Plugins

Text-to-image through a StepFun-compatible endpoint: mounts a no-npm-dependency Python MCP server behind an NDJSON bridge as mcp__simple_image__simple_image and returns the generated image URLs. Needs Python 3.8+ on PATH and STEP_API_KEY.

Configurable, provider-agnostic image generation: mounts the bundled MCP server behind an NDJSON bridge as mcp__image_generation__image_generate and mcp__image_generation__image_config_status, and registers the bundled image-generation skill. Needs baseUrl and apiKey; without them the MCP server mounts nothing.

DOCX to PDF without Microsoft Office, LibreOffice or a network: converts a Word .docx with the dxpdf Rust/Skia engine, repairs the schema-invalid OOXML that WPS Office writes, and restores the blank pages Word inserts at odd/even section breaks. Needs Python 3.8+ on PATH with pip install dxpdf lxml.

Route native DeepSeek Harness subagent_fork child sessions onto your Google AI Pro (Antigravity) subscription (Gemini 3.8 Flash, Claude Sonnet 5.5, Claude Opus 5.5, GPT-OSS 120B) with zero API cost, live SSE streaming, full tool calling, prefill cache warm-up gating, ~200ms HTTPS quota checks, and a chat header button + slide-out settings/effort drawer.

Brings McDonald's China official MCP (mcp.mcd.cn) into DSH. Paste a token you generated yourself and the plugin handshakes, calls tools/list, and registers every remote tool as a native DSH tool (mcd_ prefix) — upstream additions and removals need no code change, just a restart. Ships four always-on tools (mcd_status, mcd_test, mcd_tools, mcd_call) for self-checks and direct fallback calls, plus a settings page with a one-click connection test. The token is stored only on this machine in ~/.dsh/dsh-mcd-mcp.json (mode 600); it never enters the repository or the conversation context.

Private Windows DSH cold archives and guarded fresh-environment recovery. v0.7.3 classifies the known plugin Home directories `.dsh-paper-reader`, `dsham-move-backup` and `music` as captured plugin data, restores them as data-only, and resolves matching legacy pending-review entries during read-only planning; unknown or incomplete trees remain critical and blocking. Strict dependency, transaction and verify/commit gates remain. Verified ZIP integrity does not certify private business recovery.

A memory plugin for DeepSeek Harness: set its name, personality and character; it forgets only trivial things, keeps what matters, and marks and remembers what is special to you.

Gives a DSH agent five layers of persistent memory kept as plain Markdown files — persona, speaking-style rules, daily diary, session archive and open loops — rebuilds a system-prompt section from them within a minute of a file change, and adds three tools (memory_read, memory_write, memory_search) plus an importer for migrating an existing memory directory.

One Settings page for every DeepSeek Harness session: browse active and archived sessions by workspace, generate a title from the whole conversation for live or archived sessions, archive or restore, move a cold session to another workspace, and permanently delete an archived session (with its subagent descendants after confirmation).

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.

Adds a session_rename tool so an AI can retitle the DeepSeek Harness session it is running in.

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.

Puts a working delete-session row back into the Codex UI sidebar by providing the remote.workspaceRegistry.deleteSession service that @michengai/dsh-codex-ui checks for, unhides the per-session ... button its stylesheet hides, and permanently removes the session log, workspace records and projection cache, with subagent and forked conversations selectable in a family-tree dialog. Delete engine adapted from Lzcdebear/better-dsh-session-deletetool (MIT); both can be installed together (distinct route prefix).

Adds a Delete Session row to every sidebar session '...' menu (the official harness has archive/pin/rename/fork only, no durable delete). One-click Force delete ejects the session from the host live catalog, stops its background work, then removes the durable log, projection cache and registry bookkeeping - no ghost rows on any connected desktop web UI. Workspace files the session created are kept; conversation logs are permanently deleted. Third-party cloud mirrors (e.g. phone apps via Agents Anywhere) hold their own copies. Bilingual zh/en UI, trust-fenced local route. Requires DSH >=0.2.1-rc.1.

Automatically archives idle DSH conversations after a configurable number of days and clears the projection cache of archived sessions, without deleting transcripts or attachments.

Settings card that lists sessions whose history references attachment objects missing from the local attachment store, then backs the log up and replaces those dangling references with text placeholders so the session can send again. The repair script needs Node 22.15 or newer, and the card has been exercised on Windows only.

Export current or historical DSH conversations, including selected turns, to TXT, Markdown and searchable PDF through a Windows native save dialog.

MIT adaptation of H1a3x's prompt-template manager with default/workspace inheritance, direct per-Agent Preset template assignment or no injection, localized settings and revision-checked automatic saves. Verified on Windows Desktop RC5 and 0.2.2-rc.1; cannot be installed alongside upstream H1a3x/dsh-prompt-inject.

Claude-style dark and light skin for the DeepSeek panel of dsh-webchat: warm neutral palette, terracotta accents and Claude brand marks, applied as injected CSS with no guest DOM edits and no network requests.

Web search insights for the agent: it keeps a local log of what was looked up and which pages came back, and counts searches, returned pages, citations and pages opened per turn and across the session, with the numbers shown in a small entry under the answer and a right-sidebar panel.

Holds every final answer before delivery, has one independent minimal-context model call re-derive and check it, and releases only the verified text; the draft is never shown. Reviewed answers cost one extra model request and appear a few seconds later.

Writes the session's estimated cost into the composer token-usage pill, and adds a per-bucket amount plus an estimated total to the usage dialogs, using DeepSeek's official peak and off-peak rates in yuan.

Shows per-turn, per-session and per-workspace conversation costs in CNY, priced by DeepSeek's official tiers with peak/off-peak hours and per-request Search/Fetch billing.

Sidebar footer badge that shows whether DeepSeek is inside a peak-pricing window right now: a state dot, the current multiplier (x2 during peak, 5 zhe / half price off-peak), a countdown to the next switch, and a 24-hour schedule bar whose white line marks the current time. Peak windows are 09:00-12:00 and 14:00-18:00 Beijing time; weekends and Chinese public holidays count as off-peak. The holiday table currently covers 2026 only. Client-only, no network requests.