Skip to main content
The agent starts every conversation knowing only what’s in your files. Mirror gives you three ways to tell it more:
  • Instructions are standing guidance that the agent always reads, such as how to build and test the project or which conventions to follow.
  • Skills are procedures that the agent loads only when a task needs them, such as how to review a change or cut a release.
  • Memory is an optional, private record of how you work. The agent builds it up across sessions.

​
Instructions

Put project guidance in an AGENTS.md file. Mirror reads it before the agent starts, so it’s the place for anything you’d otherwise repeat in every prompt:
Mirror reads instruction files from:
  1. ~/.mirror/AGENTS.md, for guidance that applies to all your projects.
  2. Every directory from the repository root down to the directory where the workspace runs. A file in a subdirectory adds to the root’s file.
Outside a repository, Mirror reads the global file and the current directory only. Mirror also reads CLAUDE.md, so a project that already has one works without changes. If a directory has both, Mirror uses AGENTS.md. Keep instructions short and specific. Mirror reads at most 32 KiB of instructions in total and truncates the rest. To see which files were read and how much of the context window they use, run /context.

​
Skills

A skill is a directory containing a SKILL.md file. The file starts with a name and a description, followed by instructions for the task:
The agent sees each skill’s name and description, and reads the full instructions when a task matches the description. Unlike instructions, skills cost almost nothing until they’re used, so you can keep many of them.

​
Where to put skills

Mirror also finds skills in subdirectories of these locations, up to six levels deep. If two skills have the same name, the one closer to the working directory wins.

​
Use a skill

The agent picks a skill on its own when your request matches its description, so a clear description matters most. To ask for one explicitly, type $ to browse skills, or mention one by name anywhere in your prompt:

​
Skill metadata

​
Memory

Mirror can keep a private memory of how you work, including your preferences, facts about your repositories, and approaches that worked or failed. Memory is off until you turn it on, and it stays on your machine. Mirror stores memories in ~/.mirror/memory. Memories the agent should always see are kept as bullets in ~/.mirror/PINNED_MEMORIES.md, which you can also edit by hand.