Agent Usage Rules
When to Apply
- Multi-step tasks requiring specialized expertise
- Research, analysis, or validation tasks
- Tasks matching agent descriptions
Available Agents
| Category | Agent | Purpose |
|---|---|---|
| Research | codebase-explorer | Map project structure |
| frontend-researcher | UI/React patterns | |
| backend-researcher | API/database analysis | |
| module-researcher | Deep-dive on specific modules | |
| pattern-researcher | Identify existing conventions | |
| dependency-researcher | Package analysis | |
| ui-researcher | Figma design extraction | |
| Questioning | requirement-analyst | Clarify requirements |
| Planning | solution-architect | Architecture design |
| task-planner | Break down into tasks | |
| test-spec-generator | Generate test specs | |
| Validation | code-reviewer | Review code quality |
| test-automator | Write and run tests | |
| security-auditor | Security scanning | |
| documentation-writer | Update docs | |
| Refactoring | refactor-cleaner | Dead code cleanup, consolidation |
| Documentation | project-scanner | Generate project docs |
| Implementation | full-stack-developer | Find code locations |
| Utility | review-generator | Create review summaries |
| summarize-agent | Compress content |
Requirements
- Use agents immediately when task matches description
- No prompt confirmation needed - just invoke
- Launch independent agents in parallel
- Use
subagent_typeparameter correctly
Parallel Execution
Task(codebase-explorer, "...") | Task(pattern-researcher, "...")
Launch multiple agents in single message for independent tasks.
Multi-Perspective Analysis
For complex decisions, get multiple viewpoints:
Task(frontend-researcher) + Task(backend-researcher) + Task(security-auditor)
References
.claude/agents/- Agent definitions- Task tool documentation