Plugins

Forgetting-curve in-place dialogue compression for DeepSeek Harness: recent turns stay verbatim, middle turns are structurally dehydrated (zero-LLM), ancient turns fold into a bounded summary pair — unbounded conversations on a bounded memory budget.

Nightly self-improvement for your DeepSeek Harness agent: harvest past sessions offline, replay recurring tasks on your own API budget, and consolidate validated skills behind a held-out gate. Brings microsoft/SkillOpt's SkillOpt-Sleep engine to dsh as native tools plus a bundled skill - no weight training. Requires Python 3.10+ and the SkillOpt-Sleep engine (pip install skillopt).

Enhanced auto-compaction defaults for DeepSeek Harness agent presets: apply tuned threshold/retention budgets and per-model policies with one command, idempotent with backups and rollback.

Token cost tracker for DeepSeek Harness -- track usage, calculate costs, budget warnings

Host-level proactive wake-ups for DeepSeek Harness: let the model schedule its own alarms (once / interval / cron with random jitter), wake cold sessions on time, gate model-initiated follow-ups with quiet hours and a daily budget, and reclaim silent wake

Project documents and long-term memory as plain Markdown in a dedicated Obsidian vault: six mem_* tools (search, read, write, log, brief, admin), one budgeted recall injection at the first step of a session, and automatic distillation of completed turns. Obsidian does not need to be running, and the vault stays readable with the plugin uninstalled.

Universal Model Context Protocol (MCP) bridge for DeepSeek Harness: connects agents to stdio/sse/websocket MCP servers with a 10k dormant scale engine, sub-millisecond BM25 tool search, dynamic token budgeting, and a real-time web UI dashboard.

Host-level proactive wake-ups for DeepSeek Harness: the model schedules its own alarms (once, recurring interval, or cron, with optional random jitter), cold sessions are woken on time by session resume, model-initiated follow-ups are gated by quiet hours and a daily delivery budget, silent wake turns end with proactive_reclaim and fold out of model context as tombstones, declared schedule files glob-sync JSON schedules into host alarms idempotently, min_idle_seconds defers wakes until the destination session quiets down, and a web panel (settings section plus per-session tab, SSE live refresh) manages alarms and wake history — the heartbeat engine for AI companions and roleplay characters with their own daily routines and world evolution.

Merges other sessions' context into the current turn: evidence-weighted conflict resolution, retraction of claims whose premises were retracted, trust labelling that resists cross-session prompt injection, and budget-bounded packing. Ships a benchmark.

Pre-visit due diligence and company research in DeepSeek Harness for customer background checks and risk briefs, with Qichacha MCP tool-call budget approval and entity confirmation.

给每次 AI 对话装一根 Token 保险丝：为每条用户消息设置独立总预算，统一约束输入、缓存、思考与回答。A per-conversation total-token safety fuse for DeepSeek Harness, with a whale slider and automatic reset on the next user message.

AST-based context compression, test log filtering, and token budget guard for DeepSeek Harness: strips redundant syntax, prunes noisy log lines, and enforces a token ceiling so the context window never overflows.

Cost panel for DeepSeek Harness: meters every request from session events, prices it against a built-in catalog of 136 models across 16 providers, and shows the current model, account balance, session, today and month spend plus a monthly budget bar next to the composer. Bilingual zh/en.

Read-only Grafana tools that query metrics through the data source proxy: instance health, data source listing, instant and range PromQL queries, current alert state, and provisioned alert rules. Range queries downsample to a point budget; an explicitly supplied step that would exceed it is rejected rather than silently changed, so the model never receives data at a resolution it did not ask for.

Resumes turns interrupted by a DSH host restart: hooks agent activation (once a session is live), reads the crash-repair turn/end, and sends a continuation that requires verifying real-world state first rather than assuming a side effect succeeded or failed. Permanent failures (AUTH, QUOTA, context overflow) are skipped, with a cross-restart budget, adaptive backoff and a minimal-context payload.

Panel to edit dsh's native AGENTS.md from the web settings page: every layer in the instruction chain is clickable, three activation scopes, and a visual budget meter.

Model-callable compact_agents tool that force-compacts context in every live session (main session, ordinary sub-agents and AgentTeams members alike), queues busy targets and compacts them when the turn ends, reports the shadowed node count and estimated tokens per target, and ships a browser form for the trigger ratio, retained ratio, controlled-phase output budget, auto-continue budget and an end-of-turn pre-compaction ratio that compacts near the trigger line once a turn ends, so a reply no longer opens with a wait for the summarizer.

Agent billing and usage observability for dsh web: session-log direct-scan recording (zstd, per-event timestamps, watermark + lastSeq double dedup), multi-account rate profiles per provider and model, per-window live badge reading the window session id via React fiber, full glassmorphism dashboard with 14-day sparkline, budget alerts, relay-station balance queries and a 332-model official price library.

Alibaba Bailian (DashScope) model catalog preset + auto-adapter for DeepSeek Harness — reasoning effort levels, thinking budgets, context windows, auto-adaptation of existing Bailian routes, a configurable context ceiling

Live token cost for every chat in the DSH Web UI: the chat, its subagents and the whole session tree, priced from a bundled models.dev snapshot of seven providers and 139 priced models, with cache read and write rates wherever the provider publishes them and DeepSeek from its own peak and off-peak table. Every finished answer carries its own price in the transcript, and a chat that is not currently open is priced from the log instead of disappearing. Spend is costed per interval from an append-only JSONL ledger in the project folder, so logged work keeps the tariff it was billed at. On top of it: budget-aware work plans packed against a money limit, routing scenarios that show which models are worth connecting, a generated plan document in English, Chinese or Russian, and a warning when spend belongs to no plan unit.

Tiered persistent memory (user/env/project) with per-tier char budgets and semantic query routing; full layers refuse writes instead of silently evicting.

Team cost captain: meters every model call (subagents included), attributes each call to a member by workspace rules, prices it with the built-in DeepSeek 2026-09 catalog (peak/off-peak, CNY/USD) or user overrides, and appends it to a local JSONL ledger. Adds hot-reloadable member budgets, 80%/100% webhook alerts, optional over-budget enforcement, a team_cost_report tool with CSV export, and a settings-page team dashboard served over a Typert RPC service.

Persistent project memory, event-sourced task lifecycle, context budgeting, and cooperative multi-agent handoffs.

Local token usage and cost dashboard for DeepSeek Harness: four-segment token breakdown (miss/cache-hit/cache-write/output), cache savings estimate against the miss/hit price gap, per-session drill-down with context-growth curves and anomaly flags, global filters with shareable URLs, monthly budget progress with month-end projection, and day/month aggregate fast-path queries (O(days)) with incremental rescan.