dsh-jev-prune
DeepSeek Harness compaction plugin where Jev decides which tool results are still needed and code writes deterministic receipts.
Overview
dsh-jev-prune replaces the decisions in DeepSeek Harness's two-layer context reclamation with Jev's structured judgments while leaving the compaction algorithms untouched. DSH's default trimming is purely by size, chopping the middle of large tool results, and its region compaction asks the model to write a summary, which can hallucinate.
In layer one, Jev decides per tool result whether it will still be needed: needed results are never trimmed however large, and stale ones are trimmed however small (above a minimum size), falling back to DSH's default when no judgment is available. In layer two, region compaction injects a deterministic receipt generated by code with no model inference. The design rule is that what should not be generated by a model is not. The judgment backend is pluggable (Jev, rules or a self-hosted model), and CI is set up.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Will this tool result still be needed? | yes · no |
How it works
- DSH triggers result pruning or region compaction
- Jev judges each tool result as needed or stale
- Needed results kept verbatim; stale ones trimmed
- Region compaction inserts a code-generated receipt instead of a summary
Strengths
- No model-written summaries
- Large but needed results are protected
- Pluggable judgment backend with fallback
Caveats
- Specific to DeepSeek Harness
- Tool output sent to TypeSafe when using Jev
Getting started
Install as a DSH plugin from the dsh-plugin.org listing or GitHub and configure a judgment backend.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- DeepSeek Harness plugin
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev-judged context compaction for DeepSeek Harness: semantic tool-result pruning + deterministic receipt compaction
logicrw/awesome-jev-projectsJev-judged context compaction for DeepSeek Harness: semantic tool-result pruning + deterministic receipt compaction.
hellogumbo/awesome-jev
Frequently asked questions
Does a model write summaries?
No, region compaction uses a deterministic code-generated receipt.
What does Jev decide?
Whether each tool result will still be needed.
Can I avoid Jev?
Yes, the backend can be rules or a self-hosted model.
Related projects
DeepSeek Harness plugin that asks Jev which tool results can be pruned early, delaying or avoiding lossy compaction.
ProjectbetaTypeScriptDeepSeek Harness compaction backend that asks Jev which tool calls and outputs to keep, preserving everything else verbatim.
ProjectbetaJavaScriptDeepSeek Harness port of fast-jev-compaction: Jev keep/truncate/drop verdicts per tool call replace lossy summaries.
ProjectbetaTypeScript- fast-jev-compaction★ 7,616
Claude Code plugin and npm library that replaces the compaction summary with Jev keep, truncate or drop scores for every tool call.
PluginbetaTypeScript Codex plugin that builds handoff packets by asking Jev whether each old read-only tool record should stay, keeping retained text verbatim.
ProjectbetaJavaScriptCursor CLI plugin that asks Jev keep-or-drop nouls for each tool call and result, then re-injects kept facts after compaction.
CLIbetaTypeScript