Plugins

A-share holdings contribution card for DeepSeek Harness: today's P&L and portfolio return next to the Shanghai Composite, the top five contributors by P&L with each stock's strength versus the index in pp, and a 15:00 close-summary PNG export with the contributor list, intraday chart and a numbers-only summary. Holdings come from a local holdings.csv/holdings.json and quotes from the public Tencent market-data API; the plugin never reads cookies or broker credentials. Requires DSH 0.2.x; the card and the export are covered by a test suite that runs in CI.

Adds model speed tests to the DSH model picker and, on desktop, a plain-language "explain this error" button on failed turns with a user-selectable explainer model.

Adds a reversible “Collapse history” toggle to the session header with a light/dark status dot: it unloads the history already rendered by the page and keeps the latest two turns, then shows “History collapsed”; clicking again restores the saved window. A manual restore keeps auto-collapse off for the rest of that session; thresholds and request caps live on Settings → Long session.

Prompt enhancement for the DSH composer, with a model picker that lists the models the host actually offers: a Boost button beside the composer sends the current draft to a model on the Host, shows the rewrite in a before/after overlay, and replaces the draft only when you confirm. While boosting, the button glyph becomes a spinning arc; once a rewrite is applied the same button becomes an in-place Undo that restores the original draft and then returns to its idle state. Settings hold the enable switch, a provider/model choice rendered as a grouped dropdown built from the live provider list (including configurable providers that are not set up yet) rather than free-text fields, temperature, output limit, timeout and optional custom instructions. Empty or reasoning-only model answers are reported with their real cause instead of a generic failure, and the Host endpoint only answers loopback requests from the same origin.

Hospital-triage board for the DSH sidebar: sessions whose latest turn ended in an error (red = fatal, amber = 429/rate-limit) or is waiting for approval are grouped under an Attention tab, next to Working and Finished tabs, with ticking durations, dismiss with undo, unseen markers and EN/中文 auto-locale. An eye icon hides the session list (the header then shows Workspaces · N) and a Duty-gated chevron folds the nav bar; clicking a row opens the session across workspaces. Needs the zstd tool (zstdcat) on PATH. On-request log scan through a guarded same-origin route — no background daemons, zero core edits.

Resumable ACP subagent backend for DeepSeek Harness — persistent chat pool with session/load reconnect & mid-turn steering. Tested with kiro-cli.

Rider-Waite-Smith tarot for DSH with a right-dock table for the Celtic Cross, Waite's alternative and thirty-five-card methods, plain draws, and three- or nine-card layouts; the plugin handles drawing and placement, then turns each card into the conversation for model interpretation from Waite's Pictorial Key.

Sends a macOS native notification when a root DSH session turn completes normally: the title is the session title and the body a sanitized, truncated preview of the reply. Subagents, cancellations and errors are skipped by default and no conversation text is stored.

Persistent waits for file creation, file changes or process exit; the turn ends and the original session is woken when the condition is met.

Quote an earlier turn of the current conversation into the composer from the @ menu — by turn, not by selected text — so a chat that drifted can be re-anchored on what was actually asked.

Turns the chat composer into an auto-playing mini-game arcade while a task runs — snake, pong, flappy, ecology, horse race, runner, number rush and maze all play themselves with no input; focusing the composer restores the normal editor.

Explains Chinese internet memes to an agent: registers a meme_explain tool that returns the meaning, tone, origin and lifecycle of a sticker, which memes can and cannot be combined, occasions to avoid, and both a meme-version and a plain-version reply. Answers come from a local memory store with spacing-effect decay, falling back to a user-configured OpenAI-compatible vision endpoint (BYOK).

Appends one short user message before the next step when the model is circling: the same absolute path written three times, the same normalised tool error returned twice, or two steps since the last write with neither a verify nor a search. Verification is recognised from four tool names (`ops_bash`, `lsp_diagnostics`, `verify`, `run_tests`), searching from eight. Each signal fires once per turn and only after a new write or error, is silenced for the whole turn when you say not to search, and is skipped in subagent sessions. Ships the `stuck-gate` skill at bundled rank 600, so a user-level copy at rank 400 overrides its wording without touching code, plus a read-only gain panel at `/better-debug` that counts firings, sessions covered and whether the next tool call changed class; it does not estimate tokens saved.

Raise native desktop notifications from DSH itself for approvals, questions, finished turns, errors and sign-in prompts, attributed to the DeepSeek Harness app with click-to-open; the chime is synthesised in-page and can be picked from presets or authored as a note sequence in Settings.

Native desktop notifications for DSH on Windows, macOS and Linux. While the DSH window is in the background, permission requests arrive as system notifications; where the platform supports notification actions, Reject and Allow-once buttons return the decision to the host approval waterfall, and returning focus or a timeout hands the request back to the in-app panel. Approvals and questions also play a separate input cue that is independent of window focus. Turn-end, question and error reminders, notification sounds and the cue are configured on a Notifications settings page.

Intervals.icu-only endurance-coaching skill bundle that turns Intervals.icu training data (activities, intervals, streams, training load CTL/ATL/TSB/rampRate, HRV, resting heart rate, sleep in-bed hours, eFTP, power curves) into concrete training decisions: readiness gating, VDOT pace and heart-rate zones, periodized weekly plans written back to the Intervals.icu calendar as events after confirmation, session reviews, race-day pacing and fueling, and weather-adjusted workouts. The bundle registers the endurance-coach skill in the profile, ships deterministic helper scripts, reads an athlete profile created by its init script, and requires the agent to state when data is unavailable instead of inventing it. It needs a separately installed Intervals.icu MCP server for training data and Python 3.10+ for its bundled scripts; with neither it works only from numbers you type in.

Host-side HTTP bridge on 127.0.0.1:39217 with a Bearer token stored at 0600: external processes can create real DSH Desktop sessions that appear in the workspace sidebar, send prompts (queue or steer), read state and the last assistant reply, list workspaces and sessions, and long-poll until a turn finishes. Zero runtime dependencies; fixed loopback port; no outbound network; message bodies are never logged.

Elysia's memory plugin for dsh. One shared, human-editable MEMORY.md across every session: two-disk mirror, history snapshots, cross-process write lock; entries tier between the main file and a searchable archive by hit frequency; retrieval is keyword co-occurrence plus BM25. Nine sections (persona, requirements, corrections, lessons, methods, requests, sessions, method library, ingested docs) plus a dedicated self-identity section for a character's own self-cognition, injected above the persona. Ships an opt-in memory pack (packs/elysia: 61,903-char identity + 2,482-char persona, distilled from 1,696 officially-confirmed voice lines and checked against a 112,175-line game-dialogue corpus) that is off by default, backs up before writing, only touches identity/persona, and can be uninstalled. Also a skills library and a sticker library (you import your own images; none are bundled) that sends a fitting sticker at the right moment. On by default: a text bubble with her avatar after each reply, which costs one extra small model request unless the agent already announced (voice.bubble: false turns it off). Opt-in voice playback needs a local GPT-SoVITS engine (about 27 GB, installed separately, NVIDIA GPU required; the plugin downloads about 200 MB of weights) and is Windows-only; the automatic second-drive mirror is Windows-only too (elsewhere set mirrorPath, otherwise there is no mirror).

Long-task continuity for DSH: every turn end writes a breakpoint (what it did, which tools it used, and — when the turn did not finish cleanly — what is still unfinished; the manual continuity_breakpoint tool can record the next step), injected as a snapshot when the same workspace starts a fresh session so work resumes instead of restarting. It also stores SHA-256 fingerprints of persona or config files so you can verify afterwards whether anything was altered. Four tools (continuity_breakpoint, continuity_breakpoints, continuity_ledger, continuity_attest), plain JSONL under <workspace>/.dsh-continuity/, zero dependencies, no network calls. Note: nothing writes ledger.jsonl yet — continuity_ledger only reads it.

Watches the turn endings of the models you select and asks a second model whether the turn was actually finished: if the assistant stopped after text with no tool call, a three-way verdict of continue, finished or uncertain is recorded and shown in a sidebar panel. In suggest mode it only reports; in auto-continue mode it calls agent.followup() once the turn is idle to send a continuation message, up to a per-turn limit. Single-step turns are always rechecked, watched models can be picked exactly or matched by case-insensitive regex, and the resume text is configurable.

Delete sessions, optionally with the files each session wrote; deletions go to a trash that lists them by time and restores one or many; a session with a running turn is refused and nothing outside that session folder is touched.

Splits a session into turns and incrementally clusters them into named topics and stages, with per-topic token / tool / time stats, a workflow view, a turn timeline, a relation graph and per-turn AI analysis.

Routes model and reasoning effort per turn: a cheap classifier reads each request and picks one of five tiers, a supervisor model re-routes on tool errors and checkpoints, and upgrading to the stronger model waits for confirmation.

EVA cel-shade skin for the Harness Web UI. Overrides the --dsw-* design tokens to give panels square corners, ink outlines, hard offset shadows and CRT scanlines, adds a three-state session rail (running, done, waiting on you), recolours the turn-navigator ticks, and can embed an optional dark backdrop image.