Send to app
Send one initiative, bug, or todo to a coding tool with the work context already included.
This is the beginner path for agent workflows. It does not require building a worker or writing API code.
Before you send work
The initiative, bug, or todo should have a clear title and enough description for the next step. In spec-driven development, this item is the spec the target tool receives. Add taxonomy, comments, or a document-backed description when those details matter to the implementation.
Configure launch targets
Open Preferences and select the AI coding tools or IDEs you use. Send to app only shows launch targets you have selected, so the menu stays focused on the tools you actually use.
Enabled tools appear in the Send to menu. The launch path depends on the tool: app deep link, terminal command, MCP context, or queued worker execution.
For Terminal launches, add project folders in Preferences -> Terminal folders. If no folder is saved, Desktop asks you to pick one during launch and saves it for next time. Terminal launch uses the one:// deep link scheme and is macOS-only from Desktop.
Send one item
- Open an initiative, bug, or todo.
- Choose Send to.
- Pick the target app, terminal flow, or local agent option.
- Choose the mode that matches the work.
- Review the generated handoff before the tool starts.
The handoff uses the work item and linked context.
Modes
| Mode | Use it for |
|---|---|
| Plan | Scope unclear work before implementation starts. |
| Code | Start implementation when the path is clear. |
| Review | Inspect completed work, a pull request, or agent output. |
| Fix | Address a specific failure or review issue when the target supports it. |
When the work item is Open, Planned, Blocked, or Idea, the launch flow can move it to In Progress in the same action. Completed work supports review-style handoff.
App, terminal, or worker
A direct app launch opens the coding tool now.
Terminal opens the work in a saved local folder and runs a configured command such as Claude Code or OpenCode. Terminal launch is macOS-only from Desktop.
A local worker queues work for a running Codex worker instead of opening it in a foreground app. The worker creates an Agent Session, claims it, reports progress, and returns the result to the original initiative, bug, or todo.
What to check after launch
The coding tool or worker should produce evidence that belongs back on the work: a comment, plan, pull request, external URL, blocker, or completion note.
Treat agent output as delivery evidence, not automatic approval. Review the code, tests, pull request, or linked result before marking the underlying work done.
For the Deliver Work concept page, see AI Handoff. For the CLI equivalent, see Use in Terminal.