Plugins

A DeepSeek Harness page companion with animated characters, a clock-out countdown, estimated earnings, hydration reminders, quiet time and salary masking, with separate cards opened from the context menu.

Session Console: a sidebar dashboard over every conversation that shows run state, token spend, cache hit rate, context pressure and a heat score, resolves a plain-language sentence such as "how is the example project going" to the matching session, and dispatches that sentence as a queued task into it. Ships five agent tools (overview, find, tail, dispatch, cancel) over the same data.

Predicts the next user message as composer ghost text using the configured DSH model; Tab fills the draft without sending, with time-aware context and cached suggestions.

Context ring gauge in the DSH composer toolbar: view usage, open a breakdown panel, and run /compact from the panel, with a fireworks animation.

Hy-Memory 2.0 as a native DeepSeek Harness memory component: three-tier memory (personal / project / session), three scenes (chat / working / coding), automatic recall and capture with tool traces, a Settings > Memory panel, and MemHub interoperability.

Inject soulbridge context into DSH sessions and feed user turns back as observations.

Runs Hindsight long-term memory on your own machine: after a one-time uvx --from hindsight-embed@latest hindsight-embed profile create server setup (~450 MB of Python dependencies) and with your own LLM key, the plugin manages the local memory server itself (health probe, start, stop, logs), registers ten hindsight_local_* tools for status, page search, page read, recall, remember and reflect, injects a knowledge-page index once per session, writes each finished turn back into the bank, and extracts facts with your own LLM key. Do not enable it alongside the official @vectorize-io/hindsight-coding-agents plugin: both read the same bank and would inject duplicate context. At session end it writes a handoff snapshot that the next session injects as previous progress. The data directory is configurable (dataDir) and scripts/relocate.mjs moves an existing install to another drive, creating directory junctions where the server hard-codes its paths.

Read-only persona injection for DSH: when a session opens inside your workspace, your persona files and the latest diary entries are prepended to the system prompt, and those source files are never modified. It keeps a short cross-session feed of recent messages under <root>/live/feed.json and can write a QQ hand-off file at <root>/live/web-for-qq.md; nothing is uploaded. Variable blocks (clock, recent actions, other sessions) are delivered as a per-step context message, so the prompt prefix stays byte-stable and keeps its cache.

Cache-first context compaction with an optional adaptive economic policy, observation-only by default, without cache-hit or cost-saving guarantees.

TeleAgent (China Telecom TeleAI "Stellar" agent) model provider for DeepSeek Harness with the working v1 gateway signing: two-round HMAC-SHA256 keyed by the login JWT signature segment, which fixes the 40104 rejections the older local-v1 signing produced. Signs in through a dedicated browser window over the DevTools protocol — the plugin intercepts the login callback the same way the official desktop app does, so nothing has to be copied by hand — and also accepts a pasted token. Ships points balance, daily check-in, an in-composer account chip, and a configurable context window (100K by default, 16K–2M, since the gateway declares no catalogue and the official client ships one shared 100K preset for all three tiers). Cannot be installed alongside anye-99/dsh-codearts-auth: both register the same teleagent provider, and this plugin reuses its credentials and derives its SSE/tool-call parsing from it. It also registers the same teleagent provider as anbulang/dsh-connect-teleagent, so install only one of the two.

Right-click any selected assistant reply and add it to the composer as a removable quote chip. The selected text is delivered to the model as hidden context, so your message bubble keeps only your own words.

MCP server panel for the DeepSeek Harness Web GUI: the sidebar lists every @deepseek-ai/dsh-mcp-client row — including rows inserted by other plugins — and derives its connection state from the loader fiber, the mcp__<server>__* tools still registered, a TCP probe of HTTP endpoints and the last connection test; it shows the row's configuration, enables or disables it by editing only that row of the patch file, reconnects once after a failed attempt, and reports unavailable servers to the agent through a system-prompt note plus a call gate that refuses their tools instead of letting each call fail at the transport.

Open delivered files, changed-file rows and inline file mentions in the system default app with a left click; right-click takeover is configurable: all menus, only where DSH has none, or never. Desktop app only: the Web client has no system-open capability, so clicks stay with DSH.

Shows elapsed time, tokens, tokens per second and a shrinking countdown estimate in the chat header for the running turn, and estimates output tokens, context size, cost and duration above the composer before you send.

Adds contextual guidance for explaining technical misconceptions in conversations.

File-state cache skill for coding agents: one explore call replaces ls+grep+read, reads remember they already happened, and edits return a diff plus the exact changed line range instead of a full re-read. State is a JSON cache on disk, so it survives across tool calls and restarts. Requires Python 3.8+ (standard library only).

Bundles the PrismSearch MCP server and its agent skill. Web search over Bing and Baidu with no account or API key, plus Google and Brave when API keys are set and DuckDuckGo, Sogou, 360 and Yandex when enabled in config; automatic engine fallback and an aggregate mode. Also image search, WeChat Official Account article search, and fetch_page, which returns web pages and PDFs as Markdown in segments.

Registers the hosted SpecStream MCP server with the harness MCP manager and keeps the API key in the credential store. The server returns live web data as structured rows, each value with its source URL, and a run starts only after the quote's maximum is approved.

Localizes the prompts DeepSeek Harness sends to the model — system prompt sections, runtime context, tool descriptions and parameter docs — into Simplified Chinese. Its Cordis API dictionary is derived from zjl1989-li/dsh-harness-zh-cn (MIT), which translates the same prompt layers at runtime; do not enable both at once.

Notes-backed context compression for DeepSeek Harness: replaces LLM-summarized compaction with a model-written persistent notebook plus recoverable raw history, so the rollover boundary itself costs zero model calls.

Keep conclusions alive across context compaction: the model writes a task checkpoint and the plugin re-injects it into the system prompt's first node — the span compaction never touches, so it survives any number of compactions. Four-tier injection (0/95/175/≤12000 chars) scales with session relatedness; checkpoints live in the plugin's own folder, never in your workspace.

Soft-branches a self-contained sub-question into a child session and returns only a bounded digest, so the main context grows by a short summary instead of the whole investigation.

Lists resumable Codex chats under the DSH composer, grouped by the concrete project each one belongs to, and inserts a mechanically extracted recovery context into the draft without sending it. Project ownership is resolved from the Codex project registry with path-boundary matching and labelled by how reliable it is: recorded session ownership, registered directory, unreadable metadata, or a directory registered under several projects (left unknown rather than guessed). Also reports when the stored autosave is behind the log tail, which usually means the turn ended without a stop. The chat list requires python3 >=3.8 on PATH; Windows is not supported yet, and without python3 the list stays silently empty.

Adds a relay_check tool that runs relaycheck against an LLM API relay to detect model swaps, misreported usage, truncated context and injected prompts; keys are read from environment variables.