Large skills no longer slow down your workspace

Skills with many files, like a gallery of reference images, used to be expensive to explore. Each file an agent opened made the server load the entire skill, so an agent walking a large one could slow things down for everyone using that server, and in the worst case interrupt them. Now an agent opening a file gets that file and nothing else, however large the skill is.

How it works

• One file costs one file. Every version of a skill is stored as a list of its files, so listing a folder reads the list and opening a file fetches only that file's bytes.
• Fair sharing. One agent exploring a large folder waits its turn instead of crowding out other agents and people on the same server.
• Same skills, no changes needed. Existing skills move over on their own, and what they contain doesn't change.

Also shipped

• Publishing a new version stores only the files that changed, and restoring an earlier version is instant.
• A run's record is now stored as separate files, so opening a run's logs no longer unpacks anything.
• Uploaded archives and documents are now handled inside strict limits, so a malformed file can't affect anyone else.
• geni pushes go straight to storage, so large pushes are faster and more reliable.