Give the work a role.
Choose a developer, tester, reviewer, planner, or orchestrator. Each definition describes its mission, inputs, outputs, constraints, and checks.
Explore the rolesAn open-source agent playbook
Give your AI collaborators a shared way to work. Reusable roles, practical prompts, and explicit standards for building software with more clarity.
MIT licensedDocs & workflow helpersMade to adapt
A contract, before a conversation.
01 / A little structure goes a long way
When every task starts from a blank prompt, expectations drift. Agents gives developers and teams a common starting point for the work, the boundaries, and what “done” means.
Choose a developer, tester, reviewer, planner, or orchestrator. Each definition describes its mission, inputs, outputs, constraints, and checks.
Explore the rolesUse blank and filled templates, worked input/output examples, and standalone role guides. Adapt them to your project and your AI tool.
Open a templateShared policies cover local or Linear tracking, structured handoffs, and worktree behavior. Reusable skills add helpers for checks, testing, and coordination.
Browse the skills02 / Meet your next collaborator
A few places to start. These are instructions for your AI tool; choose a role, read its contract, and bring your own project context.
Showing all 6 featured roles.
Create a new role package with explicit boundaries, useful examples, and a visible quality rubric.
agent-making-agentWork across frontend and backend with task boundaries, validation checks, and a structured handoff.
fullstack-developerUse a read-only review workflow to surface prioritized findings and actionable remediation tasks.
deep-code-review-agentVerify UI changes, address coverage gaps, and report results with evidence and clear next steps.
frontend-testerTurn a product goal into a documented plan with atomic tasks and explicit acceptance criteria.
product-managerCoordinate the developer, tester, and reviewer lifecycle using shared tracking and handoff contracts.
sprint-orchestrator-agent03 / From intent to instructions
Read its mission and boundaries. Check which skills or connected tools it needs.
Adapt the usage template with your goal, repository, constraints, and acceptance criteria.
Give the prompt to your AI tool. Review its output against the role’s checks before continuing.
Agent: agent-making-agent
Goal: Create an agent that drafts
release notes from merged pull requests.
Inputs: Use the Agent Making README
as the standards source. Group release
notes by feature area.
Constraints: Documentation only.
No application code edits.
Output: README.md, USAGE_TEMPLATE.md,
EXAMPLES.md, and USER_GUIDE.html.Built in the open
This is a living toolkit. The role packages and helpers are available now; the broader standardization work is ongoing.
04 / Make it your own
Clone the toolkit, pick a role, and tell your AI tool where to find it. Start with one useful workflow.
Full setup instructionsgit clone https://github.com/slobodanmargetic988/Agents.git02Read a role. Adapt its template.Start with Agent Making Agent.03Add tools when your role needs them.Skill installation and MCP setup are separate.Role definitions work as local instructions. Workflow helpers may need additional tools, configuration, and permissions. Review each role before running it.
MCP setup guide