Prompt Commands

Prompt Commands are reusable slash-triggered prompt presets. They let a workspace turn repeated work such as PR review, ticket planning, release-note drafting, and HTML prototyping into durable commands that accept a short brief at runtime.

What Prompt Commands Are

A Prompt Command is a stored set of instructions that runs through the normal Dawn agent conversation path. The user invokes it with a slash command, then adds the specific brief for that run.

/pr-review https://github.com/example/app/pull/248
/ticket-plan ADO-1842
/prototype redesign the billing settings page using the existing admin style
/release-notes changes merged since last Friday

Everything after the command slug becomes the raw brief. The command body supplies the durable workflow instructions, so users do not have to repeat the same long prompt every time.

Prompt Commands vs Product Slash Commands

Product slash commands

Commands such as /help, /whats_new, and /link are built-in product actions. They are deterministic and do not depend on a model run.

Prompt Commands

Commands such as /prototype, /pr-review, and /ticket-plan are stored prompt workflows. They start a normal Dawn run with command instructions attached.

Use Slash Commands when you need help with built-in product commands. Use this page when you want to create, customize, or invoke reusable prompt workflows.

What A Command Contains

• Slug: the slash entry point, for example pr-review or prototype.
• Name: the readable title shown in the command browser and command suggestions.
• Description: the short explanation users see when deciding which command to run.
• Brief hint: the usage hint shown after the command, for example PR URL, branch, or review focus.
• Tags: labels that make commands easier to filter and organize.
• Command body: markdown instructions that define the reusable workflow.
• Status: enabled commands can run; disabled commands stay visible but are not available as runnable presets.

Global And Workspace Commands

Global commands

Global commands are Dawn-provided defaults. They are read-only in the product UI and give every workspace a useful starting command pack.

Workspace commands

Workspace commands belong to one workspace. Managers and owners can create them from scratch or copy a global command into an editable workspace version.

If a workspace command uses the same slug as a global command, the workspace command overrides the global one for that workspace. This lets teams keep Dawn defaults while customizing the workflows that matter locally.

Current Global Command Examples

• /prototype creates or revises a grounded HTML artifact from a short brief.
• /pr-review reviews a pull request or diff for correctness, regressions, and missing tests.
• /pr-summary summarizes what changed, why it changed, risky areas, and files touched.
• /ticket-plan turns a ticket, issue, work item, or card into an implementation plan grounded in connected code and docs.
• /ticket-triage classifies priority, severity, owner, missing information, and next action.
• /release-notes drafts user-facing release notes from merged PRs, tickets, and connected project context.
• /project-brief summarizes project state across tickets, docs, PRs, design files, and memory.
• /reflect audits recent workspace context and suggests memory updates, skills, commands, or workflows.

The exact command list can change as Dawn ships new global defaults. Type / in a supported chat surface or use /help to see what is available in the current workspace.

For per-command syntax and behavior, see the Command Reference.

How To Create Or Customize A Command

1. Open Dawn Web.
2. Go to Commands.
3. Use Add a Custom Command to create a workspace command, or open a global command and choose Copy to workspace.
4. Set the slug, name, description, brief hint, and tags using the form fields.
5. Write the durable workflow instructions in the command body markdown.
6. Save the command and run it from a supported Dawn chat surface with /command-slug your brief.

Users do not need to write frontmatter manually in the editor. The command details are managed by the UI; the markdown editor is for the command body.

Agent-Assisted Command Authoring

Dawn agents can help create, copy, and update workspace Prompt Commands when catalog authoring tools are available. This lets a good one-off prompt become a reusable command without manually rebuilding it in the command editor.

Useful requests include:

Create a workspace prompt command called /security-review from the review process we used in this thread.

Copy the global /pr-review command into this workspace and make it stricter about pre-existing issues.

Update /release-notes so it always presents user-facing changes before internal changes.

The agent can draft the command body, choose a slug and description, and update the workspace command record. You should still review high-use commands before telling the team to rely on them.

Auditability And Stale Writes

Command authoring is auditable. When an agent creates or updates a command, Dawn records the tool event so the change can be traced back to the conversation that requested it.

Dawn also protects shared command content from stale edits. If a command changed after the agent last read it, Dawn should not silently overwrite the newer version. Reload the latest command and ask Dawn to apply the update again.

Runtime Behavior

• The transcript keeps the literal slash invocation, so people can see exactly what was typed.
• Dawn passes the command body and raw brief into the run so the agent follows the stored workflow.
• The selected agent still controls available skills, memory, connected sources, tools, and local capabilities.
• A command shapes the task. It does not grant extra permissions or broaden source access.
• If the command has no brief, Dawn still passes the command body, but the command should be written to ask for missing required input when needed.

Revisions

Commands are durable workspace assets. Every save creates a revision, so teams can inspect previous versions and restore an older version when a command changes in the wrong direction.

• Use revisions to review how command behavior changed over time.
• Restore a prior revision instead of manually reconstructing an older prompt.
• Treat edits to high-use commands like changes to shared workspace process, not throwaway prompt tweaks.

Writing Good Prompt Commands

• Use a command for repeated workflows, not one-off requests.
• Keep each command focused on one kind of outcome.
• Tell Dawn what context to inspect first and what output shape to produce.
• Call out when the command should ask for clarification instead of guessing.
• Use the brief hint to teach users what to put after the command slug.
• Copy and customize global commands when they are close to the workflow you need.

Troubleshooting

• A command does not appear in autocomplete: confirm it is enabled and available in the selected workspace, then type / at the start of the composer.
• A copied global command still looks unchanged: confirm you are editing the workspace copy, not the read-only global default.
• The wrong version runs: check whether a workspace command with the same slug is overriding a global command.
• The agent cannot access a source the command mentions: assign the source, skill, or memory access to the agent. Commands do not grant extra access by themselves.
• The command output drifted: inspect revisions and restore the last known-good command body.

List the effective prompt commands available in this workspace and explain which are global versus workspace overrides.

Suggest three useful custom prompt commands for this workspace based on connected sources and repeated workflows.

Turn the repeated workflow from this thread into a workspace prompt command draft.