---
title: "Choosing an Agent"
description: "Interactive selector and complete reference for picking the right agent for any task."
canonical: "https://orchestkit.yonyon.ai/docs/agents/choosing-an-agent"
---

# Choosing an Agent

Interactive selector and complete reference for picking the right agent for any task.

import { LazyAgentSelector as AgentSelector } from "@/components/lazy";
import { Callout } from "fumadocs-ui/components/callout";

# Which Agent For Your Task?

OrchestKit's 36 agents cover the full software development lifecycle. Use the interactive selector below to find the right specialist — filter by category, task type, or use the guided quiz.

<Callout title="Quick Start: Top 5 Agents" type="info">
  **Most common picks:**
  `backend-system-architect` (APIs), `frontend-ui-developer` (React),
  `security-auditor` (OWASP), `workflow-architect` (LangGraph),
  `debug-investigator` (bugs). For full-stack features, use `/ork:implement`.
</Callout>

<AgentSelector />

---

## Common Scenarios

### "I need to build a new feature end-to-end"

Use `/ork:implement`. It spawns up to 14 agents across design, implementation, and validation phases. You choose the scope (full-stack, backend-only, frontend-only, or quick prototype).

### "Someone submitted a PR, I need to review it"

Use `/ork:review-pr 123`. It spawns 6 parallel reviewers (code quality, type safety, security, tests, backend, frontend) and synthesizes their findings into a single structured report.

### "The app is broken and I do not know why"

Spawn `debug-investigator`. It follows a scientific method: reproduce, gather evidence, form hypotheses, test them, isolate the root cause, and recommend a fix -- without modifying any code.

### "I need to set up CI/CD and deploy to production"

Start with `ci-cd-engineer` for pipeline setup, then `deployment-manager` for the release. The `release-engineer` handles versioning and changelogs.

### "I want to design an architecture before coding"

Use `system-design-reviewer` to evaluate your plan against five dimensions (Scale, Data, Security, UX, Coherence). It returns a structured review with APPROVE / REQUEST CHANGES / REJECT verdicts.

### "I need to audit security before a release"

Run `security-auditor` for OWASP Top 10 and dependency scanning. Add `ai-safety-auditor` if the codebase includes LLM integrations. Use `security-layer-auditor` for defense-in-depth verification.

## Read-Only vs Read-Write Agents

Some agents are intentionally restricted to read-only operations using agent-scoped hooks that block Write and Edit tools:

| Read-Only Agents | Why |
|-----------------|-----|
| `security-auditor` | Should report vulnerabilities, not fix them |
| `debug-investigator` | Should diagnose problems, not change code |
| `code-quality-reviewer` | Should review code, not rewrite it |
| `system-design-reviewer` | Should evaluate plans, not implement them |

All other agents have full read-write access to the tools listed in their frontmatter.

## What's Next

- [Multi-Agent Patterns](/docs/agents/multi-agent-patterns) -- How agents work together in parallel
- [Writing Agents](/docs/agents/writing-agents) -- Create your own custom agent