Custom commands and agents

Custom commands

A command is a prompt you write once and call by name. It is a Markdown file:

---
description: Review a file for correctness bugs
agent: reader          # optional: run as this custom agent
mode: ask              # optional: ask, governed or direct
argument-hint: <file>
---
Review $ARGUMENTS for correctness bugs, not style. Start with $1.

Saved as .albertcode/commands/review.md in a repository, it is /review:

your-project · gpt-4o · governed ❯ /review src/cart.py

• $ARGUMENTS is everything typed after the command; $1 to $9 are its words, quoted as a shell would. With neither in the text, what you typed is added at the end.
• When two places define the same name, the repository wins. A command can never take a built-in's name: /plan is always /plan.
• A command is text. Running one runs no program and reads no file: what it expands to is sent as your own message, and the mode and the approvals apply as if you had typed it.

Custom commands appear in the / list and in /help.

Custom agents

An agent is a role a task works as. It can only narrow what mini-AlbertCode may do, never widen it. Keep agents in .albertcode/agents/ in a repository, or in agents/ in your settings folder. Choose one with /agent (or ctrl+x a) in the terminal, or the agent chip in the browser; Albert goes back to the default. A custom command's agent: line uses one for that message only.

• With no tools line, an agent changes only the instructions.
• An agent that cannot edit answers questions only: a change asked of it is refused before anything starts, with the reason.
• The agent is fixed when a task starts.