See the context a Session submits to its model in the DeepSeek Harness Web UI: the system prompt, the tool schemas and every message, drawn as a grid of token squares.
インターフェース · 生産性ツール
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:LionNatsu/dsh-context-explorer
Context Guardian as a native DeepSeek Harness (DSH) bundle: deterministic, seq-pointer-preserving compaction with model-facing recall and search. Companion to dsh-tool-guardian.
メモリ・ナレッジ · 開発ツール · サービス連携 · 生産性ツール
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:LuminariSoftwares/context-guardian
Context offloading for DeepSeek Harness: consumed tool calls and their results are replaced in the model-visible history by a short stub, while the originals stay in the session log for replay. A between-step policy releases the accumulated backlog in batches, and an offload_context tool lets the model unload what it has finished with.
Agent 連携 · 生産性ツール
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:Hanochhu/dsh-context-offload
DSH plugin: context reclaim at tool-result creation time, and at every later seam where context grows. Lossless noise purification, per-content-type bounding with retrieval pointers, tool-definition folding, system-prompt section-order stabilisation, and prompt-line trimming. Five Cordis seams, zero runtime dependencies. Also takes over the host's writable-request path, so the official `dsh-llm` file patch is no longer needed, and reports — without ever applying — which other official file patches are present. Never rewrites committed history, so the provider prefix cache is never invalidated.
その他の拡張
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:xiamubumu/dsh-context-reclaim
Interactive behavioral shaping for a DeepSeek Harness (DSH) session: rewrite the model-visible conversation history in place (reply text and reasoning chain separately), with an audit trail and undo. 交互式行为塑形:就地改写会话的模型可见历史,回复与思考链分开编辑,并带改写审计与还原。
生産性ツール
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:zeranhub/dsh-context-shaping
Structured context for DeepSeek Harness: split the conversation into typed sub-contexts (reasoning / tool calls / replies) and index them with a table of contents. 结构化 dsh 上下文:按类型拆分小上下文(思考/工具调用/正式回复),以结构化目录为索引。
Agent 連携
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:kkkkkklze/dsh-context-structurer
Keeps the chat context-usage ring honest: an empty usage sample from a failed attempt no longer collapses the reading to 0% across a retry, and the ring's percentage matches the figures beside it.
その他の拡張
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:icrefin/dsh-context-usage-guard
Manually resume an interrupted DeepSeek Harness conversation: when a turn stopped by hand, by a network or provider failure, or by the output-token limit, one command picks the run up exactly where it stopped. | 手动继续被中断的 DSH 对话:当轮次因手动停止、网络或模型请求失败、输出 token 上限而中断时,一条命令即可从断点续跑。
その他の拡張
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:CaiBai-Fish/dsh-continuechat
Contract check for DeepSeek Harness plugins: statically verifies that registered tools' output.render() returns ContentBlock[], and warns when a session event type falls outside the kernel vocabulary. Warn-only. 契约体检插件(不检测恶意)。
金融データ · 生産性ツール
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add github:imtokenxinluo/dsh-contract-check