インターフェース · DeepSeek Harness プラグイン — 71 ページ

The dsh SDK profile bundle: stdio JSON-RPC serving and process lifecycle over dsh-base

The standalone minimal SDK profile bundle: JSON-RPC, one DeepSeek adapter, persistent shell, editor, and JSONL sessions

Web Session-log export command and shared download dialog (desktop fork: the trigger lives in the session toolbar's overflow menu, not its own header capsule)

JSONL durable session persistence backend for the DeepSeek Harness

JSONL durable session persistence backend for the DeepSeek Harness

Browser control: drives Edge/Chrome via the Chrome DevTools Protocol so the AI can automate the web (navigate, click, type, screenshot, evaluate JS). Settings live in the 'browser-control' namespace with a /api/browser status surface

Browser control: drives Edge/Chrome via the Chrome DevTools Protocol so the AI can automate the web (navigate, click, type, screenshot, evaluate JS). Settings live in the 'browser-control' namespace with a /api/browser status surface

Model-facing clipboard tools: clipboard_read and clipboard_write over the OS clipboard (PowerShell Get-Clipboard/Set-Clipboard on Windows)

Model-facing clipboard tools: clipboard_read and clipboard_write over the OS clipboard (PowerShell Get-Clipboard/Set-Clipboard on Windows)

Computer-use for DeepSeek Harness: control the Windows PC via Windows-MCP, provisioned with a bundled uv and an isolated Python runtime under $DSH_HOME/runtime

Computer-use for DeepSeek Harness: control the Windows PC via Windows-MCP, provisioned with a bundled uv and an isolated Python runtime under $DSH_HOME/runtime

Deliver AI artifacts to the user: model-facing send_file tool copies a produced file into <files>/deliver and emits dsh/notify (kind=deliver) so the mobile native layer can offer one-tap save to Download/DeepSeekHarness

Deliver AI artifacts to the user: model-facing send_file tool copies a produced file into <files>/deliver and emits dsh/notify (kind=deliver) so the mobile native layer can offer one-tap save to Download/DeepSeekHarness

Model-facing file_name_search tool: fast recursive filename (glob substring) search under a directory, ignoring node_modules/.git

Long-term memory for DeepSeek Harness: remember/recall/list/forget tools plus a memory skill, persisted under $DSH_HOME/memory/memory.jsonl

Task notifications + scheduler for DeepSeek Harness: Windows notifications when the AI starts/finishes a task, plus user-defined repeating scheduled tasks with optional commands, driven by the `notify` settings namespace

Phone control for DeepSeek Harness mobile: model-facing tools let the AI read the current screen's interactive elements and tap / swipe / type / press keys / scroll / open apps through the native accessibility service (local HTTP bridge at 127.0.0.1:3090)

Model-facing remind tool: schedule a delayed desktop notification (remind me in X minutes) bridged through the dsh/notify event

Shared snippet library: model-facing snippet_save/list/get/delete tools plus /api/snippets for the UI picker — one store for the user and the AI

Usage & balance for DeepSeek Harness: replays session events to estimate CNY spend (DeepSeek deepseek-chat pricing) and queries the live DeepSeek /user/balance, served over /api/usage for the sidebar budget bar

Multimodal vision tool for dsh: analyze images or screen captures using DeepSeek vision models

Model-facing web_fetch tool: fetch a URL and return its readable text content (scripts/styles stripped, whitespace collapsed, length-capped)

Display DeepSeek API account balance under each assistant turn

The dsh browser-surface bundle: the web patch layer over dsh-base plus the runtime glue plugin (frontend dist serving, web-surface prompt, bash runtime variables, URL line)