Plugins

Competition math (IMO/Putnam/USAMO/AIME) solved with a pure-reasoning pass, an adversarial verifier in a fresh subagent context, and calibrated confidence output (high / medium / honest "no confident solution"); optional LaTeX-to-PDF rendering.

Explicit model routing for Main, Subagents, web_search, and image generation, plus a catalog-driven Composer and Plan Review picker with Model, Effort, Context, Fast, and Thinking variants.

Three-layer local memory for DeepSeek Harness: L1 snapshot memory (MEMORY.md/USER.md) read on demand through the memory_* tools + L2 searchable knowledge base (facts + SQLite FTS5, Chinese search via sliding-window tokenization) + L3 import from Hermes/Claude Code/Codex/WorkBuddy and from approved Agent Hub documents (memory/global, memory/projects, accepted decisions) with content-hash dedup. v0.2.0 removed every automatic injection: memory enters the context only when the model calls one of the 12 memory_* tools, and data stays on the local drive.

LongCat-2.0 LLM adapter: 1M context, binary thinking mode, tool calling, and streaming over the OpenAI-compatible endpoint.

Grok-compatible web search with retained source pagination, Context7 and Exa documentation lookup, bounded page extraction, site mapping, offline research planning, and read-only diagnostics.

Cute token-usage pet in the session header: live context occupancy, per-session usage and breakdown, plus date/weekday, weather, 3-day forecast and severe-weather alerts, all theme-aware.

agentmemory for DeepSeek Harness (dsh): full memory_* tools, capture hooks, and context injection over the local REST server

Durable citation-thread learning companion for DeepSeek Harness conversations

Adds persistent execution contracts, recursive work graphs, critical clarification, and evidence gates for long or underspecified Harness tasks.

Zero-dependency MCP client bridge that connects stdio and HTTP MCP servers and registers their tools as mcp_<server>_<tool> for the agent.

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.

Incognito sessions: a standalone ephemeral sub-agent with its own preset (no parent context) and full toolset, draggable floating window, burned to zero traces on close.

Cross-session semantic memory for DSH: indexes session history locally with EmbeddingGemma 2 embeddings and recalls relevant past context, injected as clearly-framed untrusted context. No npm runtime dependencies; synthesis is opt-in.

Reversible context folding plus external memory for DeepSeek Harness: tool output collapses into expandable lines, session facts become auditable records.

Prune low-value exchanges or single messages from a session: select them inline in the chat, drop them from the model-visible surface and the transcript, undo in one step.

Read-only heads-up display over a DeepSeek Harness conversation: one state word in the centre, fixed seats around the ring for terminal, code, search, documents, web and a catch-all, and a sub-agent ring when more than one agent runs.

Composer-side helpers for the DSH Web UI: next-message prediction shown as grey ghost text that Tab fills in, a one-line per-turn recap above the input (placeable below instead) that never enters the model context, a right-sidebar panel for follow-ups from past sessions, recap history and side questions with their own model picker, and config-driven reply length, style and actionable framing.

Inspects the model's streamed reply before it is committed and either discards it or holds it for review when it reads as a refusal, so that text does not stay in the session context. Detection combines built-in phrase rules (an opening refusal, a capability denial, policy wording, first-person intent, refusals quoted in reasoning) with a plain-text word list and optional regular expressions. Interception can sit at the llm/stream waterfall, at the HTTP fetch layer before encryption (not TLS decryption, and it cannot tell which session a call belongs to), at both, or in a preview mode that shows a transient preview while the original chunks stay in memory until the whole reply has been checked. The same plugin also guards tool calls on tools/pre-execute: a command that deletes a drive, filesystem or user-profile root, formats or repartitions a volume, empties the Recycle Bin, fetches and executes a remote script, or writes to a raw device is denied, and a recursive delete outside build directories is recorded by default. Each tier can be set to deny, ask, record, or off, and the disk guard has its own switch. Both halves are heuristics: they can miss phrasings, the disk guard sees the arguments of a call rather than what a running script does, and they fail open.

A slim tool strip beside the conversation: one-line notes into a sticky note, a follow-up or a reminder (the sticky-note and follow-up modes need the dsh-sticky-wall and dsh-followup plugins installed); selected text quoted, saved or handed to the model; saved prompts into the composer in one click; a recap of this thread or the previous one; a read-only panel for context use, tokens, subagents and open follow-ups; plus custom entries that open a path or run a command.

Windowed handoffs on top of DSH native compaction: each compaction appends a bounded window message (goal, todos, recent user instructions, notes), with session-private notes, paged history recall and a /ctx switch; notes and history tools appear after the first rotation.

Runs the standalone opencode2api backend (jasonxu114514/opencode2api, Windows x64/ARM64, no Docker or OpenCode CLI) with DSH and on exit stops only the process it launched; its settings card installs, updates or rolls back the backend from checksum-verified GitHub releases, copies the initial admin password and syncs /v1/models into an opencode2api provider with context, vision and reasoning efforts, re-syncing every 10 minutes. Frozen archive: 26.10.31 is the last version, and the author points to dsh-magpie as the replacement.

Estimates what each conversation costs, with built-in official prices from 16 vendors (DeepSeek peak/off-peak hours and long-context tiers included), a breakdown panel with a currency picker under the composer, and each conversation's cost in the sidebar.

Adds a right-click command "通过 dsh 打开" (open with dsh) to the Windows Explorer right-click menu for a folder, a folder's empty background and any file: it launches DeepSeek Harness on that folder, creates a new session every time, and puts the clicked file into the new draft as an @ reference. Ships a packaged-COM menu for the Windows 11 modern menu plus a registry entry for the classic menu; first install asks once for a certificate trust (native/trust-cert.bat).

Layered context compression for DeepSeek Harness: folds old turns into model-written L1–L3 summary blocks, maps what each layer holds, and wakes the model when the context crosses a size line or keeps growing. By default it requires DSH's experimental Session Search plugin (session_event_read / session_event_search); without it the tools stay disabled unless requireSessionQuery is turned off.