01
One model, every agent
Adapters for Claude Code, Codex, and Cursor normalize instruction files, MCP servers, and installed skills into a single workspace model.
Claude Code, Codex, and Cursor each keep their own instructions, MCP servers, and skills. Aura reads all of them, normalizes what it finds, and shows you where they disagree.
curl -fsSL https://tryaura.sh/install | shOr install with npm, or read the script before running it.
01
Adapters for Claude Code, Codex, and Cursor normalize instruction files, MCP servers, and installed skills into a single workspace model.
02
A check is synchronous and reads only the normalized model. It cannot touch the filesystem or run a command, so one scan evaluates every rule reproducibly.
03
A fix returns a plan of write, remove, move, and symlink operations — never actions. Aura owns the diff, the dry run, the backup, and the undo.
04
Every path in a plan, including symlink targets, must resolve inside the workspace or the Aura-managed part of your home directory. Plans that escape are rejected.
05
Adapters, checks, snippets, skills, MCP catalog entries, and presets are all contribution slots on a plugin, namespaced under the plugin's own ID.
06
Run the same scan locally and on every pull request. Findings carry stable IDs and resolved severities, so a gate means the same thing in both places.