You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
These are basically pattern match rules on "events" like the model creating a file, a file being added to the context, etc. They allow prompts to be automatically customized by context.
There's an interesting blog post here by someone who argues for the construction of "stdlibs" of rules that help power up the model:
The Cursor format is a bit odd - it appears to be YAML embedded inside Markdown with header context. Presumably it's done this way so rules can be created using a GUI. I don't personally use Cursor so don't care about compatibility. Other people who work in teams might care more. It strikes me that the format doesn't seem to be terribly intuitive, nor documented, so Aider could perhaps invent a better format.
A simple git-like approach would be to have an .aider directory that can contain hook scripts. When an event occurs like a response being received from a model, hook scripts would be invoked as a sub-process which speak a simple file/stdout based protocol, perhaps with some other commands existing in the environment. For instance, a script added to .aider/hooks/on-edit.d/do-the-thing could:
Fix common errors made by a model.
Automatically extend the context based on file type or location.
Recursively invoke a model on the first output of the model (perhaps by calling back into Aider using a command supplied in an environment variable?)
Generate a suggestion or automatic submission of the next command.
And so on. On Windows Aider could notice the .ps1 extension to send the script to PowerShell.
The text was updated successfully, but these errors were encountered:
The Cursor IDE is basically VSCode + something like Aider. It has a concept called Rules:
https://docs.cursor.com/context/rules-for-ai
These are basically pattern match rules on "events" like the model creating a file, a file being added to the context, etc. They allow prompts to be automatically customized by context.
There's an interesting blog post here by someone who argues for the construction of "stdlibs" of rules that help power up the model:
https://ghuntley.com/stdlib/
Example rules include:
The Cursor format is a bit odd - it appears to be YAML embedded inside Markdown with header context. Presumably it's done this way so rules can be created using a GUI. I don't personally use Cursor so don't care about compatibility. Other people who work in teams might care more. It strikes me that the format doesn't seem to be terribly intuitive, nor documented, so Aider could perhaps invent a better format.
A simple git-like approach would be to have an .aider directory that can contain hook scripts. When an event occurs like a response being received from a model, hook scripts would be invoked as a sub-process which speak a simple file/stdout based protocol, perhaps with some other commands existing in the environment. For instance, a script added to
.aider/hooks/on-edit.d/do-the-thing
could:And so on. On Windows Aider could notice the .ps1 extension to send the script to PowerShell.
The text was updated successfully, but these errors were encountered: