People generate pet projects using Composer only.
But is Composer ready for the simple tasks?
I’ve chosen 3 simple tasks for Cursor AI Composer (as of 0.41):
- create a file
- create directory
- ls project directory
Since it is a Cursor AI Composer, not Cursor AI Chat, I expect the job to be done by Composer. I don’t expect Composer to gives me instructions back instead.
In below tests the directory of current project was empty.
Composer had also an empty context and reset (↻) before each command is applied.
create a file
command | gpt-4o | cursor-small | claude-3.5-sonnet | o1-mini |
---|---|---|---|---|
create file foo.txt |
||||
create foo.txt |
||||
hey, create file foo.txt |
||||
please, create a file foo.txt |
||||
hey, create foo.txt |
create directory
command | gpt-4o | cursor-small | claude-3.5-sonnet | o1-mini |
---|---|---|---|---|
create directory 'foo' |
||||
create foo/ |
||||
hey, do create directory "foo/" for me, please |
Note: you can create even nested directories if you ask to place a file there.
Example that works sometimes:
- claude-3.5-sonnet:
hey, please create file mydir/foo.txt
- gpt-4o:
please create file doc/lections/readme.md
if your filename or directory name sounds a bit “templatish” like mydir/subdir/foo.txt then model is likely to avoid creating, but will feed you back with instructions on how you do the job
ls
project directory
command | gpt-4o | cursor-small | claude-3.5-sonnet | o1-mini |
---|---|---|---|---|
list the content of the current project directory |
||||
dir the current project directory |
||||
ls the current project directory |
Note: o1-mini at least once printed out the file which was actually deleted long before. Prompt/context caching issue?