Winnow
Keep relevant Claude Code tool output in context and recall hidden blocks on demand.
What Jev decides
The probability that each block belongs in the agent context.
Winnow judges large Read, Bash, and Grep results block by block, hiding only confident low-relevance sections behind recoverable stubs.
- Jev role
- Judge whether each tool-output block is needed for the current task.
- Input
- The current task, tool arguments, and blocks of tool output.
- Output
- Per-block relevance probabilities used by local keep-or-hide rules.
- Decision frequency
- For eligible large Read, Bash, and Grep results.
- Fallback
- Uncertain and error-bearing blocks stay visible; hidden text remains in a local recall cache.
How it works
Overview
A Claude Code hook divides large tool results into blocks and asks a judge whether each block matters to the current task. Confidently irrelevant blocks become short stubs with local recall keys; errors and uncertain blocks remain visible. Jev is the default judge, while a Claude Haiku adapter is available when Jev access is unavailable.
- Language
- Python
- License
- MIT
- Author
- GhalebDweikat
Sources
Related projects
fast-jev-compaction
Prune stale coding-agent tool history without rewriting the context that remains.
What Jev decides
Whether each old tool call and its result should remain available.
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.