Apache-2.0 DeepSeek Harness plugin installed via `dsh plugin add` that adds a Context dashboard tab and `/context` command for context-window composition, history, compaction events and token costs.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-08-21
License
Apache-2.0
01What can it help you accomplish?
See exactly what fills the agent's context window right now
A six-color stacked composition bar scaled against the model's full context window (system prompt, tool schemas, user messages, injected context, assistant replies, tool results), remaining headroom, and the top-5 most expensive tool schemas
Developers running DeepSeek Harness (dsh) who need to find out which part of the conversation is eating the context budget
Track how the context window grows and shrinks across a session
One stacked bar per model request with Turn/Step granularity, hover tooltips, pinnable full breakdowns (estimated vs. provider-reported tokens), and ✂ markers where compaction or pruning reclaimed tokens
Engineers debugging long DeepSeek Harness sessions who want to watch compaction and prune events instead of guessing
Inspect what any single request was actually assembled from
Context browser (also via the `/context` command): six collapsible category sections expanding into every element with its token price and actual content — system prompt, tool schemas, message text, reasoning, tool-call arguments and outputs
Plugin and skill authors who want to verify what context their injections add to each DeepSeek Harness request
02How to install into DeepSeek Harness
Prerequisites
- A DeepSeek Harness (dsh) installation — the plugin is added from any existing dsh installation
Installation steps
- 01
Install from any DeepSeek Harness installation: `dsh plugin --profile web add dsh-context`
$ dsh plugin --profile web add dsh-context
- 02
Update later with: `dsh plugin --profile web update dsh-context@latest`
$ dsh plugin --profile web update dsh-context@latest
- 03
Start the web UI with `dsh web` — no build step, no restart
$ dsh web
Verify the integration
Not specified by the author
03DSH integration and capability boundaries
Native DeepSeek Harness plugin installed via `dsh plugin add`, adding a Context dashboard tab to the dsh web UI plus a `/context` slash command
Context tab dashboard
any open DeepSeek Harness session in the dsh web UI→context stats (turns, steps, injections, compactions, prunes), current composition, per-request history, context events, and the model-visible message list
`/context` slash command
`/context` typed in a dsh chat (or picked from the `/` menu)→a centered dialog with the Current Composition card and the Context browser — same composition bar, legend, and per-step browsing as the tab, without leaving the chat
Context events log
the session's compaction, prune, injection and model-switch events→every event with its token delta, turn/step attribution and timestamp, filterable by category (Inject / Compact / Prune / Switch)
Context browser
Live (next request) or any retained step from the picker→per-element breakdown of what the request was assembled from, down to actual content (system prompt text, tool JSON schemas, message text, reasoning, tool-call arguments and outputs)
elements older than the loaded chat window page older history in automatically when expanded
04Who is it for? When not to use it?
Good for
- Developers running DeepSeek Harness (dsh) who need to find out which part of the conversation is eating the context budget
- Engineers debugging long DeepSeek Harness sessions who want to watch compaction and prune events instead of guessing
- Plugin and skill authors who want to verify what context their injections add to each DeepSeek Harness request
Not for
- The dashboard is delivered as a tab in the dsh web UI and must be installed with the `web` profile (`dsh plugin --profile web add`); usage is tied to running `dsh web`.
- The repository was created on 2026-08-14 and is at v0.19.2, so the plugin is very young and its interfaces may still change.
05Compatibility, maintenance and safety notes
- Steps before a compaction are reconstructed from the removed-message archive, so their makeup can be only approximate; the card says so when that happens.
- The dashboard is delivered as a tab in the dsh web UI and must be installed with the `web` profile (`dsh plugin --profile web add`); usage is tied to running `dsh web`.
- The repository was created on 2026-08-14 and is at v0.19.2, so the plugin is very young and its interfaces may still change.
Apache-2.0 · actively maintained (latest release v0.19.2, 2026-08-20)
06Frequently asked questions
How do I install dsh-context in DeepSeek Harness?
Run `dsh plugin --profile web add dsh-context` from any dsh installation, then start the web UI with `dsh web`. No build step and no restart are needed. Update later with `dsh plugin --profile web update dsh-context@latest`.
Is it a native plugin or an MCP server?
It is a native DeepSeek Harness plugin: after install it adds a Context / 上下文 tab to the dsh web UI and a `/context` slash command inside chat sessions. There is no MCP setup.
What can I see about my context window?
The six-part composition of the current window scaled against the model's full context window, remaining headroom, the top-5 most expensive tool schemas, one stacked bar per model request, every compaction / prune / injection / model-switch event with token deltas, and the model-visible message list with per-message token cost.
Are the token counts exact?
The breakdown shows estimates alongside provider-reported actual prompt/output tokens when available. Steps before a compaction are reconstructed from the removed-message archive, and the card notes when a step's makeup is only approximate.
Does the plugin send my data anywhere?
The README describes pure visualization of context data inside the dsh web UI — browsing history and expanding elements may page older chat history in automatically, but no external service is mentioned.
07Related DSH workflows
api-relay-audit
by toby-bridges
Local security audit for AI API relays and LLM proxies: detects prompt injection, model substitution, tool-call rewriting, SSE anomalies, error leakage, and Web3 wallet risks.
awesome-deepseek-harness-plugins
by zhiyuan-fan
Curated DeepSeek Harness (DSH) plugins, extensions, tools, skills, clients, runtimes, integrations, and verified references — English and Chinese.
awesome-ai-pedia
by awesome-ai-pedia
One-stop AI resource encyclopedia for dsh — LLMs, agents, RAG, multimodal, MLOps, AI tools, vibe coding and learning roadmaps, constantly updated.
awesome-deepseek-harness
by libukai
The ultimate guide to DeepSeek Harness — quick-start, curated resources, hand-picked plugins and useful tools.
08Data and sources
The best [DeepSeek Harness plugin](https://www.deepseek.com/harness/) for Agent's context insights and management.
dsh plugin --profile web add dsh-context
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-08-21. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.