Customizing Commands

Customizing Commands

Use this as the command-specific entrypoint for local mod slash commands. For broader mod work, recipes live in ../creating-mods/references/commands.md, ../creating-mods/references/architecture.md, ../creating-mods/references/ui.md, and ../creating-mods/references/plan-mode.md.

Mod files live in:

textCopy code
~/.letta/mods/

Use a focused file name, e.g. ~/.letta/mods/review.ts or ~/.letta/mods/commands.ts.

First decide whether a command is right

If the command is a reusable workflow like /goal, put the workflow instructions in a skill and keep the mod command as a small launcher/prompt.

Workflow

1. Inspect ~/.letta/mods/ for related command files.
2. Preserve unrelated mod code; create a focused new file if merging is messy.
3. Register with letta.commands.register() and guard with letta.capabilities.commands.
4. Return the unregister function, or a disposer that calls it plus any timer/panel cleanup.
5. Tell the user the exact file path changed and to run /reload.

Default prompt command

tsCopy code
export default function activate(letta) {
  if (!letta.capabilities.commands) return;

  return letta.commands.register({
    id: "review",
    description: "Review current git changes",
    args: "[focus]",
    run(ctx) {
      const focus = ctx.args.trim();
      return {
        type: "prompt",
        content: focus
          ? `Review current git changes. Focus on ${focus}.`
          : "Review current git changes. Focus on correctness issues.",
        systemReminder: true,
      };
    },
  });
}

Command result types

tsCopy code
type ModCommandResult =
  | { type: "prompt"; content: string; systemReminder?: boolean }
  | { type: "output"; output: string; success?: boolean }
  | { type: "handled" };

• prompt: sends content to the agent. Use for normal slash shortcuts.
• output: prints local text and does not contact the agent.
• handled: command handled its own side effects/UI; common for panel commands.

Rules

• Command IDs omit the slash: id: "review", not "/review".
• Use lowercase slugs with letters, numbers, and hyphens.
• Do not register built-in command IDs.
• runWhenBusy: true commands must not return prompt while the main agent is busy; use scoped conversation helpers/panels and return handled.
• showInTranscript: false commands should usually return handled, not prompt.
• Do not import Letta Code app internals.
• Do not do surprising side effects on startup; mods activate on app start and /reload.

More recipes

• Simple output command, panel command, busy-safe conversation command: ../creating-mods/references/commands.md
• Complex command architecture, state, cleanup: ../creating-mods/references/architecture.md
• Panel/status UI patterns: ../creating-mods/references/ui.md
• Worked plan-mode command/tool composition: ../creating-mods/references/plan-mode.md