Plugins

Layered context management: at each turn end older tool output becomes a deterministic card, the model can fold older ranges into its own summaries in up to three tiers (compress), and every original is archived locally for decompress and search_context. Works alongside DSH-Mem0: superseded runtime-context snapshots (such as recalled memories) are never archived, carded or folded, and are replaced by a one-line placeholder at turn end when the prompt cache is already broken, when the saving outweighs the cache miss, or above 75% of the window; a strataBudget service (pressure, allowance, countTokens) and a strata/fold-landed event are provided for other context plugins; strata_status warns when DSH's automatic compaction is off or its official summary keeps failing. Its L1 layer, archive rotation, prune guard and preset takeover are forked from dsh-live-compact 1.0.6 (MIT); uninstall dsh-live-compact or billion-context first — they replace the same compaction rows and Strata will not take over while either is installed.

Wall-clock session and turn timing for the DSH Web GUI: a timer chip in the session header shows the session total, every finished turn shows its own duration above the action row, and a sidebar panel totals today, the last 7 days, the last 30 days or all sessions. The numbers are folded from turn/start and turn/end in the session log, so paging, compaction and restarts do not change them.

Cache-first context compaction with an optional adaptive economic policy, observation-only by default, without cache-hit or cost-saving guarantees.

Notes-backed context compression for DeepSeek Harness: replaces LLM-summarized compaction with a model-written persistent notebook plus recoverable raw history, so the rollover boundary itself costs zero model calls.

Keep conclusions alive across context compaction: the model writes a task checkpoint and the plugin re-injects it into the system prompt's first node — the span compaction never touches, so it survives any number of compactions. Four-tier injection (0/95/175/≤12000 chars) scales with session relatedness; checkpoints live in the plugin's own folder, never in your workspace.

Copies the compaction checkpoint into the sticky-notes archive when context is compacted and wakes the session to write what was lost.

One settings card that gives every DSH agent mode its own compaction threshold: it registers <mode>-compact-<ratio> preset copies with compaction-basic.thresholdRatio rewritten, re-registers when the value or the selection changes, and can adopt one copy as the default mode (standard > cordis > minimal > ptc).

Replaces the built-in compaction provider: at a configurable fraction of the context threshold (85% by default) a detached task summarizes everything up to the previous turn, and the next step boundary commits that held summary as a checkpoint with zero model calls, so the swap itself takes milliseconds. /bcompact \[message] starts the same generation on demand and then sends the message. A settings page sets the trigger ratio, the summarization model, and how many consecutive failures stop background generation until a half-open probe gets through. Subagent sessions, a changed prefix or an expired summary fall back to the built-in synchronous path.

Compress every idle session's context at a fixed time each morning on DeepSeek Harness: a daily run (08:30 by default), a catch-up run on the next launch after a missed slot, and a /compact-all command for on-demand compaction, with per-session outcomes reported.

Claude-Code-compatible memory for every model in DeepSeek Harness: layered CLAUDE.md/AGENTS.md instructions with @import and rules, a persistent MEMORY.md memory with full writing guidance, refresh on resume and compaction, optional recall, background ext

溢出兜底与续命：上下文超限、内置压缩放弃后自动再压一次并重试；仍失败时把原会话整理成交接包，在新会话接着聊。不换引擎、不加界面；续命默认显式触发，原会话只读。 · Overflow fallback and session revival for DeepSeek Harness: one last compaction retry, then continue the dead session in a fresh one with a handoff. No engine swap, no UI, o

StartLux Decision: a cheap judgement layer for DeepSeek Harness — all nine decision points (tool authorization, model routing, delegation, compaction, turn continuation, retrieval, skill pick, session title, COT translation) are implemented and individual

Adjusts the context-window percentage at which DSH triggers its native automatic compaction: a slider on the plugin page in the plugins manager sets it from 1% to 100% (default 80%, arrow keys adjust by 1%), and the retained-history budget shrinks with lower thresholds so early compaction is not defeated; the compaction run itself stays with the native engine.

Recovers a session the host already refused for exceeding the context window: it re-checks the gateway message the built-in classifier misses, takes the real window out of that message, and folds the oldest span into one checkpoint without a model call (tool-result pruning, the host compaction engine when it can shrink the projection, or a mechanical digest fold), then retries the failed step; a context_rescue tool and /rescue, /rescue-status expose the same path by hand.

Read-only viewer for archived DSH sessions: browse the archive with workspace filtering and title/ID search, and read the transcript folded from the durable log — including messages from before a context compaction — without unarchiving.

Browse and search provider models to choose an independent summary model for DSH Standard mode, or let Auto follow the conversation route; save extra automatic pressure-compaction passes, while manual compaction attempts one useful pass. Includes Chinese/English theme-aware settings; start a new Standard-mode conversation after installation to apply the configuration.

Compaction ledger for the DeepSeek Harness: audit every context compaction (what was shadowed, when, at what token cost) and let the model recall pruned content on demand

Distills long sessions into background observations and reflections, injects them as a system-prompt section, and renders compaction summaries from prepared memory; includes a Settings page and source-backed recall.

Audit what context compaction actually lost. Reads the raw session event log and checks session-constraint survival (the standing instructions the user gave — do-not-X, must-Y, only-in-Z — and whether the summary still carries them; arXiv 2608.11242 measured ~17% survival across compactors), summary hallucination (paths, versions, ports and identifiers the summary invents that were never in the history it replaced), ghost references ("see above" pointing at removed text), token accounting (a summary nearly as large as what it claims to free), and unaccounted prunes. Four agent tools including a cross-session sweep that ranks conversations by how much compaction hurt them. Deterministic — no LLM, so a wrong answer is a rule bug with a name.

See what a conversation actually costs: how many model requests, how big the system prompt and tool table are, cumulative tokens, estimated spend, and how far you are from automatic context compaction. Reads the local session log only; never calls a model, never adds a word to the system prompt. Token and spend figures are estimates from character counts, not provider usage.

Budgeted verbatim user anchors that survive DeepSeek Harness compaction

Percentage-only compaction and model-configurable GPT and DeepSeek token counting for DSH

Plain-language trust dashboard for the dsh trust family: aggregates the nine channels (injection interception, write denials, egress blocks, custody-chain verification, tool-surface and prompt drift, leftover scans, forgotten compaction constraints, profile composition) into one right-sidebar panel with a single status light and event cards that translate engine jargon into sentences a non-security user can act on; the read and egress engines run from the package's own npm dependencies so those channels work even without the sibling plugins, and absent plugins simply show as not installed.

Compaction integrity: directive lines (never/must/必须 style markers) extracted from the compacted range are verified against the summary — verbatim or by content-word coverage — and every constraint the session silently forgot is reported and written to a JSONL ledger; pinned constraints from config are checked on every compaction.