Plugins

Wire-level LLM inspector: captures the raw HTTP request and response bodies of model provider calls and attributes each capture to the session that caused it through the llm/stream waterfall. The request pane renders the body as a collapsible tree with per-node sizes — one measured request was 98.6% a single 973-entry array — and the response pane reassembles the OpenAI-style SSE chunk stream into reasoning, content, tool calls and usage. Shown in a Conversation View tab and a loopback viewer page.

Ask the user a question about a local image; the DSH Web card offers zoom, pan, box, brush and lasso annotation, NG/OK/unsure verdicts and a note, and the agent receives annotations back in source-image pixel coordinates.

Conversation outline panel for the DSH web surface: user questions as root nodes (index, time, tool count and text volume per turn) with the Markdown headings of assistant replies hanging beneath them, plus level slider, search scope, bookmarks, copy and export as Markdown, section jump, keyboard navigation, a right-edge density rail and per-session reading-position memory. Read-only UI: it never touches prompts, tool results or provider traffic, and writes nothing outside its own localStorage keys.

Adds a speaker button immediately right of the Like button on every finalized assistant reply; click it to hear the reply through the browser speech engine, and hover it for a speed and voice panel.

Windows-native computer use for DeepSeek Harness: two tools drive and read any desktop app. One batch call runs find, click, type, key, read, wait and screenshot steps in a single engine pass, so a task whose steps are known up front costs one model round trip; reading returns window text rather than an image, and input and capture prefer UI Automation patterns, Win32 messages and offscreen PrintWindow so the user's focus and pointer stay untouched.

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

让 dsh 在 Windows 上使用用户自安装的 Git Bash，并绕过其沙箱限制 | Enable dsh to use a user-installed Git Bash on Windows by bypassing its sandbox restrictions.

Cross-session memory for your dsh agent: remembered, recalled, and re-verified instead of assumed.

Password-optional LAN entry for the DSH Web GUI. DSH stays loopback-only while a settings-card plugin serves its HTTP and WebSocket traffic on port 3301, storing the password only as a scrypt verifier.

Query and manage JumpServer through conversation: assets, users, accounts, permissions, sessions, command audit logs, command filters, and RBAC roles, authenticated with an AccessKeyID/AccessKeySecret pair (HTTP Signature).

A continuation of simon300000/dsh-auto: the Auto Approve permission preset for the DSH Web UI reviews each approval with one single-shot model call instead of a reviewer subagent, auto-approves only the actions that pass and hands the rest to human approval, and remembers confirmed decisions as project- and global-scope allow/deny lists, with an approval timeline in the right sidebar.

Keeps a redundant sandbox-escalation argument from failing a tool call. Escalating tools (pwsh, bash, write, edit) advertise the full sandbox_permissions enum, but DSH only accepts a level strictly wider than the one already in effect — a check its own source calls 'deliberately not a schema constraint'. A model that reflexively passes the argument therefore picks the level it is already at and the call dies before running with 'sandbox escalation to "workspace-write" is not strictly wider than this call's current "workspace-write" mode', costing some models a whole turn of retries. This plugin registers one tools/execute waterfall listener and, only on that documented rejection and only when the arguments really carried an escalation field, re-issues the identical call once without it. Safe by DSH's own documentation: the rejection precedes any execution ('nothing has run'), and the corrected call cannot match again, so the retry is loop-free. Reproduced and verified end to end — before: isError true, the command never ran; after: the command's real output, isError false, one tool/call and one tool/result. Zero dependencies.

Manage Proxmox VE through conversation: 116 read + 116 write tools covering VMs, containers, storage, nodes, clusters, backups and firewalls, with API-token or username/password (ticket) auth.

Sound alerts for the DSH web UI: a chime when an answer completes and a three-tone prompt when your approval or question is pending, with optional custom WAV stored only in the local browser (pure client-side, zero assets).

Training attestation — reproducible, attributable, auditable off-site standardized training for asset pricing. Hard-gates four failure modes (half-refresh, hash mismatch, unknown-read-as-pass, master pollution). Ships one skill.

Automatically orchestrates complex DeepSeek Harness requests. It detects when delegation is useful, builds a dependency-aware plan, assigns bounded researcher, architect, backend, frontend, tester, and documentation workers, collects structured evidence, and sends the results to a final reviewer. The bundle atomically enables the workflow engine and model-facing subagent tools; disabling the plugin disables its complete tool group as well.

Layered cross-session memory for DSH that pushes only what changed - the first round injects every active entry, later rounds push only added / updated / removed ones, and nothing at all is injected when nothing changed, with keys that merely repeat an entry id left out of the text so the baseline is about 26% smaller. Facts and decisions carry semantic keys (one active truth per key), the model may only write to an inbox that a human promotes, an entry can carry a review date that raises a one-time reminder when due, and search is relevance-ranked with journal lines down-weighted and Chinese matched by bigram. A sidebar tab tells the lifecycle story instead of listing files - entries are grouped by stage (awaiting confirmation / in use / archived) along four axes (topic - one level, human-assigned, type, tag, date), each group can be ticked for batch promote / withdraw / archive / restore / delete / assign topic, and topics can be renamed store-wide or searched on their own. The archived stage reads in three levels - the category says why an entry left (distilled into the docs / useless for good / not needed right now, the first two being defaults you can add to), below it the entry topic stays collapsed until you click it, and entries with no topic sit right under the category. The tab also shows the global and workspace instruction files plus the current injection size. A zero-dependency CLI ships with it, so the store is usable without DSH.

System-audio visualizer for the turn navigator rail: the rail reacts in real time to whatever the default playback device is rendering, with four motion modes (spectrum, pulse, sway, travelling wave), a 16-band logarithmic spectrum from a 2048-point FFT on the host, and live sensitivity, strength, bass-punch and wave-speed controls. Audio is captured host-side through WASAPI loopback, so there is no permission prompt, no microphone and no screen capture. The browser half adds one attribute, one stylesheet, some CSS variables and a sidebar control row, and never patches the official component; switching it off removes the attribute, the stylesheet and the variables, returning the rail to its default styling. Windows only.

Flip the DSH page over to a glass task board showing live jobs, subagents and session state, with an infinite canvas that unrolls a card into draggable controls, embedded web-app frames, and a conversation box that edits the canvas.

Report stage — academic paper writing for asset pricing and machine learning research. Ships one skill.

Data stage — data discovery and quality assessment for asset pricing and machine learning research. Ships one skill.

Literature review stage — search, synthesis, and citation-chain tracking for asset pricing and machine learning research. Ships one skill.

Model training stage — end-to-end analysis, model training, and estimation for asset pricing and machine learning research. Ships one skill.

PPT stage — presentation generation (Beamer/Quarto) from the paper for asset pricing and machine learning research. Ships one skill.