Send to Agents
Select Send to... on an initiative, bug, or todo to hand it off. The menu lists your published workflows, your agents, and the coding tools you enabled.
Pick a destination
Workflows
Sends the item into a published workflow: a repeatable sequence of agent steps and optional human review gates, such as plan, build, and review. Only workflows that accept this work type appear. To set one up, follow Run your first workflow.
Agents and cloud agents
Assigns the item to a single agent as a one-step run. Choose this when you do not need a multi-step workflow. Progress appears in Monitoring and stays linked to the item.
Coding tools
Opens a coding tool, such as Cursor or Claude Code, with a prompt prefilled from the item. Only tools enabled in Preferences appear here. Terminal launches are covered in Use in Terminal.
Choose a mode
The mode tells the agent or tool what to do next:
| Mode | Initiatives and todos | Bugs |
|---|---|---|
| Plan | Break it down first | Figure out how to fix it |
| Build / Fix | Start building | Fix the bug |
| Review | See what is left to do | Check whether it is fixed |
| Research | Investigate before deciding | Investigate the open question |
Completed items only offer Review and Research.
If the item is Open, Planned, Blocked, or an Idea, sending it can also move it to In Progress.
What gets sent
The prompt is built from the item's title, description, linked documents, comments, and details. In spec-driven development, this item is the spec the agent works from, so add missing context before you send it.
What sent content cannot change
The agent reads the item's text, comments, and documents as context, not as instructions it must obey. Nothing in that content can change the agent's sandbox, credentials, allowed commands, network access, or permitted side effects. Those are set by the agent's configuration.