Aliases in AGENTS.md: Short Commands for an AI Agent

I picked up a simple idea from an AI engineer's YouTube video: add short aliases for recurring instructions to your global prompt. I tried it in my own setup. It turned out to be handy.

"Explain it more simply," "shorten your answer," "just explain, don't change anything." You can type these requests in full every time. Or give them short names once and use those when talking to the agent.

I keep the definitions in my global AGENTS.md. Claude Code uses CLAUDE.md for these instructions. The agent needs to load the file into its context. A Markdown file sitting on disk won't change its behavior on its own.

My Five Aliases#

Here's a block you can add to your instructions. In this version, an alias works both as a standalone message and within a longer request:

## Aliases

When I use an alias as an explicit instruction, expand it and act as if
its full definition were given to you directly. This applies both to
standalone messages and to aliases within a longer request.
Do not expand aliases in quotes, code, or text that merely discusses them.

scr = `Simplify, compress, and repeat your response`
e18 = `Explain this like I'm 18. Simplify your language. Shorten your response`
foc = `Focus on what matters most here. Whats the true signal? Whats the true value? Boil your response down into the most important thing we need to focus on`
exp = `Explain only. You may search and inspect, but do not change anything`
qst = `If anything in my request is unclear, ask clarifying questions before proceeding. Don’t make assumptions`

scr asks the agent to shorten and rephrase its answer. e18 asks for a simpler explanation, as if it were talking to an eighteen-year-old. foc asks it to focus on what matters most. With exp, it can inspect code and look up information, but it shouldn't change anything. qst asks it to clarify anything unclear first instead of making assumptions for me.

For example, a request might look like this:

Explain how this code works. e18

Or, if the agent has already given a complicated answer, I can send just e18 in the next message to ask for the same explanation in simpler, shorter terms.

There's a caveat with exp: it's an instruction to the model, not a tool-level restriction on writing. If you need to prevent changes reliably, enforce that through access permissions or a sandbox.