Overview & result
Jev-pruner is a Claude Code hook that intercepts long Bash results after execution but before the main model receives them. Jev scores bounded output chunks; code keeps diagnostics and archives removed text for recovery.
From the original post
A Claude Code plugin that uses TypeSafe's Jev to trim noisy Bash output
How Jev fits in the loop
- Intercept eligible Bash output and divide it into bounded chunks.
- Ask Jev Noul relevance questions using task and transcript context.
- Keep required chunks, preserve diagnostics and archive any pruned original text.
How to reproduce
- Install the documented Claude Code plugin and run its tests with mocked Jev responses.
- Send a large Bash output through the hook and inspect kept text plus the archive footer.
- Read the archive when necessary; verify that errors and protected document-like outputs are left intact.
Why this build matters
A concrete context-control design where Jev decides relevance but code preserves recoverability and failure safety.
Limitations
- The hook skips outputs below its threshold and preserves many document/error classes; it does not prune every tool result.
- Model failures or archive-write failures leave original output untouched; reduced context does not prove faster task completion.