Plugins

Skill center for DSH: browse skills by region, install them from a zip archive, and manage them afterwards with enable/disable, uninstall, trash restore and search.

Prompt-polish template injector for DSH: click ✨ in the composer to inject a "polish my prompt" template (current draft auto-embedded), no send, no network, no model — the main agent rewrites with full context. Replaces narrative-prompt-polish (2026-09-14 migration).

Indexes the current workspace into a relation graph and exposes one agent tool, codegraph_explore, for reachability, callers, and impact, plus a Web sidebar graph pane.

Figma design context for the agent: reads a frame's node tree, auto-layout, paints and typography, resolves the variables it is bound to, and renders a screenshot to compare code against — plus four bundled design-to-code skills. Connect with one button against Figma's own consent page; the plugin ships its own OAuth app, refreshes tokens automatically, and never asks for or displays a token.

Appends actionable hints to failed tool results, such as the nearest existing directory when a path does not exist.

Hard stop for runaway turns: blocks repeated identical tool calls and ends the turn when per-turn step or time budgets are exceeded.

Ein local-first persönliches Zustandsprotokoll und eine Entscheidungsmaschine für dsh: bewahrt jede Eingabe bytegenau mit Provenienz auf Zeilenebene auf, argumentiert gegen die eigenen Schlussfolgerungen, indem sie zuerst Gegenargumente erzeugt, versieht jeden gespeicherten Fakt mit eigener TTL und meldet, wo eine Entscheidung auf gar nichts beruht.

Local relationship-memory workbench with a confirmation gate: every AI write — extracted memories and newly created contacts alike — lands in a pending queue until the user confirms it in the GUI; materials are archived verbatim with quote-level provenance checks on AI extraction, and confirmed memories aggregate into per-contact timelines, gift occasions and reciprocity views. Ships a native relationship agent preset for conversation capture; JSON storage by default with an optional SQLite backend, zero production dependencies.

Model-callable compact_agents tool that force-compacts context in every live session (main session, ordinary sub-agents and AgentTeams members alike), queues busy targets and compacts them when the turn ends, reports the shadowed node count and estimated tokens per target, and ships a browser form for the trigger ratio, retained ratio, controlled-phase output budget, auto-continue budget and an end-of-turn pre-compaction ratio that compacts near the trigger line once a turn ends, so a reply no longer opens with a wait for the summarizer.

Session management for DSH: archive, restore and delete persisted sessions, cap how many sessions each workspace keeps, and review disk usage in a size governance panel.

Session health check for DSH, read-only: scans the whole session library (totals, largest sessions, warnings for oversized sessions and event-count blowups) and deep-inspects a single session log (tail-balance / crash-tail check, tool-call/result pairing, event-type distribution).

Route each WeChat group session to a different Agent preset by group name, reusing dsh-wechat-assistant without changing its code.

Model-driven context compaction for DeepSeek Harness. The agent decides when and what to compress, commits durable session-log replacements, and ships reversible tool-result compression with byte-exact hash retrieval.

Rotate the outbound LLM session ID after each successful DSH compaction without creating a new DSH session.

Cross-version session migration for DSH: sweeps legacy session stores (v0 plaintext, .zstd and .zip exports), re-pins the header cwd to the target workspace and drops sessions in so the persistence layer upgrades them on open; settings sidebar with env/import/list/convert tabs plus a CLI (check / fix / import).

Merges other sessions' context into the current turn: evidence-weighted conflict resolution, retraction of claims whose premises were retracted, trust labelling that resists cross-session prompt injection, and budget-bounded packing. Ships a benchmark.

One settings page that lists every subscription sign-in the harness can offer and walks you through it — ChatGPT (Codex), Claude, GitHub Copilot, OpenRouter, Kimi and xAI — reading the list from the authorization seam instead of hardcoding providers, and showing credential records no registered flow claims any more.

Create and configure Agent presets from a single settings section — persona prompt, a checked knowledge-base entry list injected into the system prompt, individual switches for four knowledge-base tools, per-Agent narrowing of MCP connections, and batch knowledge-base import.

Registers the Xunfei Loomy cloud channel as an OpenAI-compatible supplier in dsh-router, adding model listing and streaming chat completions.

Shows the DeepSeek account balance as a row in the sidebar footer with a refresh button, and opens a popover with the balance card, daily token and estimated-cost charts over a 7/30-day window, a per-model table, and a link to the platform usage page. The balance comes from the official /user/balance endpoint; usage is aggregated from local DSH session logs and priced with the published peak/off-peak rate card in CNY, so it is labelled an estimate and its scope is stated on the panel.

Shows token and CNY cost per turn and per session in the harness's own statistics form, a whole-account bill over 1/7/30-day windows with cost-per-unit-of-work trends, and an advisor that re-measures its own suggestions after you adopt them.

Replaces the chat input model picker with a provider-grouped, collapsible list: groups models by provider, folds vision-bridge mirror providers back into their upstream, and carries the reasoning-effort options each model declares in settings. Tested on DSH Desktop only; the DSH web build is untested.

Prompt scene picker for DSH: choose a scene from a dropdown and it opens a new session with the prompt already filled in (with {variable} placeholders), so you only tweak and send.

Keeps the question you are currently on pinned above the composer, with a dropdown of every earlier question that scrolls the conversation back to that turn; questions folded into process groups are expanded first, and Escape, scrolling or clicking outside collapses the panel.