Build an agentic system around your work
Move beyond one-off prompts. Combine project context, specialized agents, repeatable workflows, connected tools, and automated checks into a system that helps agents work consistently across tasks, projects, and teams.
Download
Design the system behind your agents
Give agents the context, capabilities, and controls they need to take on real work reliably.
Apply team practices automatically
Use instructions to give agents your coding standards, architecture decisions, and file-specific guidance without repeating them in every prompt.
Optimize AI credit usage
Focused instructions and reusable workflows can reduce corrective turns, repeated context, and discarded work so more of your usage moves the task forward.
Make recurring work repeatable
Package proven procedures in agent skills so agents can follow the same steps on demand instead of you describing the process again.
Extend and control agent capabilities
Give agents specialized roles with custom agents, connect external systems through MCP, and use hooks when an action must run consistently.
Build a setup around your workflow
Customization types work together as layers. Instructions supply project standards, a custom agent defines a focused role and tool set, and a skill captures the steps for recurring work.
Add an MCP server when the workflow needs an external system, and use hooks for checks that must run every time. Keep the setup in your repository so the whole team starts from the same context.
Learn how customizations combine
Scale a proven setup across teams
Once a workflow works well, bundle its instructions, skills, custom agents, hooks, and MCP servers as a plugin. Teams can install the complete setup as one package, version it as practices evolve, and reuse it across projects.
Share customizations with plugins
Measure, refine, and get more from each turn
Agent Debug Logs visualize model turns, token counts, tool calls, and subagent handoffs. Trace where an agent repeated work, missed context, or skipped a useful tool, then refine the customization and compare the next run.
This feedback loop helps you keep instructions focused, make repeatable work reusable, and identify avoidable turns that consume AI credits without moving the task forward.
Explore Agent Debug Logs
Choose the right customization
Use guidance for context, workflows for repeatable work, tools for new capabilities, and hooks for actions that must run consistently.
Instructions
Apply project standards automatically or target guidance to specific files and tasks.
Agent skills
Teach agents task-specific workflows with supporting scripts, templates, and resources.
Custom agents
Configure a specialized role with its own instructions, tools, and language model.
MCP servers
Connect agents to external tools and data sources through an open protocol.
Hooks
Run deterministic checks and automation at defined points in the agent lifecycle.
Start with the context you repeat most
Add an instructions file for the guidance you give every day. As new patterns emerge, build on it with a reusable workflow, specialized role, external capability, or reliable check.