Developer Toolbox
A collection of essential development workflow agents that integrate seamlessly with Claude Code.
What's Included
Agents (7)
Agent Purpose Triggers On
commit-helper Generate conventional commit messages "commit message", "staged changes"
build-verifier Verify dist/ matches source after builds "changes not appearing", "verify build"
code-reviewer Security audits and code quality reviews "code review", "security audit", "OWASP"
debugger Systematic debugging with root cause analysis "error", "TypeError", "stack trace", "bug"
test-runner TDD workflow and test creation "write tests", "TDD", "coverage", "jest"
orchestrator Coordinate complex multi-step projects "coordinate", "multi-step", "complex feature"
documentation-expert README, API docs, architecture diagrams "document", "README", "API docs"
Rules (1)
Rule Purpose
agent-first-thinking Behavioral interrupt - consider agents before manual work
Installation
Via marketplace
/plugin install developer-toolbox
Or local development
/plugin install ./skills/developer-toolbox
After installation, restart Claude Code to load the agents.
Usage Examples
Commit Helper
"Help me write a commit message for these staged changes"
Build Verifier
"My changes aren't appearing in production, verify the build output"
Code Reviewer
"Review this authentication code for security vulnerabilities"
Debugger
"I'm getting TypeError: Cannot read property 'map' of undefined"
Test Runner
"Use TDD to implement this user validation function"
Orchestrator
"Coordinate a refactor of the authentication system across 5 services"
Documentation Expert
"Create comprehensive API documentation for this REST endpoint"
Agent Design Philosophy
All agents follow the "MUST BE USED when" pattern for reliable auto-discovery:
description: | [Role] specialist. MUST BE USED when: [trigger 1], [trigger 2], [trigger 3]. Use PROACTIVELY for [broad task category].
Keywords: keyword1, keyword2, error-message-fragment
This ensures Claude Code discovers and proposes the right agent automatically based on user requests.
Agent-First Thinking Rule
The included agent-first-thinking.md rule encourages using agents by default:
The Inversion:
-
Wrong: "I'll do this manually unless it's big enough for agents"
-
Right: "I'll use agents unless there's a reason not to"
Triggers:
If about to... Use instead...
grep/glob 3+ times Explore agent
Read 5+ files Explore agent
Same edit across files Parallel agents
Audit multiple items Parallel swarm
Customization
Each agent can be extended by editing its markdown file after installation:
Find installed agents
ls ~/.claude/plugins/cache//developer-toolbox//agents/
Or copy to user-level for customization
cp [plugin-path]/agents/code-reviewer.md ~/.claude/agents/
Combining Agents
Agents work well together:
"Review this code for security issues, then write tests for the critical paths"
→ code-reviewer first, then test-runner
"Debug this failing test, document the root cause, and commit the fix"
→ debugger → documentation-expert → commit-helper
Version History
- 1.0.0 (2025-01-20): Initial release with 7 agents and 1 rule