Repository navigation
Replies: 1 comment
|
Another example of the same thing. An agent wrote:
It rendered as a file chip labelled This one is narrower than A guard on the basename would catch it without touching |
|
Another example of the same thing. An agent wrote:
It rendered as a file chip labelled This one is narrower than A guard on the basename would catch it without touching |
Uh oh!
There was an error while loading. Please reload this page.
Inline code like
`size/provider/file.jpg`renders as a chip labelledfile.jpg. In my case it was part of an image URL, not a file in the workspace. The chip drops the first two segments, so the reader can't see what the agent wrote, and clicking it opens a file that doesn't exist.No text rule can catch this.
size/provider/file.jpghas the same shape assrc/main.ts, and the current tests treatvendor/jquery-3.6.0.min.jsandshare/man/ls.1as files on purpose. Absolute paths have a cheap fix (require a known filesystem root or the workspace, #19012). Relative paths don't.I can see two ways out:
projects.searchEntriesandprojects.listEntries. It costs a round trip per message, though, and the chip would appear late or flicker. Remote clients would feel that most.size/provider/file.jpg, and not just the file name. There's no new I/O. A wrong guess then costs one bad click target, and the text stays readable. The catch is longer chips for real files deep in the tree.I lean towards 2, since what actually hurts is losing the text the agent wrote. Option 1 is more correct, but the cost looks too high for this. Is either direction acceptable?
All reactions