Plugins

Session-scoped delivery receipts for local artifacts: compare current file hashes with a recorded baseline, keep agent claims separate from explicit user acceptance, flag changed or unavailable files, and export a Markdown receipt through a tool, slash command, and conversation panel.

Personal project assistant for bounded project context and memory, explicit task handoffs through an existing WeChat bridge, reminders, and source-linked summaries. The WeChat path needs the separate pan17/dsh-wechat bridge and your own scan-code binding.

Owner-facing AI studio for DeepSeek Harness: project roles, task acceptance, worktree-based agent runs, host verification, independent review, decision records, and human-triggered merges.

Read-only Automations tab in the conversation view. It aggregates what other plugins are scheduled to run: last run time, next run time and a rough token-cost estimate, plus today's total spend and remaining balance. It owns no tasks and starts nothing.

Native roundtable panel with a product manager, engineer and critical user, two discussion rounds, follow-up questions, votes, answer retries and summaries.

Douyin patent and trademark complaint workflow for DeepSeek Harness with a left-sidebar task panel: checks user-supplied links, materials and organization identity in a dedicated browser, records each submission state, queries platform history read-only, and shows local task counts in the sidebar; never discovers targets or submits without explicit authorization.

Turns a plugin repository's existing screenshots into a 6-second silent looping hero reel: a VP9 webm that keeps real alpha, a matted gif, and a cover frame, plus a manifest with per-scene holds and sha256s. The timeline is exact by construction (integer holds plus transitions sum to durationMs, and the last scene transitions back into the first so t=0 equals t=duration). Zero runtime dependencies: PNG decode/encode on node:zlib, an SDF card compositor, and ffmpeg as the only external binary, discovered at runtime. Ships a conversation-view panel (pick, reorder, generate, copy the README markdown) and a CLI with a storyboard.json contract.

Local GPT-SoVITS voice for DeepSeek Harness: boots the engine with DSH and kills it on exit, speaks only the line marked for voice, and ships a default voice, so you only need to point runtimeDir at a local GPT-SoVITS install.

Dictation that ends up in the repository instead of the chat log: hold the button (or import a WAV), and the plugin archives voice-memos/<id>/ with audio.wav, a rendered waveform.png, memo.md (frontmatter, per-segment timestamps, ready-to-paste quote blocks) and meta.json. Audio is resampled to 16 kHz mono in the browser before it is sent. Transcription stays local — the harness speechToText service with silence-split windows, or a command worker such as faster-whisper — and when no engine is available the memo is still filed with an explicit status and reason rather than a fabricated transcript. Ships a conversation-view 口述 tab with click-to-seek playback and per-segment quote copy, three agent tools (voice_memo_save / voice_memo_list / voice_memo_read) and HTTP routes. Zero runtime dependencies, 52 node:test cases, real screenshots in docs/.

Prove that a local image edit left every pixel outside its mask untouched. Original, generated and mask PNGs go in; a greppable `MASK_PIXEL_LOCK=PASS|FAIL|EMPTY` verdict, a difference heatmap, per-channel statistics, a bounding box and the largest changed pixels come out, plus a self-contained HTML report and `report.json` with a SHA-256 file hash, a canonical protected-pixel digest and a proof digest binding both to the policy. Supports edit and lock mask polarities, a mask threshold, tolerance and alpha-ignoring; PNG color types 0/2/3/4/6, bit depths 1-16, all five filters and Adam7. Zero dependencies, local only — no model call, no network, no upload.

Windows NapCat deployment guide and read-only loopback port checks for DSH, with a user-initiated administrator launcher.

Archives DSH conversation output into SiYuan notes: a single siyuan_archive call files content under `<archive path>/<year-month>/<type>/<date>_<time>_<topic>_<content>` with a metadata table and auto-maintains an index document at the notebook root; 6 siyuan_* host tools (list notebooks, full-text search, read document, archive, delete block, remove document) and a Web settings page for base URL with connection test, API token in the host credential store, archive target and title template, plus notebook/document-level read/write/delete permissions with default deny and child levels able to only tighten.

Windows desktop control for DeepSeek Harness agents: window and UI Automation automation, local Windows OCR, direct browser reads over the Chrome DevTools Protocol (DOM text, JavaScript, selector clicks), humanised mouse and keyboard input, browser tab management, and a live control panel. It probes the machine on first run, writes its own config, and degrades gracefully when a capability is missing. Windows 10/11 only.

Study companion for your materials with guided practice, evidence-grounded feedback, Leitner spaced review, 18 teacher tools, and a Chinese/English student panel using the current DSH session model; requires DSH >=0.2.1-rc.2 <0.3.0-0, Python 3.11+, and Python runtime dependencies.

Connects DeepSeek Harness to MCP Bids public US federal opportunities and connected TED notices, with source-linked document tools and personal API-key authentication.

Mounts the pen.dev desktop app MCP server in a DSH session so the model can read and edit the .pen document open in Pen through the get_app_state, read_skill, execute, get_style and browser tools; finds the server binary by probing the usual install paths, and adds a pen-dev skill plus a listener that turns pen.dev plan-limit failures into a short bilingual notice. Unofficial, and ships no pen.dev binaries.

The composer predicts what you are about to type. Inline ghost text sits on the caret, measured against the editor's own font and line box, and Tab turns it into real text (Esc ignores it). The prediction is one short call through the session's own model router, seeded with your draft and the last few turns of the conversation, so an empty composer also gets a next-prompt guess written in the user's voice; suggestions that merely restate the assistant are dropped. Host side is one HTTP route with sub-path endpoints (loopback-only, 64 KB body cap, 2-minute result cache, 20-second miss cache, one in-flight call per session); client side is one 0-height overlay registered on conversation.input.overlay, with insertion guarded by the composer's draftRev CAS and a setDraft fallback. IME composition and slash commands are skipped. 16 node:test cases, zero runtime dependencies, real screenshots in docs/.

Content-catalog guard for WeChat mini-programs: validates catalog.json so every activity price is an integer amountFen, every template renders prices through money(), required fields / duplicate copy / image paths are checked, and the result is an issue list plus a unified diff that can be applied with patch -p1. 29 rules over five groups (catalog shape, required fields and uniqueness, amounts, rendering, images). Three fixes are automatic (legacy yuan amount, string Fen, absolute image path) and preserve the file's own indent, EOL and trailing newline; everything else is reported with a hint instead of being guessed. Dry-run by default; apply re-checks the written file so the report describes what remains. One implementation serves a tool, a conversation panel and a CLI with exit codes 0/1/2. Zero runtime dependencies, no model call, no network.

Auditable token-cost ledger for DeepSeek Harness: wraps the llm/stream waterfall and records every call with provider, model, input/output/cache-hit/cache-write tokens, usage source and integer nano-CNY cost from an explicit price table. Aggregates by hour, day and session; exports JSON or CSV with a no-write dry-run preview; keeps a manually entered official platform bill in a parallel column with actual-minus-estimate diff. Unknown provider/model routes stay explicitly unpriced. Zero runtime dependencies; no balance query, notifications or animations.

Version-adaptive tooling for DeepSeek Harness: it reads the running DSH version from the plugin manager, mirrors app-boot plugin-compatibility rules (engines.dsh first, then @deepseek-ai/dsh* peers, prerelease-inclusive comparison, exact-version exemptions) to score every installed plugin, and statically flags source still using APIs renamed or deprecated between v0.1.6 and v0.2.1. Ships a conversation-view panel, an agent tool and HTTP state/scan/report routes. Read-only, offline.

English port of Windows Computer Use for DeepSeek Harness: 10 desktop tools (computer, computer_batch, open_application, windows, ui_elements, clipboard, switch_display, app_skill, request_access, list_granted_applications) with UI Automation, per-app access control, English settings UI and mini progress card overlay, Esc pause/resume, and DSH 0.2.1 unpacked koffi.node auto-discovery. Fork of copylee711/dsh-computer-use; it reuses the same plugin id and tool names, so the two cannot be installed together.

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.

Codex-style drag-to-reference for DeepSeek Harness: drag a conversation row from the sidebar onto the chat area and it becomes a native session-reference chip in the composer draft, so you can keep typing around it before sending. Drag a workspace (folder) row instead and the draft gets that folder's path, so a whole directory can be handed to the model. The target session is matched by session id rather than by title, so duplicate or blank titles cannot insert the wrong conversation. The reference cap comes from the host (3 by default) and is read at runtime instead of being hard-coded, so a host-side raise is followed without changing the plugin. If the chip cannot be inserted, the plugin falls back to sending a reference-only message instead of failing silently.

One header button stops running turns, goals and background jobs and writes the plan to disk; after a restart the same button picks the work up again.