compaction 55

Terminal-style input history for the DeepSeek Harness web composer: edge-first arrows with exact draft/caret restore, browser-local persisted history, Ctrl+R reverse search, workspace-scoped recall, and fully configurable keys — plus sliding-context aware

DeepSeek Harness plugin that uses configured model providers for image analysis and context compaction.

Lossless context folding for DeepSeek Harness: the oldest span is replaced by a brief, and the originals stay in the session log.

A live context-window donut for DeepSeek Harness: token usage, compaction savings, and cost at a glance

Deterministic context-pruning compaction backend for dsh (DeepSeek Harness) — zero-LLM summaries, /dcp command, works out of the box. Design references Opencode-DCP/opencode-dynamic-context-pruning.

Keeps long coding-agent sessions lean: wrap a stretch of work in a named task, and when it ends, fold that whole span into one short titled summary. The conversation stays readable, context costs stay low, and any fold's original content can be read back

Context compaction that keeps its receipts: a five-section handoff summary, a durable segment index, review tools that read compacted text back, optional working notes, and a read-only extraction exit. Mounts as the compaction-basic row's replacement; cha

Context compression tool (context_compact): the agent writes the replacement checkpoint itself and hands it to the host compaction engine, which skips the LLM summarizer call; automatic compaction stays on the official engine.

Model-driven context-window rollover bundle for the DeepSeek Harness: fresh working context, durable model-managed notes, a small verbatim recent tail, and targeted history recovery — no summarization.

DSH plugin: fixes compaction failure on local qwen3.8-27b gateways — xhigh thinking burns the entire output token budget, so thinking is off for compaction-only, with the model's non-thinking sampling parameters

会话级自动上下文压缩插件：回合中每步前 + 回合结束检测上下文用量，超过会话阈值自动 compact，摘要注入上下文后自然继续（DeepSeek Harness / DSH）

Model-scoped DSH plugin: GPT Responses lifecycle, Native V2 compaction and Hosted/Alpha Search; Grok native Web/X Search

GLM Mode: a fully tuned coding agent preset for zhipuai/glm-5.3-flash in DeepSeek Harness — PTC/Code Mode tool presentation, GLM-specific auto-compaction (75%/22%), native multimodal, and a guard family (reasoning loop guard + tool spin guard). The preset

Session context governor for DeepSeek Harness: prompt size and per-step cost in dollars (cache-read/cache-write aware), cache-hit, context bands, a per-session compaction threshold derived from the model window, DeepSeek balance and peak/off-peak tariff,

DeepSeek Harness plugin: tells the model its real, host-measured context occupancy every turn — and keeps compaction summaries from carrying false "context is nearly exhausted" claims.

Personal dsh preferences: a tunable companion to the official compaction engine (threshold, retention, summarization model/reasoning, turn-end compaction), llama.cpp model import for the Models page, and tab/notification helpers (status-light favicon, bro

DSH plugin: counts completed compactions per session and raises a floating reminder — with bilingual UI — only after repeated compactions plus heavy extra token spend.

Pure-incremental directive-driven compaction for DeepSeek Harness: /compact-directive <requirement> summarizes the session middle per your requirement, /trim-directive <requirement> trims the whole conversation per your requirement

Adaptive Reversible Context (ARC) for DeepSeek Harness — provider-aware, cache-conscious context governance with reversible recovery.

Compaction quality / hallucination detector: when a conversation span is compacted, check on two axes that key information survived 鈥?FIDELITY (assertions in the summary must be found verbatim in the original leaf events, else flagged as fabricated) and C

Step-0 observation probe for dsh-compaction-audit: logs every compaction/* session event and probes whether shadowedSeqs still resolve to readable events in session.events. Observe-only; never injects, never blocks.

Model-autonomous context compaction for DeepSeek Harness: mind_map surveys the agent's own context surface (seqs, token weights, clear-boundary markers and the clear-mind playbook), clear_mind replaces a chosen span with a model-written checkpoint through

Manual context compaction for DeepSeek Harness: keep the latest N conversation items with stop or batch mode and a per-session compaction archive — embedded in the context meter popup when the meter panel slot exists, or a composer button otherwise.

Windowed, reversible context management for DeepSeek Harness; replaces Basic compaction.

Jev-scored context compaction for DeepSeek Harness -- keep the messages that matter, compact the ones that do not

dsh plugin: a compact_now tool letting the model trigger compaction itself at a safe point, plus a context_status tool so it can check real token usage against its context window, instead of only the automatic token-threshold trigger.

Model-free /trim for DeepSeek Harness — drop the oldest, least valuable span of context on demand (the `/trim` command) or automatically when a request hits the model's context wall, without any model call.

Reasonix-style cache-aware compaction backend for DeepSeek Harness (DSH). Replaces/enhances compaction-basic with compact_ratio, one structured summary checkpoint, and a stable recent tail.

Manual context compaction for DeepSeek Harness: keep the latest N conversation items with stop or batch mode and a per-session compaction archive — embedded in the context meter popup when the meter panel slot exists, or a composer button otherwise.

Codex-style lossless-retrieval context manager for the DeepSeek Harness: token-budgeted sliding window, persistent working notes, and cold-history search over the append-only session log

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.

Manual budget-targeted context compression for DeepSeek Harness: the conversation model picks ranges, a cheap flash-tier route writes the checkpoint summaries.

DSH plugin: mark messages as invalid so an operator annotation digest steers what a compaction checkpoint absorbs. Cache-preserving: appends after the cached prefix, never rewrites history.

Deterministic, zero-LLM, byte-budget context compaction for DeepSeek Harness, with byte-exact recall of compacted history.

DSH plugin (NInfer engine only): fixes compaction failure on local qwen3.8-27b gateways served by NInfer — xhigh thinking burns the entire output token budget, so thinking is off for compaction-only, with the model's non-thinking sampling parameters; the

DeepSeek Harness plugin for bounded, provenance-linked context steady state with recoverable raw history

Absolute token trigger for DeepSeek Harness automatic compaction: condense at a fixed token budget or at the model window ratio, whichever comes first.

Active Context Pruning (ACP) for the DeepSeek Harness — model-driven context management as a CompactionEngine backend.

Stop DeepSeek Harness compaction losing its summary: drop the tool schemas from the summarization call, because GenerateOptions has no toolChoice to say "don't call a tool".

DSH host plugin: makes /compact use a user-chosen provider/model instead of the session's chat model, configurable in Settings.

High-fidelity, faithful, bilingual, recursive compaction backend for DeepSeek Harness — a drop-in upgrade over dsh-compaction-basic.

Host- and agent-layer DeepSeek Harness plugin that prices every automatic context rewrite before it lands, asks the human by default, and reports what the provider had to re-read cold.

Adds a Compact-and-run option to the DSH Web plan-review card: compact the history submitted before the plan, then approve it.

DeepSeek Harness（dsh）Web GUI 插件：在上下文占用面板新增「压缩」与「交接」按钮。Adds Compact / Handover buttons to the dsh context panel (unofficial plugin).

在上下文监控卡片内直接调节自动压缩的触发阈值。host 半在运行时替换 @deepseek-ai/dsh-compaction-basic 已挂载引擎的 thresholdRatio 并持久化，client 半把滑块画进卡片面板。

DSH 上下文交接压缩后端：程序抽取硬事实（路径/命令/报错逐字保留）+ 模型只补四节因果，PTC 感知。衍生自 fan56/dsh-dcp（MIT），主体已重写。

Jev/System-One semantic pruning of stale tool results for DeepSeek Harness; keeps conversation text verbatim and forgets stale tool output before summary compaction is needed

Active Context Pruning (ACP) for the DeepSeek Harness — model-driven context management as a CompactionEngine backend.

Auxiliary-call reasoning policy for dsh: the harness marks its own background calls (compaction checkpoints, session titles) with a `purpose` and names no reasoning level for them, so `dsh-llm` fills the gap from the route's declared default and they thin

Keeps a local model's context at a size the GPU handles well: measured prefill speed, hard ceiling, early compaction for DeepSeek Harness