Specialized Agents for Every Role

Each agent is purpose-built for a specific part of the development lifecycle. They collaborate, hand off work, and keep your team moving.

@planner
Plan
→
@builder
Orchestrate
→
@developer
Build
→
@critic
Review
→
@qa
Validate
🤖

@developer

Implements features, fixes bugs, writes code

How it works

You describe what you need — @developer analyzes the codebase, plans the approach, writes the implementation, and runs tests. It commits clean, reviewed code that follows your project's conventions.

Key capabilities

Multi-file changes Test generation Build verification Convention adherence Context-aware refactoring
📋

@planner

Breaks down features into structured tasks and stories

How it works

Describe a feature in plain language. @planner reads your PRDs, understands the codebase architecture, and generates implementation tasks with acceptance criteria. Interactive Q&A lets you refine scope before committing.

Key capabilities

PRD-to-task decomposition Story assignment Semantic task grouping Estimation Interactive walkthrough
🔍

@critic

Reviews code for quality, correctness, and consistency

How it works

After @developer writes code, @critic reviews it — checking for bugs, style violations, performance issues, and architectural drift. It catches what automated linting misses.

Key capabilities

Multi-dimensional review Security analysis Performance checks Architecture guardrails Actionable feedback
✅

@qa

Validates completed work through testing sessions

How it works

When a task reaches "ready for test", @qa launches a testing session — starting the dev environment, running test suites, and performing exploratory testing. Pass triggers auto-merge. Fail delegates back to @developer with detailed reproduction steps.

Key capabilities

Automated test execution Exploratory testing Pass/fail workflow Fix delegation Regression detection
🏗️

@builder

Orchestrates the full build pipeline

How it works

@builder coordinates the other agents — delegating implementation to @developer, review to @critic, testing to @qa. It manages the story processing pipeline, tracks progress, handles failures, and ensures every change is verified before commit.

Key capabilities

Pipeline orchestration Failure recovery Session management Progress tracking Multi-story execution

How Agents Collaborate

A coordinated handoff pattern keeps work flowing through the pipeline.

@planner → creates tasks → @builder → delegates to → @developer → @critic → @qa → merged ✓

Each agent has a focused responsibility and clear handoff points. @planner breaks down the work, @builder coordinates execution, @developer writes the code, @critic catches issues early, and @qa validates before merge. When something fails, it flows back through the pipeline with context intact — no lost work, no repeated explanations.

Model Flexibility

Each agent can use different AI models. Bring your own keys, set org defaults, and override per-agent.

🔑

BYOK

Anthropic, OpenAI, Google, and more — connect your preferred providers.

🏢

Org Defaults

Set organization-wide model preferences that apply to all members.

👤

Personal Overrides

Individual users can override org defaults with their own preferences.

🎛️

Model Picker

Org → Provider → Model hierarchy for precise control.