Plugins

Analyzes usage and cost across models, providers and workspaces over any date range: hourly single-day and daily cross-day trends, a 53-week heatmap, cache-hit and model/workspace shares, plus request logs, CSV export and a ledger that survives session and workspace deletion. Highlights: multi-dimensional usage analysis, model and workspace insight, configurable cost and balance, full history with incremental sync, and auditable day-to-day operations.

A better fold bar: automatically folds tool calls and Think segments into step fold bars, and collapses each finished turn into a single turn header with live metrics (duration, TTFT, token, tok/s, cache hit) — only the final summary stays visible; customizable leading icons with animated poker cards built in (suit-cycling card animation while running, a deck when collapsed, a fan when expanded), a settings row for the official fold mode, compatible with both new and old DSH versions. Zero intrusion into DSH source.

Connects Huawei Cloud CodeArts to DeepSeek Harness with built-in OAuth2 sign-in (PKCE + DPoP) or a pasted token, after which CodeArts models are available in the DSH conversation window. Reads per-account credit balance, expiring credits and daily check-in from the CodeArts credit API, counts requests and tokens locally, and shows a per-account credits popover under the chat input box; the credits line and the upstream queue hint can each be turned off in settings.

Turns a one-line DSH draft into a complete task brief (role / goal / constraints / output format). The engine is a byte-verified port of linshenkx/prompt-optimizer (three modes, 10 templates identical to upstream); history is consent-gated with version chains and structural comparison; every run reports its token usage. Nothing is written back or sent without a click.

Expert-mode agent preset for DeepSeek Harness (v0.9.2, npm dsh-expert-mode, bilingual EN/ZH): a chief coordinator plus 17 domain-expert subagents with automatic task delegation. Features: taskboard scheduler (file-system task state machine pending/ready/running/done/failed, dependency DAG, atomic claim, retry, crash recovery), quality gates (5-stage pipeline for high-risk tasks: requirement clarity, implementation, verification, independent review, integration, with 2-round rework limit), Five-Anchor constraint (review/convergence/anti-drift/collaboration-check/resource-awareness per turn), Near-distance Guidance (identity/task/output template per expert), progressive disclosure (~28% token savings), expert persistence, inter-expert file message bus (direct P2P comm, zero coordinator relay), cross review, experience pool, fast-track for simple tasks, fault recovery with auto-retry. Experts: data analyst, copywriter, legal review, product manager, frontend, UI/UX, architect, social media ops, growth hacker, quant finance, finance, backend, DevOps, database, QA, security.

Evaluation harness for DSH plugins: YAML cases drive real headless agent runs, assert on tool calls, args, results and token usage, with a baseline gate for CI regression.

DeepSeek Harness's own fast mode: the same model on the same reasoning effort, fewer output tokens, all prompt-side. Only the output-discipline switch is on by default (late mode, active from step 4); every other switch is off and can be enabled on its own; turn on the tool-output trim switch when tool output is long. Measured on one machine, not a leaderboard: 30 task pairs with long output at the trim switch 0.756x wall clock and 21% fewer output tokens; 24 pairs on 6 real pallets/click PRs 0.805x and 23% fewer. On official SWE-bench Lite instance definitions with a self-built harness it does not show a gain (8 instances, three independent runs pooled to 72 pairs: 0.933x wall, 0.903x tokens, wall still crosses 1; single runs of the same configuration differ by 30%), and stacking switches on top of the trim switch is worse than the trim switch alone. Numbers and repro in the repo.

Reuses the CodeBuddy CLI sign-in to bring its models into DeepSeek Harness, with a settings card showing account, token expiry, and remaining credit.

Usage tiers and shareable stats cards: 🐟→🐳 ranks by monthly token burn with a locally-computed estimated percentile, precise pricing for 46 models across 6 vendors (including size-tiered Chinese pricing), backfill of pre-install sessions, and old-vs-new comparison across the 2026-08-17 rate change. Local-only, no telemetry.

Settings page for per-model capabilities: reasoning-effort levels with their wire spellings, input modalities, and token capacities, written as declarations or sparse catalog overrides.

One-command VPS deployment for DeepSeek Harness: a zero-dependency login gateway with automatic HTTPS that proxies the native web UI, a token-protected setup wizard, and one-click DSH upgrades with automatic rollback; as a plugin it adds Settings → VPS Deploy to install DSH on (or remove it from) a VPS over SSH from a form, plus version, upgrade, gateway status and a QR code to add DSH to a phone's home screen.

Header quota badge plus a usage-monitoring tab: live balance/quota remaining and reset countdown per provider, with requests, tokens (input/output/cache-read/cache-write), cache hit rate, estimated cost and average TPS tracked per session, provider, model and client, stored locally in SQLite with cross-device import/export.

Host tools for Texas Hold'em on kuroneko.chat — seat token, hosted/manual control, legal actions, and act loop for one GitHub identity.

Settings card for the DSH LLM auto-retry engine: tune retry count, backoff and jitter live from Settings -> General, with configurable retryableCodes chips - defaults add INVALID_REQUEST so OpenAI-style HTTP 400 reasoning_text errors in thinking mode get retried automatically.

Local-first four-bucket Token observability: persistent session/provider/model/day dashboards, trends, budget and anomaly signals, public-rate estimates, safe aggregate exports, and explicit opt-in usage or session-trajectory analysis.

A right-side floating panel and /context command for DeepSeek Harness — a live donut chart of context token usage, allocation, and estimated cost.

Plays Inception Mercury's diffusion (denoising) stream live in the dsh Web UI answer bubble: for routes that support it, the request is re-sent with `diffusing: true` and the real frames are replayed from noise to final text, while only the converged answer reaches the session log. Support is detected per provider/model with a probe that is rejected before any token is generated, a floating per-model toggle turns it on, and attachments, auxiliary calls or unsupported routes fall back to the original adapter. Replaces the assistant-message renderer for all models while enabled.

Plugin coordination layer: plugins declare their capabilities, commands route to the owning plugin without a model call, tool schemas load on demand to cut prompt tokens, run history consolidates into long-term memory while idle, and failures are classified before any retry.

Usage and cost view for DeepSeek Harness: aggregates real token usage from existing session logs, locks each cost at the price of its event time, and shows month and today spend against a budget from a sidebar card or an input-bar pill.

Wraps work in named tasks and, once a task closes, folds its whole span of messages into one summary at the next step boundary; the original content can be restored on demand.

Live token stats under the DSH Web composer: streamed tokens/s, output token and first-byte latency, plus per-step estimate-vs-actual reconciliation using the official BPE tokenizer, to tell a slow stream from a stalled one.

Cross-session memory with zero-token deterministic recall (bigram-Jaccard fused with FTS5 BM25 via RRF, Chinese-aware tokenization), reinforcement that strengthens repeated observations, and procedures crystallized into SKILL.md files refined in place through an active-stale-archived lifecycle with reversible archive and rollback. Ingest and disposal each expose a user-chosen autonomy tier; conflicts and high-importance writes always go to human review, and nothing is ever auto-deleted. Requires Node 22.5+.

REST + SSE gateway exposing running Harness sessions to third-party clients: API-key auth, token streaming, workspace grouping, and adopting GUI sessions to keep chatting.

Registers two MCP servers that drive the installed AutoCAD and SolidWorks desktop apps over COM, plus two on-demand skills. Windows only, Python 3.11+; the pinned upstream servers install themselves on first load, and only the sides whose CAD application is registered get a row. Each server advertises two meta-tools instead of its full catalog (a discovery shim here for SolidWorks, upstream DISCOVERY_MODE=search for AutoCAD), taking the schemas loaded before the first request from 23,404 to 256 tokens and from 83,732 to 365 respectively.