developer-toolbox

Name: developer-toolbox
Author: jezweb

Seven specialized agents for code review, debugging, testing, documentation, commits, and build verification. Includes commit-helper, build-verifier, code-reviewer, debugger, test-runner, orchestrator, and documentation-expert agents with auto-discovery triggers Each agent follows a "MUST BE USED when" pattern to trigger automatically based on user requests like "code review", "TypeError", "write tests", or "document" Agents can be chained together for multi-step workflows (e.g., debug → document → commit) Includes an agent-first-thinking rule that encourages using agents by default for tasks like repeated edits, multi-file reads, or parallel audits

INSTALLATION

npx skills add https://github.com/jezweb/claude-skills --skill developer-toolbox

Run in your project or agent environment. Adjust flags if your CLI version differs.

SKILL.md

$2a

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