プラグイン

Command palette (⌘K) for the DeepSeek Harness Web GUI: keyboard-first quick entries over sessions, settings, and actions, with a customizable hotkey and a register extension point.

In-place redaction and row-level rollback for session logs, preserving the seq-equals-row-index invariant that makes deleting a middle row corrupt the entire log. Originals are copied to a quarantine sibling before any write, and the reader's own semantics are replayed as a pre-write gate.

Preventive content policy for tool results: rewrites matched text, and drops or truncates whole fields by path before a result is normalised, rendered or persisted, so an unreviewed payload never enters it. Ships inert with no built-in category or domain blocklist.

AnySearch-backed search tools for agents: general and vertical-domain search, parallel batch queries, vertical-domain capability discovery, and page-to-Markdown extraction, plus an API key panel in Settings. It registers its own tools instead of a web_search provider, so it cannot be installed alongside @anysearch/anysearch-dsh.

Ships the Zhouli-mode agent preset, which runs a session in three phases (declaring roles on opening, citing authorities while working, a six-item self-audit at the close), plus one tool with seven actions over a bundled 49,384-character 《周礼》 corpus that returns every passage with a stable 篇序.职官序 coordinate, a per-office establishment index and collation between two Kanripo witnesses.

Read-only bridge that surfaces Claude Code's existing project memory in DSH — git-root-aware resolution, credential redaction before injection, and an on-demand recall tool. Zero runtime dependencies.

Adds per-route context window and compaction threshold fields to the Models settings page and a vendor-aware thinking-effort picker to the composer, writing llm-pi-ai capacities, reasoningEfforts, and a compaction-basic patch row.

Adds Open workspace entries to the sidebar session-row menu and the session header actions slot, opening the session workspace folder in the system file manager.

Voice input plugin (dual-face): mic button beside the composer send action, live recording bubble streaming PCM to a local Qwen3-ASR python service managed by the host, plus an ASR environment settings page (clone runtime/model repos, create venv, run ser

Multi-pet desktop companion: four MAM status lights watch every active session (approval / running / done-unread / error) and each card click-through to its conversation; four voice groups speak on state changes with duration-timed subtitles; import pets from four sources (folder / zip / Codex dir / Petdex registry) with card-style hot-swap and an integrity guard; five action bindings, drag physics, three scales. The built-in fox Foxbell ships in the package — install and go.

Semantic risk grading and progressive authorization: classifies tool calls into safe/risky/redline, asks before irreversible actions, auto-allows only approved-and-succeeded signatures.

Global Quick Actions beside every session-backed composer in DSH. One press sends a preset or your own saved text, with per-action send confirmation.

A plugin for dsh: quantify every dsh Agent interaction as a visualized performance metric and automatically sync it to WakaTime -- use data to showcase your productivity with Agent.

CWL — Context Window Lifecycle for DeepSeek Harness: structured context eviction (arXiv:2606.11213). Graduated, deterministic, zero-LLM eviction of exploration/action episodes when context pressure exceeds budget — no summarization lossiness, no hallucina

Model switching, model capability configuration, fixed compaction models, subagent routing, and input history enhancement.

Multi-agent frugality defense plugin: read-ledger dedup metrics, compaction-immune rules, completion gate, cheap-model review lane.

Batch recursive extraction: magic-byte format detection (disguised extensions), nested archive auto-unroll, multi-password pool with per-layer dynamic discovery, SFX/split-volume pairing.

Explains which layer killed a failed turn: times every model request (first byte, longest silence, total) and combines that with the error to name the culprit (dsh idle watchdog vs Node undici timers vs SDK timer vs server crash, context overflow, gate queue, tool timeouts, failed compactions) and the exact setting to change; verdicts in the DSH terminal plus a /why command, subagents included.

DSH Web GUI 插件，用于对会话消息进行编辑、重试、高级重试和删除等功能的实现

Context-window attribution for DSH Web: a Mini Utility Dock panel that ranks the user messages, assistant turns, reasoning blocks and tool results of a session by the tokens each one occupies, so it is clear which entries hold the window. It also shows the share by entry type, the heaviest entries across the live sessions of the host, and, where the host provides them, provider usage and the estimated system/tools/messages composition. The panel is read-only: it does not estimate cost, modify a session, or trigger compaction.

Context compaction / headroom for DeepSeek Harness — compress the live context so the prompt stays small without losing the state that matters (layer 3).

dsh port of the shared handoff kit: handoff and task-id-bootstrap skills plus host-side automation — baseline state injection, per-turn snapshots with an Auto Log trail, task=<id> routing, continuation re-injection, a compaction guard with clear-required alerts, and read-only subagent handling — all writing the same .agents/state/ as the Codex, Claude, and pi editions for one-repo multi-agent handoff.

Serves the obra/superpowers skill catalog to the harness and injects its using-superpowers bootstrap as a persistent, compaction-safe system-prompt section.

Codex-shaped coding tools and agent presets for GPT models on DeepSeek Harness: exec_command, write_stdin, apply_patch and view_image, plus OpenAI Chat Completions and Responses API routes, bridged to DSH-native plan, ask, approval and compaction. Long-running commands become DSH background jobs with bounded head/tail output readable through job_output, and each subagent type can be pinned to its own preset, provider, model and reasoning effort.