プラグイン

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.

Per-workspace custom system prompts for DeepSeek Harness (DSH): set one active prompt per workspace, captured into a session at creation time, editable from a session-header dialog. 给 DeepSeek Harness 的「工作区自定义提示词」插件：每个工作区一条生效提示词，新建会话时固化注入，会话头部弹窗增删改查。

Adds a sidebar music browser with multiple public sources, persistent background playback, queue advancement, and reviewed MusicFree-compatible source extensions.

Audits every plugin HTTP route in a profile for a browser-trust fence: plugin routes win the web server's longest-prefix match ahead of the /api gateway, so they never see its trust check and must pin the Host to loopback themselves. Grades PASS/WARN/FAIL per route and fails a fence that compares Origin to Host without pinning it (bypassable by DNS rebinding). Ships as a CLI for CI and a route_fence_scan tool.

Runtime security guard for DSH: loader import confinement, HTTP Host header validation, and source patch for VM sandbox escapes (4 CVEs). AI-assisted.

Tools guard that moves agent-issued `rm` targets to the macOS Trash instead of deleting, with a shell-aware lexer covering compound and disguised commands; a switch in Settings → General turns it off.

Background security auditor for DeepSeek Harness: scans agent outputs for secret leakage, checks command safety before execution, and surfaces findings in a persistent audit log.

Auto-adjust model reasoning effort per task (off/low/high/max): task classification, \[effort=...] markers, in-turn error escalation, DeepSeek peak-pricing-period capping with countdown notices, and manual-selection priority.

Real AST-based scan of installed plugins for malicious-intent patterns (dynamic eval, cross-plugin writes, exfiltration-shaped network calls), with an optional periodic schedule and auto-quarantine on critical findings.

Tiered task queue as the single source of truth for DSH↔Hermes work (reset > approve > review): enqueue tools for all submission paths, lease/claim model with concurrency 1, priority aging against starvation, atomic busy mutex, and a Hermes-side consumer (cron */1 + monitor gating) that drains tasks, writes review-handoff requests with single-slot checks, and invokes the reset agent.

Hands DSH reset requests to an external ops agent via a versioned JSON protocol: preflight snapshot, pre-restart maturity gate (no pending approvals, free disk, cooldown), restart, health check, and interrupted-work recovery, with the result delivered back to the requesting session.

Auto-approval agent for DSH permission requests: rule layer (denyAlways/allowlist) plus Hermes Pro semantic adjudication with quick-fail retries, fail-closed to human when Hermes is unavailable, rejection feedback returned to the requesting session, and approve tasks enqueued into the task queue with a shared busy mutex.

ServerChan3 (Server酱) push notifications when a turn finishes — title, model, project directory, git branch, status and a reply excerpt, with subagent filtering and fire-and-forget delivery.

Task-complete and question notification sounds for the DeepSeek Harness Web UI: a soft chime when the agent finishes a turn, a ding-dong when it asks you something (questions, plan reviews, approvals). Synthesized Web Audio, zero assets, mute toggle and volume slider in the session header and settings.

DSH Source Control tab in dsh-better-sidebar that mirrors VS Code Git workflows.

Git status badges for DSH: sidebar workspace-row badges and an input-row chip showing the current conversation's workspace git state, with event-driven freshness over SSE.

Task Runner mode for DeepSeek Harness: the main agent decomposes large tasks into independent subtasks, dispatches them to subagents (default local model) in fresh short contexts, and synthesizes results; concurrency and worker/orchestrator models are configurable in the settings panel.

✨ One-click prompt polisher for the DSH composer — an agent refines your rough draft in a side chat over multiple turns, and you copy the result back when you're happy.

Turns a product or UI request into an implementation-ready design spec with /ui-workflow and /ui-rerun. Analysts run through a read-only ui_analyst tool, and the lead agent writes docs/ui-workflow reports.

Intercepts turn-stopping to force one critique-and-improve round after each completed turn before the agent stops.

Plan-and-execute orchestration over ReactLoopAgent — turns a goal into reviewable steps with a web UI plan card.

Captures every plan the agent presents through exit_plan_mode as markdown under <workspace>/.dsh/plans, opens a review window, and keeps a Plans history tab recording the outcome of each plan.

Folds the DSH skill catalog prompt surface with KV-cache-stable rendering, BM25/bge-m3 hybrid skill search, and optional auto route hints.

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.