Agents Overview

17 specialized agents that handle every aspect of software development—automatically orchestrated through predefined workflows.

Quick Reference

Development & Implementation

Agent	Purpose
planner	Research, analyze, create implementation plans before coding
fullstack-developer	Execute implementation phases with strict file ownership
scout	Parallel file search across large codebases
scout-external	External search using Gemini CLI and OpenCode
debugger	Root cause analysis, log investigation, issue diagnosis
tester	Test execution, coverage analysis, quality validation

Quality & Review

Agent	Purpose
code-reviewer	Security audits, performance analysis, code quality
database-admin	Query optimization, schema design, performance tuning

Documentation & Management

Agent	Purpose
docs-manager	Technical documentation, API docs, architecture guides
project-manager	Progress tracking, cross-agent coordination, status reports
journal-writer	Document failures and setbacks with brutal honesty
git-manager	Conventional commits, security scanning, token-optimized

Creative & Research

Agent	Purpose
ui-ux-designer	Award-winning UI with Three.js, responsive layouts
copywriter	High-converting marketing copy, viral content
brainstormer	Explore approaches, challenge assumptions, debate decisions
researcher	Multi-source research, documentation analysis

Integration

Agent	Purpose
mcp-manager	MCP server integrations, tool discovery, execution

How to Use

Automatic (recommended): Commands orchestrate agents automatically

/cook [feature]     # planner → code → tester → reviewer → git-manager
/fix:hard [bug]     # scout → debugger → planner → code → tester
/plan [task]        # planner + researcher

Explicit: Request specific agents in prompts

"Use scout agent to find auth files, then planner to create migration strategy"

Under the Hood

Orchestration Patterns

Sequential (default): Agents run in order, each building on previous output

planner → code → tester → code-reviewer → git-manager

Parallel: Independent agents run simultaneously

scout (dir1) ┐
scout (dir2) ├─→ Aggregate → planner
scout (dir3) ┘

Hybrid: Mix of sequential and parallel for complex tasks

Agent Communication

Agents share context through:

Shared files: docs/, plans/, code standards
Handoff protocols: Each agent receives previous output, performs task, passes results
TodoWrite: Real-time progress tracking visible to user

Handoff Example

planner output → plans/auth-feature.md
    ↓
code reads plan → implements → creates files + tests
    ↓
tester runs tests → validates coverage
    ↓
code-reviewer audits → security + quality report
    ↓
git-manager commits → conventional commit + push

Troubleshooting

Agent not activating?

Check command matches task complexity (/fix:fast vs /fix:hard)
Verify workflow files exist in .claude/agents/
Try explicit invocation: “Use [agent] to…”

Slow response?

Use parallel orchestration when tasks are independent
Scope tasks more specifically
Use simpler commands for simple tasks

Conflicts?

Review orchestration order in workflow files
Check handoff protocols between agents

Key Takeaway

17 agents work together automatically—use commands to orchestrate them, or invoke explicitly for specific tasks. No manual coordination needed.