yangyu666/dsh-jev-prune ↗★ 6

dsh-jev-prune

基于语义和收据机制压缩与修剪上下文 适合在长对话中面临上下文压力、需要智能精简历史记录的用户。

套件
dsh-jev-prune
相容性
待驗證
Harness 依賴範圍
^0.1.5-rc.2 || ^0.2.0-rc.2
版本
0.1.1
授權
MIT
最近更新
2026年10月6日

安裝

$npx -p @deepseek-ai/dsh dsh plugin --profile web add dsh-jev-prune@0.1.1

Configuration

Start with a dry run in the profile's plugin entry:

- id: jev-prune
  config:
    dryRun: true
    judgeOn: pressure
    softLimit: "55%"
    compactReceipts: true
    compactOn: pressure
    compactSoftLimit: "70%"

After inspecting /jev, set dryRun: false to apply reductions.

OptionDefaultPurpose
keepModebudgetRank candidates; trimming budget follows context pressure.
preserveRecent / compactPreserveRecent4 / 1Independent recent-node protection.
headChars / tailChars600 / 200Verbatim content retained by layer 1.
compactQuantile0.34Tail fraction on each judgment axis; use the intersection.
compactToolsRead-only allow-listLimits which tool pairs may be compacted.
receiptTextChars400Bounds verbatim assistant-visible notes in receipts.
dryRunfalseJudge and report without applying changes.

Full configuration and pressure-gate details · Configuration example

In-session usage

Entry pointPurpose
/jev, jev_prune_statusLedger for both layers: cached judgments, cumulative savings, takeover state, tool-name index
jev_prune_nowForce one layer-1 trimming pass
jev_compact_now (supports dryRun)Force one layer-2 receipt compaction and list every gate's exclusion counts; in dryRun it also prints the full receipt text
jev_restoreSafety valve: fetch back the original text that a checkpoint moved out (read-only)
jev_probe_shapesPrint the real event shapes and the resolved tool names, for adapting to a different DSH version

In normal operation both layers are driven automatically by context pressure; no manual step is needed.