claude-plugin-software-engineering-team

claude-plugin-software-engineering-team

Opinionated specialist agents and team workflows for Claude Code. Focused on doing fewer things well rather than covering everything.

Philosophy

Inspired by Trail of Bits' approach: agents encode expertise, not procedures. Each agent knows how to think about its domain, not just what steps to follow. They include reference material (anti-pattern catalogues, checklists, smell guides) that make them expert-level rather than superficial.

Key principles:

• Agents don't overlap — each has a clear domain boundary and explicitly states what it doesn't do
• Severity is standardised — Critical/Warning/Suggestion across all agents, with consistent meaning
• Memory is enabled — agents learn your codebase patterns across sessions via memory: project
• Cost-conscious — review agents use Sonnet (fast, cheap for read-only analysis); only the orchestrator inherits your session model

What's Included

Review Agents

Orchestration

Investigation

Skills (Slash Commands)

Installation

Option 1: Claude Code Plugin (recommended)

# From any Claude Code session:
/plugin install https://github.com/rayh/claude-plugin-software-engineering-team

Option 2: Copy to Global Config

# Clone the repo
git clone https://github.com/rayh/claude-plugin-software-engineering-team.git
cd claude-plugin-software-engineering-team

# Copy agents and skills to your global Claude config
cp -r agents/* ~/.claude/agents/
mkdir -p ~/.claude/skills/review ~/.claude/skills/investigate ~/.claude/skills/spike
cp skills/review/SKILL.md ~/.claude/skills/review/
cp skills/investigate/SKILL.md ~/.claude/skills/investigate/
cp skills/spike/SKILL.md ~/.claude/skills/spike/

Option 3: Copy to a Specific Project

# From your project root
cp -r agents/* .claude/agents/
cp -r skills/* .claude/skills/

Enable Agent Teams

Agent teams are experimental. Add to your ~/.claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Usage

Quick Review (After Making Changes)

/review staged
/review src/api/
/review feature-branch

Smart Review (Let Tech Lead Decide)

Use the tech-lead to review my recent changes

The tech-lead reads the diff, consults its decision matrix, and only spawns relevant specialists. A one-line bug fix gets quality-engineer only. An API change gets all four.

Bug Investigation

/investigate Users report intermittent 500 errors on POST /api/sessions after deploying v2.3

Technical Spike

/spike Should we use Redis, DynamoDB, or PostgreSQL for session storage?

Single Specialist

Use the security-auditor to review the OAuth implementation in src/auth/
Use the sre-architect to evaluate the docker-compose.yml

When to Use What

Team Topology Ideas

Beyond the included agents, these patterns work well with agent teams:

Migration/Refactor Team

• Scout: maps all usages of the thing being changed
• Implementer(s): one per module to avoid file conflicts
• Verifier: runs tests after each batch, reports regressions

Documentation Team

• Reader: traces code flows, builds understanding
• Writer: produces docs from reader's findings
• Reviewer: checks docs against actual code

Incident Response Team

• Timeline builder: reconstructs what happened from logs/metrics
• Root cause analyst: investigates code for the underlying bug
• Remediation planner: proposes fix + prevention measures

Customisation

All agents are markdown files with YAML frontmatter. Edit them to:

• Adjust severity calibration — if an agent flags too many Warnings for things you consider fine
• Add project-specific context — domain knowledge, tech stack details, team conventions
• Change models — upgrade to opus for more thorough reviews, or haiku for quick checks
• Restrict tools — remove Bash if you want purely read-only analysis

Prior Art & Inspiration

This project builds on ideas from:

• trailofbits/claude-code-config — "encode expertise not procedures"
• VoltAgent/awesome-claude-code-subagents — 100+ agent definitions
• contains-studio/agents — departmental agent organisation
• affaan-m/everything-claude-code — battle-tested config collection
• wshobson/agents — plugin-based orchestration patterns
• Addy Osmani's agent teams writeup — narrow scope = better reasoning

License

MIT