Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

fast-compaction-dsh

DeepSeek Harness port of fast-jev-compaction: Jev keep/truncate/drop verdicts per tool call replace lossy summaries.

ProjectbetaCustom license
View on GitHub ↗

Overview

fast-compaction-dsh brings verdict-based context compaction to DeepSeek Harness. Instead of asking an LLM to summarise old turns, which can lose file paths, errors, constraints or commands, every tool call and result in the compacted region is scored by jev-latest: stale calls are deleted, half-stale results are truncated to a bounded head, and everything else stays byte for byte. User and assistant text is never removed.

It is a port of tamaratran/fast-jev-compaction adapted to DSH's compaction seam. Calls and results are paired by call id; the first message and the newest six messages are pinned. The state sent to Jev is the whole compacted region with results replaced by short notes, fitted into a 25k-token budget in stages. On any failure, including a missing key, Jev errors, malformed answers or insufficient reduction, it falls back to the shipped basic summariser.

What the decision model decides

TypeQuestionAnswers
ChoiceShould this tool call/result be kept, truncated or dropped?keep · truncate · drop

How it works

  1. Pair tool calls and results; pin first and recent messages
  2. Fit the region into the state budget
  3. Jev scores each call and result
  4. Apply verdicts; fall back to compaction-basic on failure

Strengths

  • Kept content stays verbatim
  • Safe fallback to the stock summariser
  • Staged state fitting

Caveats

  • Specific to DeepSeek Harness
  • Custom licence per the listing (upstream is MIT)

Getting started

Install as a DSH compaction plugin per the README and set TYPESAFE_API_KEY.

At a glance

Type
Project
Status
beta
Language
TypeScript
Runtime
DeepSeek Harness plugin
Has code
yes
License
Custom license

Use cases and integrations

How curated lists describe it

  • Verdict-based context compaction for DeepSeek Harness — replaces lossy LLM summaries with fast keep/truncate/drop decisions from jev-latest; everything kept stays verbatim. Port of tamaratran/fast-jev-compaction.

    logicrw/awesome-jev-projects

  • Verdict-based context compaction for DeepSeek Harness — replaces lossy LLM…

    wh000wh000/awesome-jev-live

Frequently asked questions

Does it summarise history?

No, it keeps, truncates or drops tool items; text stays verbatim.

What if Jev fails?

It falls back to DSH's compaction-basic summariser.

What is it based on?

A port of tamaratran/fast-jev-compaction.

Related projects

  • deepseek-harness-jev-pre-compaction

    DeepSeek Harness plugin that asks Jev which tool results can be pruned early, delaying or avoiding lossy compaction.

    ProjectbetaTypeScript
  • dsh-context-curator

    DeepSeek Harness compaction backend that asks Jev which tool calls and outputs to keep, preserving everything else verbatim.

    ProjectbetaJavaScript
  • dsh-jev-prune

    DeepSeek Harness compaction plugin where Jev decides which tool results are still needed and code writes deterministic receipts.

    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-jev-compaction

    Codex plugin that builds handoff packets by asking Jev whether each old read-only tool record should stay, keeping retained text verbatim.

    ProjectbetaJavaScript
  • cursor-clijev-compaction

    Cursor CLI plugin that asks Jev keep-or-drop nouls for each tool call and result, then re-injects kept facts after compaction.

    CLIbetaTypeScript