Skip to content

README.md

Latest commit

 

History

History
184 lines (149 loc) · 6.01 KB

README.md

File metadata and controls

184 lines (149 loc) · 6.01 KB

English | Русский

Team

Build features with an AI agent team and built-in review gates.

Prerequisites

Agent Teams are experimental and disabled by default. Enable them before using this plugin.

Add CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS to your settings.json or environment:

// ~/.claude/settings.json
{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Or set the environment variable:

export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Restart Claude Code after enabling.

Installation

/plugin marketplace add izzzzzi/izTeam
/plugin install team@izteam

Usage

/build <description or path/to/plan.md> [--coders=N] [--fresh] [--git-checkpoints]
/brief <description> — interview first, then build
/conventions [path/to/project]

Examples:

/build "Add user settings page with profile editing"
/build docs/plan.md --coders=2
/build "Add notifications" --git-checkpoints
/brief "Add notifications"
/conventions

How It Works

/build

A Team Lead orchestrates the full implementation flow. The pipeline scales with task complexity, so simple work stays fast and complex work gets more review.

Repo Map — before dispatching researchers, /build generates a ranked symbol map of the codebase (.repo-map). This gives the team instant structural context: which files exist, what they export, and which are most-imported. Researchers skip structural discovery and focus on semantics, coders get a codebase overview in the briefing. Inspired by Aider's repo map approach.

flowchart TD
    subgraph Phase1["Phase 1: Discovery & Planning"]
        R["Parallel Research<br/>Codebase + Reference researchers"]
        CL{"Complexity<br/>Classification"}
        R --> CL
        CL -->|SIMPLE| S5
        CL -->|MEDIUM / COMPLEX| PV
        PV["Plan Validation<br/>Tech Lead checks scope & architecture"]
        RA["Risk Analysis<br/>Tech Lead + Risk Testers"]
        PV --> RA
    end

    subgraph Phase2["Phase 2: Execution"]
        S5["Coding with Gold Standards<br/>Parallel coders implement"]
        CC["Convention Checks<br/>naming, imports, schema"]
        SR{"Specialized Review"}
        AA["Architectural Approval<br/>Tech Lead sign-off"]
        S5 --> CC --> SR --> AA
        SR -.- U["SIMPLE: 1 Unified Reviewer"]
        SR -.- SP["MEDIUM/COMPLEX: Security + Logic + Quality"]
    end

    subgraph Phase3["Phase 3: Completion"]
        IV["Integration Verification<br/>build + tests"]
        CU["Conventions Update<br/>patterns & review findings"]
        SM["Summary Report"]
        IV --> CU --> SM
    end

    RA --> S5
    AA --> IV
Loading
Level When What changes
SIMPLE 1 layer, no behavior changes, <3 tasks Lightweight team, single reviewer
MEDIUM 2+ layers, modifies existing code, 3+ tasks Full team, specialized reviewers, risk analysis
COMPLEX 3+ layers, touches auth/payments, 5+ tasks Full team + deep analysis and risk testing

/conventions

Analyzes the codebase and creates/updates .conventions/ with:

  • gold-standards/ — short exemplary snippets
  • anti-patterns/ — what to avoid
  • checks/ — naming and import rules

/build uses these conventions as reference examples. You can also run /conventions standalone.

Complexity Levels

Level Team Size Reviewers Risk Analysis Tech Lead Validation
SIMPLE 4 agents 1 unified Skipped Skipped
MEDIUM 5-7 agents 3 specialized Yes Yes
COMPLEX 6-9+ agents 3 specialized + deeper scrutiny Full + risk testers Yes + user informed on key decisions

Team Roles

Role Lifetime Purpose
Lead Whole session Orchestrates delivery and team flow
Supervisor Permanent Monitors liveness, loops, and escalations
Codebase Researcher One-shot Summarizes structure and conventions
Reference Researcher One-shot Provides high-quality reference files
Tech Lead Permanent Validates plans and architecture
Coder Per task Implements and self-checks
Security Reviewer Permanent Finds exploitable vulnerabilities
Logic Reviewer Permanent Finds correctness and edge-case bugs
Quality Reviewer Permanent Improves maintainability and consistency
Unified Reviewer Permanent All-in-one reviewer for SIMPLE tasks
Risk Tester One-shot Verifies explicit risks with targeted checks

Structure

team/
├── .claude-plugin/
│   └── plugin.json
├── skills/
│   ├── build/
│   │   ├── SKILL.md
│   │   ├── scripts/
│   │   │   └── repo-map.py
│   │   └── references/
│   │       ├── complexity-classification.md
│   │       ├── risk-analysis-protocol.md
│   │       ├── state-ownership.md
│   │       ├── state-template.md
│   │       ├── summary-report-template.md
│   │       └── teardown-fsm.md
│   ├── brief/
│   │   ├── SKILL.md
│   │   └── references/
│   │       ├── brief-template.md
│   │       └── interview-principles.md
│   └── conventions/SKILL.md
├── agents/
│   ├── supervisor.md
│   ├── codebase-researcher.md
│   ├── reference-researcher.md
│   ├── tech-lead.md
│   ├── coder.md
│   ├── security-reviewer.md
│   ├── logic-reviewer.md
│   ├── quality-reviewer.md
│   ├── unified-reviewer.md
│   └── risk-tester.md
├── references/
│   ├── gold-standard-template.md
│   ├── reviewer-protocol.md
│   ├── risk-testing-example.md
│   ├── status-icons.md
│   └── supervisor-playbooks.md
├── README.md
└── README.ru.md

License

MIT