Prompts / Slash commands

Prompts store message text under a slash command such as /review. Create them in organization or project settings, then select them in a thread composer.

Create a prompt

1. Open Prompts in project settings for a project-specific command, or in organization settings to share it across the organization's projects.
2. Create a prompt and enter its Command key, such as review, without the leading slash.
3. Add an optional description and category, then write the Content.
4. Leave Auto expand and Autosubmit off for your first prompt, then save.

A review prompt could contain:

Review the changes for correctness and regressions.
Read the relevant specification and inspect the test coverage.
Report actionable findings with a file location and a concrete failure case.
If there are no findings, say so and identify any verification gaps.

Command keys contain lowercase letters, numbers, hyphens, or colons and must be 2 to 40 characters long. Keys must be unique within their project or organization.

Organization owners and admins manage organization prompts. Project admins and members manage project prompts. Other members who can view the owning organization or project can read its prompts.

Insert a prompt

Type / in the thread composer, find the command, and select it with a click, Enter, or Tab. The menu shows scope badges when both project and organization prompts are available. Add task-specific details around the selected command, attach the task if needed, then send.

With Auto expand off, the composer keeps a compact command reference. Convoy expands that selected reference into the stored prompt content for the agent when you send; the visible message retains its short command label. Typing literal /review text without selecting a command reference does not perform that expansion.

Project prompts take precedence when resolving a bare command key. To choose a scope explicitly, use /project:review or /org:review in the menu. A command you selected keeps that identity even when another scope has the same key.

Choose expansion and submission behavior

Both settings default to off. Autosubmit activates only when you select a saved prompt from the menu. Merely typing an exact command key never submits. Existing text, another command reference, or a file mention prevents autosubmit, so you can combine a prompt with your own request. An attached task and ready attachments can still accompany an automatic submission. Autosubmit also respects the normal Send button's disabled state.

Edit a saved prompt from its settings row. A selected compact reference uses the prompt content resolved when you send; expanded text is already part of your draft and can be edited there. Deleting a prompt requires confirmation.

Prompts, Skills, and runtime commands

The / menu also contains other items with different behavior:

• Skills tell the agent to load an instruction bundle through MCP. They do not paste the whole bundle into the message or autosubmit.
• Runtime commands come from the selected AI runtime. For example, /compact manages its conversation context; it is not a saved Convoy prompt. Availability and behavior depend on the runtime. Selecting /compact can submit immediately when the composer is otherwise empty.

Use a Project agent when the instructions should apply throughout a new thread. Use a prompt when you want to insert a reusable request into a particular message.