Plugins

Claude Code-style auto-memory, natively: typed memory files (user/feedback/project/reference) with a MEMORY.md index auto-injected into the system prompt (zero footprint when empty); memory_write/read/list/delete follow Claude Code write rules (dedupe-and-update, absolute dates, cross-links); workspace-scoped and user-scoped layers, cross-process file locking with orphan-lock recovery, plain Markdown on disk, no external services, no embeddings.

Aggregated web search provider that fans one query out to 11 engines (Exa, Parallel, DuckDuckGo, SearXNG, Brave, Tavily, Serper, Mojeek, DeepSeek, Anthropic, OpenAI) concurrently, URL-dedupes and merges the results. Ships a settings card for toggling backends, storing API keys in the credential store, and tuning concurrency and timeouts.

Codex-style review of session file changes: per-turn cards (optionally taking over the shell edited cards) whose totals are the turn net diff, a review tab with syntax-highlighted diffs that never wrap, real line numbers, open-at-line and a draggable split, a hover preview of the whole file with three lines of context around each change, one-click per-turn revert, and the on-disk record size with a clear button in the review toolbar. Records are sharded per session, with content-addressed snapshot blobs that dedupe and are garbage collected. Light/dark presets follow the DSH theme, typography follows DSH and installed skins (or any font on the machine, with a px stepper), and a settings switch injects a think-in-Simplified-Chinese rule into the global AGENTS.md.

Five tools for reading a video as raw material rather than as something to watch. video_outline maps a whole file as text and draws nothing at all: 4-60 equal windows, each with its cut count, speech share, intensity bar and tags, then the windows worth opening and the exact start/end for the call that follows -- zero image tokens. video_probe returns metadata and a reading plan without decoding frames. video_storyboard turns a window into one contact sheet plus a text index binding every cell to its timecode, cut count and frame time; cells whose text would be illegible at that scale are marked zoom, and dedupe global folds a picture that comes back after a cut. video_zoom re-renders 1-9 cells from a short window at a larger scale. video_read decodes only keyframes, OCRs each screen, and returns an outline plus \[mm:ss] text -- the path for lectures, slide decks and screen recordings, where the words are the content. It also reads a file with sound and no picture as a transcript (subtitle track, then DSH_ASR_CMD, then a transcript file you pass; long audio is cut at 45 s onto silence), composes six changed frames into one sheet when a file has no readable text at all, and can write the result out as SRT, WebVTT or text. Cuts, dissolves and one-frame flashes are found in a single ffmpeg pass; from ten minutes up a file is analysed on its keyframes only, 2.7x faster on a two-hour lecture; and a cell is never shrunk below a fifth of the source, so a 1440p sheet spends the same tokens on fewer, bigger cells. OCR is the engine Windows ships, tesseract with line boxes, or DSH_OCR_CMD. It takes a B station link (bilibili.com/video/BV..., a bare BV id, an av id or a b23.tv short link) wherever it takes a file: up to 300 seconds is read from the CDN in place with nothing written to disk, longer videos are downloaded once and cached within a 4 GB / 24 video budget, play URLs are cached for ten minutes, and a link that is not a single video says what it is. Needs ffmpeg, uses ffprobe when present, and has no other runtime dependencies. It is a wire-compatible reimplementation of SAOS-3624/dsh-video-storyboard: it registers the same video_probe / video_storyboard / video_zoom tool names, so the two cannot be enabled together.

Community search, similarity analysis, and install/dedupe/build-it-yourself decisions for the dsh-plugin ecosystem.

Aggregated web_search and web_fetch providers for DeepSeek Harness, sitting behind the built-in tools. Search fans one query out to AnySearch, Zhipu and Zhihu concurrently, dedupes URLs, fuses ranks with Reciprocal Rank Fusion and optionally reranks through SiliconFlow. Fetch is URL-aware: static pages pass through the built-in HTTP fetcher, dynamic ones fan out to Firecrawl and AnySearch extract and pick one body by quality. One shared circuit breaker skips an engine for the day after a quota or 429 signal, and a settings card keeps each engine API key in the credential store without ever showing its value.

Self-tending living memory in one local SQLite file — a nightly patrol dedupes, merges, decays and cross-links entries on its own, with seven-signal RRF hybrid recall (FTS5 + jieba Chinese-first tokenization, optional vector KNN, graph PPR), a typed knowledge graph, conflict detection instead of silent overwrites, and web telemetry panels.

Group installed skills and manage MCP servers from a left-panel UI: filter the model skill catalog by enabled groups (union-deduped) and toggle MCP servers independently.

Autonomic heartbeat for each root agent: silent pulses diff the session's todos, Agent Team board and goal against what the last pulse saw, and speak only on a signal — L1 report, L2 ask (ask_user_question), L4 escalate — never starting work. Each message carries a why-now line, an optional one-line physiology signature (time since your last message and how many times it has reached out today) and a compact task snapshot (800-char cap); a dedupe window, quiet hours, a daily budget, refractory/restart cooldowns and persisted vitals keep it quiet, while busy-safe followup injection never stacks. It shares one advisory speech floor with dsh-plugin-memory (~/.dsh/proactive-cooldown.json, 180 s) so the two never talk over each other, stays silent while a child subagent or teammate is still working, and surfaces a "subagents working · N" pill in the conversation header (polling pauses while the tab is hidden). v0.8 adds a diagnostics panel that explains recent quiet pulses and names the current speech-floor holder, plus panel controls for the arbitration window, presence suppression and the physiology line; its settings section carries a stroked heart mark (16×16, stroke 1.3, same spec as dsh-plugin-memory).

Voice input for the Web UI: a mic button that drives local VocoType offline speech recognition and auto-inserts recognized text into the composer (auto-deploy, dedupe, continuous dictation).

Microphone voice input for the composer: browser Web Speech API live transcription, dedupe/auto-continue, smart punctuation, language and auto-send settings.

Incremental file mounting with read dedupe: mounted line ranges are never re-sent to the model, on-disk changes invalidate and remount, with a Mounted Files tab and token-savings accounting.

Router and artifact gate for DeepSeek Harness: a per-turn skill router that ranks the live catalog, plus a write gate enforcing read-before-write, dedupe and a four-line audit on every file write.

Complementary web-search provider: every web_search merges the DeepSeek Search API (primary) with serper.dev Google results (secondary) at a fixed 2:1 ratio. Both sides are direct HTTP retrievals with no auxiliary model turn: URL-deduped sources, long DeepSeek summaries clipped into snippets, per-backend timeouts and graceful degradation to whichever backend answers.

Two-way TickTick/Dida365 (滴答清单) task sync via the official OAuth Open API: pull tasks across lists, create/update/complete/delete tasks, and one-shot title-deduped sync, with a web settings panel.

File dedupe and unify toolkit, shown as "小视频整理" in the DSH sidebar: exact matches by head/middle/tail sampling or full byte compare, near-duplicate MP4s grouped by duration and a configurable first-N-seconds frame window (default 10s, 1–5 fps, default 2; longer videos are compared on their first N seconds too), batch rename or move by extension, UUID, file size or video length, empty-folder cleanup under the selected sources, and MOV/AVI/WMV detected by file header and transcoded to MP4 keeping resolution, frame rate and audio parameters. Requires ffmpeg and ffprobe on PATH. Concurrent workers per task with a shared setting; one task at a time; per-group help dialogs; local-only API. Every task that moves files has one-click restore: per-group/target _dupfc_restore.json records, deleted files skipped and counted, files outside the record renamed to UUID and moved into resetFail; orphaned ffmpeg processes left by a crashed host are swept on plugin load.

Intercepts DSH tool results before the model reads them: dedupe cache for repeated calls, a declarative decision graph that replaces large outputs with structured digests, an injection gate, and an honest token ledger that includes a negative-return account for over-compression. Measured 1400-1600 tokens saved per hit; the cumulative ledger is currently negative and the README says why, including that it cannot skip model turns.

A DSH workbench for picking photos from local folders on macOS and Windows: local dedupe, filtering and grouping, AI grid triage and scoring with an image-capable model you supply (the plugin bundles no model, credentials or quota), then human review, calibration and export.

Feishu (Lark) IM bridge over a self-built app long connection: forwards inbound chat messages into a DSH session and posts the reply back to the same chat. Dedupe, allow-lists, ack text, dry-run. Self-trigger was split out into dsh-self-trigger.

Three-scope rule injector for DeepSeek Harness that re-publishes the rules into long sessions, with a loop-breaker, dedupe by content fingerprint, per-session persistence, and a settings-panel editor.

Keeps the NapCat QQ backend alive next to DSH: probes the OneBot 11 HTTP port at startup, launches NapCat when the port does not answer, then rechecks every 45 seconds, kills duplicate process chains, hides the console window, can create a logon scheduled task when a task name is set, and adds a settings page plus napcat_status, napcat_restart and napcat_dedupe tools. Optionally remembers the QQ password, behind a risk confirmation, so an expired fast-login token can be renewed without scanning a QR code. Windows only; without that scheduled task the launched QQ exits together with DSH.

Token usage workbench for token-plan users: folds assistant/message usage from the session log into per-day call-level detail (cross-session responseId dedupe keeps fork/resume copies counted once), converts local volume into plan credits against terms you enter (default 1M tokens = 0.2 credits, caps 600/2500/6000 for day/week/month, optionally scoped to selected channels), and shows a composer-dock readout plus a settings page with per-provider cost table, 7/30d bar and line trends and a rolling six-month heatmap. No credentials are read: the remaining percentage a provider keeps behind its control-plane API is replaced by a clearly labelled local conversion, and reference costs come from the public DeepSeek pricing page.

Long-term memory on top of mem0 and a local Chroma store: user profile and recalled facts injected through the system prompt, capture at session end, a nightly pass that dedupes, merges paraphrases and resolves contradictions, a persistent topic graph written as Markdown cards, periodic export to a git repository, and a full-page memory UI with search, bulk merge, snapshots and undo. Needs a manual Python venv setup plus LLM and embedding API keys before use.