Slash commands

Create /commands that start a known task with one line, bound to an agent, skills or a workflow, and shared across the workspace.

On this page

A slash command is a saved prompt with a name that starts with /. Typing /release-notes in a composer runs the same instructions every time, with the same agent, skills or workflow. Commands are workspace-wide and managed in Settings → Skills & commands → Slash commands.

Where commands work

Commands show in the composer’s command menu wherever you can start or steer work:

  • The Home composer and the New run dialog (press C).
  • A run’s conversation, to ask the agent working on it.
  • The agent panel (⌘ J).
  • The command palette (⌘ K), by name.

Type / to open the menu, then keep typing to filter. Each command shows its description, so write one that says what will happen.

Create a command

Select New slash command and fill in:

FieldRules
NameStart with / then lowercase letters, for example /release-notes. Digits and dashes are allowed after the first letter. Must be unique.
DescriptionShown in the composer’s command menu. One line.
PromptWrite what the agent should do when someone runs it. Text typed after the command is passed in as the request.
BindingsOptional. The agent, skills or workflow it uses.
AgentsOptional. Which agents may run it. Empty means every agent.

Then select Create command. It shows “Created /release-notes” and appears in every composer at once.

Bindings

Without bindings, a command uses whichever agent is in front of you and whatever skills it would load anyway. Bindings make it predictable:

BindingEffect
AgentAlways runs with this agent, for example writer.
SkillsLoads these skills for the task, as if mentioned with #name.
WorkflowStarts a run of this factory workflow instead of a one-off conversation. Text after the command becomes the run’s prompt.

A command bound to a workflow is the quickest way to give people a “start the usual process” button: /hotfix payments refunds double-posting starts the Hotfix workflow with that description.

Example: /release-notes

Name:        /release-notes
Description: Draft release notes from PRs merged since the last tag
Agent:       writer
Skills:      #voice-and-tone
Prompt:
  Find the latest release tag in {{factory.repos}}. List every PR merged
  since then. Group them as New, Improved and Fixed. Write one plain
  sentence per change from the user's point of view; skip internal
  refactors. Link each PR. End with "Not verified:" and anything you
  couldn't confirm from the diff. Post the draft here for review.
  Request: the text after the command.

Someone types /release-notes for 4.12, include the billing changes and the writer agent drafts it with the voice-and-tone skill loaded.

Built-in examples on Explore

Explore lists ready-made commands, including /standup, /triage, /spec and /postmortem. Install command adds a copy to your workspace that you can edit. See Install an item.

Edit, duplicate or delete

Each command’s menu has Edit, Duplicate, Publish to Explore… and Delete. Edits apply the next time someone runs the command. Runs already started keep the prompt they started with. Deleting shows “Deleted /name”; workflows and other commands don’t depend on commands, so nothing else breaks.

Commands and skills: which to use

Use a skill whenUse a command when
Agents should apply it whenever a task matchesA person starts the task on purpose
It describes how to do somethingIt describes what to do, now
It applies across many workflowsIt packages one routine request

They work well together: a command names the task and binds the skill that knows how.

API and CLI

Manage commands with the Commands API. Commands can also be published to Explore like skills; see Publishing to Explore.

    Type to search the docs, help center and blog.