-crew.git
cd claude-crew
New-Item -ItemType Directory -Force "$env:USERPROFILE\.claude" | Out-Null
Copy-Item CLAUDE.md "$env:USERPROFILE\.claude\CLAUDE.md"
Copy-Item -Recurse -Force agents "$env:USERPROFILE\.claude\agents"
Copy-Item -Recurse -Force skills "$env:USERPROFILE\.claude\skills"
Copy-Item settings.example.json "$env:USERPROFILE\.claude\settings.json"

git clone https://github.com/miladjs/claude-crew.git
cd claude-crew
mkdir -p ~/.claude
cp CLAUDE.md ~/.claude/CLAUDE.md
cp -r agents ~/.claude/
cp -r skills ~/.claude/
cp settings.example.json ~/.claude/settings.json

Open the copied settings file, set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN, then restart Claude Code.
To change the stack, edit only skills/stack-profile/SKILL.md.