Plugins

Replaces the summarizing context compactor with a handoff note the model writes itself and a fresh window that keeps only the system prompt, the note, and the current turn.

Native generation-time Mixture-of-Agents: every turn, reference models run the same session context in parallel as pre-run drafts and one aggregator executes with full tooling; config-as-model recipes, F mirror / R trimmed views, enforced subagent gating (dual-shape session indexing), derived per-recipe vision (vision refs relay image reads to text-only aggregators), tiered cadence with separate steering and patrol rosters plus frozen reuse between beats, degeneration guard dropping degenerate reference output, anonymized model face (failure notices carry no provider identity), and cache-friendly guidance placement (the rehearsal record is pinned where it was produced instead of appended to every request, so the cached prefix covers it and each step pays only for what was added since - on by default across every cadence).

Ships a minimal lesson ledger: brief.ps1 to look up past pitfalls, add-lesson.ps1 to record new ones, deployed into the skills directory and wired into AGENTS.md by the installer. The model checks it before working and turns what it hits into rules, so the same mistake is not made twice. This plugin watches whether that actually happens and shows the result in a tab and under the composer.

Interactive HTML or flowchart cards inside the conversation: the tool blocks until the user submits, then returns the clicked action, fields, comments and card state.

DSH plugin: keep the running turn's work visible — process groups, reasoning and tool results expand automatically and fold once the group ends, leaving history untouched.

Elephant Memory: your DeepSeek agent never forgets. An endless memory for the DeepSeek harness: the agent keeps every chat it has in a project and reads all of it at every turn through a fixed-size set of summary lines (after Victor Taelin's OptChat speci

DeepSeek Harness conversation polish: message and step folding, composer and sidebar sizing, folded-content search, turn timing, and confirmed local conversation history deletion.

Command DeepSeek Harness from Telegram and Feishu (Lark): one live progress message per turn, approvals and questions as buttons, files and voice, delivery that survives outages and restarts, QR setup for Feishu, and per-person roles.

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/.

Adds a comet-like light band that flows along the window edge of DeepSeek Harness — up to four in parallel, colored by each session's model, reversing direction while waiting for your reply, and returning to the composer or the workspace once the conversation ends.

A chibi whale maid desktop pet for DeepSeek Harness who eats white rice while the model generates, lives inside the DSH window or anywhere on a macOS or Windows desktop, and dives for treasure between turns. On Windows the desktop overlay is compiled at install time and skipped when swiftc or the .NET SDK is unavailable, so the pet stays inside the DSH window.

Turns a phone into a personal assistant for the DeepSeek Harness on your computer — pair once by QR code, then chat, send photos and files, watch replies stream, and collaborate from several phones at once over LAN, Tailscale or a public tunnel. Ships an Android app, an iOS build pipeline and a zero-install web/PWA client for HarmonyOS and iOS browsers.

For DSH plugin AUTHORS: run this after you finish writing a plugin and it audits the entire publish path, then hands you the exact GitHub + npm + market commands. Pre-publish gate for DSH plugins. Turns every silent failure in the publish path into a deterministic check: package.json completeness (missing exports\["."], files omitting lib/ so the tarball is empty, no "type": "module", client declared without a ./client export), UTF-8 BOM, JSON/YAML parse errors, cordis.patch.yml insert rows written at the wrong indent (the bug that makes a plugin install and never mount), insert name mismatches, missing entry exports, require() inside ESM, missing LICENSE or README install line, missing tests, unscoped npm names that return a 403 no granular token can diagnose, versions already on the registry, and missing GitHub repos. Six agent tools; every finding carries its fix. Read-only, no LLM.

Windows toast notifications for DeepSeek Harness: approve or reject from the notification while DSH is in the background, with an in-app fallback; foreground only skips the current conversation, while other conversations still notify. Also reports turn completion and errors; subagent sessions stay silent.

Router and artifact gate for DeepSeek Harness: a per-turn skill router that ranks the live catalog, plus a write gate enforcing read-before-write, dedupe and a four-line audit on every file write.

Translates PDFs with BabelDOC through an OpenAI-compatible endpoint while keeping the original layout, and returns both a bilingual and a translated-only PDF. Requires a one-time runtime setup and an OpenAI-compatible API key.

Local non-generative decision layer: phocinae_ask and phocinae_gate return typed verdicts with calibrated confidence from a small decision model, with fail-closed escalation. Needs a local phocinae-server (pip install phocinae-server) with the Phocinae-Largha-150M-v1 model on loopback port 8155.

Adds a balance-checking tool, injects a per-turn DeepSeek account balance snapshot with a low/critical power-saving and stop policy, and provides a settings page for thresholds and FX-linked currency conversion. Requires a signed-in DeepSeek account for balance reads; API-key-only profiles return an unreadable result.

Task timing for DSH agents: a task_clock tool that measures elapsed time on the host monotonic clock, resumes across DSH or system restarts with the gap estimated, and returns unknown rather than inventing a number; it does not touch presets.

Adds a Chat mode beside Work with a user-defined persona, long-term memory extracted after each turn and recalled by topic, and stickers it can save and send.

Progress tab for the DSH right sidebar: the turn status of the session and the progress long-running tasks reported, in one tab beside the conversation.

Per-conversation system prompt, turn-based context folding, sampling overrides and a low-token pure chat mode for DeepSeek Harness.

Persona plugin: a slowly moving 0-100 emotion meter, its own worldview, a per-turn diary, a relationship network, proactive messages, and an inner-voice line beside the cost row.

Counts the subagent sessions a conversation spawned and shows their token usage and CNY cost in the composer dock, grouped by parent turn.