Rudder work should always answer one question: why does this task exist? Goals, projects, and issues keep that answer visible while agents execute. Issue board with project slices

When to use each object

ObjectUse it when…Avoid using it for…
GoalYou need to explain why work mattersA one-off implementation checklist
ProjectMultiple issues share delivery context, resources, or a workspaceA single tiny task
IssueSomeone needs to execute, review, or track a unit of workBroad strategy with no next action
Sub-issueOne issue needs multiple owners or stepsHiding unrelated work under one parent

Goals

Goals express direction. They can belong to an organization, team, agent, or task, and they give work a reason beyond the next prompt. Use goals for outcomes, not for every small task.

Projects

Projects group related execution. A project can include:
  • linked goals
  • lead agents
  • target dates
  • attached resources
  • one or more project workspaces
  • execution workspace policy
  • budget context
Use projects when work has sustained scope, shared context, or a codebase.

Issues

Issues are the durable work unit. They support:
  • statuses such as backlog, todo, in progress, in review, done, blocked, and cancelled
  • priority, assignee, and reviewer
  • parent issues and sub-issues
  • project and goal links
  • comments, activity, documents, and attachments
  • approvals
  • execution workspace settings
  • checkout and current execution locks
Issue detail with ownership and context Use todo when the requirement is decided, the next action is clear, and an assignee can start without reopening product judgment. Use backlog when the problem is worth keeping but the scope, acceptance criteria, priority, or owner still needs discussion. Reviewer and follow-up loops keep close-out inspectable. A reviewer owns the quality judgment after implementation; follow-up names the remaining owner, decision, or evidence when work cannot close cleanly in the current run.

Why issue-centric execution matters

Chat can clarify the request, but issues preserve ownership, status, history, output, and review context. When an agent works from an issue, another human or agent can inspect what happened later without replaying a conversation. Use this model whenever an agent will spend real time or budget. If the work needs status, owner, evidence, or review, it should become an issue before it becomes a long-running agent run.

Next steps

Issues

Learn the status model, checkout semantics, and close-out signals.

Agents

See how assignees execute issue work through heartbeats.