DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

Context Compression Improved — DSH Plugin for DeepSeek Harness
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in
← Plugins

dsh-context-compression-improved

Context Compression Improved

DSH plugin dsh-context-compression-improved

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add dsh-context-compression-improved@0.1.1
READMECompatibilityVersions
Context Compression Selector settings UI

Description

DSH plugin dsh-context-compression-improved

Compatibility and provenance

Context Compression Improved is published as dsh-context-compression-improved and currently resolves to version 0.1.1. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
web
Release source
npm
Registry updated
9/20/2026

Versions

0.1.1stable
9/15/2026

Related plugins

Loading related plugins…

Latest
0.1.1
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
12.8 MB
Files
28
Surface
web
License
MIT
Source
npm
GitHub
★ 1
Weekly downloads
786
Security scan
✓ v0.1.1 scan passed
Last push
9/19/2026
View source ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue

Related plugins

More verified plugins in memory-context.

Memory Plugin@openviking/dsh-memory-pluginOpenViking memory and context bundle for DeepSeek HarnessContextdsh-contextA DeepSeek Harness plugin for context insight and management, with context dashboard and context command, for understanding how the context is made of, and how it evolves.Weknora@wxg-prc-cpg/dsh-weknoraWeKnora knowledge retrieval tools for DeepSeek Harness (dsh): semantic search, document reading and RAG/agent answers over your own knowledge bases.Memsearch Dsh@zilliz/memsearch-dshMemSearch plugin for DeepSeek Harness: shared markdown memory across agents, with capture, pre-step context injection, memory-recall skill, and a skill-candidate review panel.

README

dsh-context-compression-improved

The installable Product Bundle for the unofficial community DeepSeek Harness context-compression selector.

0.1.0 highlights: DeepSeek V4 Flash Vision's official tokenizer is included; users can choose the model-driven Auto Compact threshold; standard profile watermarks and compression parameters follow that choice.

dsh plugin --profile web add dsh-context-compression-improved@latest
dsh --profile web --dump-config

This one command installs the Bundle and its pinned @huggingface/tokenizers dependency; there is no separate runtime package. The Bundle contributes Host settings, the Web UI, and a reversible preset overlay. All presets except the exact built-in minimal id receive the compression stack; switching non-Minimal presets preserves the saved selector settings. Minimal pauses plugin compression without deleting the setting.

What it provides

It provides deterministic Fresh, Aggregate, History, Native tool-result, and Custom TailTrim compression; a plugin-owned context_compression_retrieve recovery tool; structured audit records; and a pinned offline DeepSeek V4 tokenizer. It uses only public DeepSeek Harness APIs and supports both 0.1.1-rc.2 and 0.1.2-alpha.5; it does not patch Harness core.

Lossy rewrites require exact same-revision counts before and after replacement. Verified model ids are deepseek-v4-flash, deepseek-v4-pro, and deepseek-v4-flash-vision-exp. Vision text counting uses the separately pinned deepseek-ai/DeepSeek-V4-Flash-Vision-Exp tokenizer. Vision images use the official image-processor arithmetic validated against official golden fixtures and are reported as tokenizer-estimate: the four alignment residues over valid intrinsic dimensions are collapsed to their midpoint, with 384 tokens retained as the per-image upper bound; malformed or unevaluable dimensions use a fixed 256-token fallback. The estimate is not exact because the absolute prompt position and the adapter's final image projection are not publicly observable. Image-bearing tool-result candidates remain exact-ineligible and intact; unknown models, unavailable tokenizer assets, unsafe tool groups, and incomplete text measurements still fail open.

Audit log records use the prefix context-compression audit and distinguish policy snapshots, skipped/disabled components, committed rewrites, fail-open errors, and observed core compaction/summary events. No audit record contains prompt or tool-result content.

The first policy-frozen record carries the complete settings/deployment snapshot. It is emitted through the Harness logger, so its retention follows the configured log sink; standard prune/replacement Session events remain the durable proof of committed rewrites. This package does not hard-code a private path below ~/.dsh.

Verified compatibility: DeepSeek Harness dsh-v0.1.1-rc.2 and dsh-v0.1.2-alpha.5, Node ^22.19.0 || >=24. No Harness core patch is required.

See the full README for profiles, defaults, audit evidence, Adaptive/cache limitations, upgrade/removal, and security guidance. Tokenizer provenance and checksums are in assets/deepseek-v4/manifest.json and assets/deepseek-v4-vision-exp/manifest.json, summarized in THIRD_PARTY_NOTICES.md. This package is not affiliated with or endorsed by DeepSeek.