dsh-reading-pad

An immersive read-only reading panel for DeepSeek Harness —— the model pushes a typeset Markdown into your right sidebar, or you flip through the chapters of your current writing project.

中文 · English

What you get

• 📖 AI-delivered prose —— ask the model to push a typeset Markdown through reading_pad_send and read it in the panel; the text is rendered as-is, zero processing.
• 📚 Manuscript chapters —— switch to the "Manuscript" mode and flip through the chapters of your current writing project, without opening the writing workspace.
• 🎨 Three low-blue-light themes —— Paper (warm white) / Sepia (cream) / Night (dark grey), one-click switch, your choice is remembered.
• 📐 30em measure typography —— line width fixed at 30em (about 30 Chinese characters), line-height 1.6, 2em first-line indent, 0.8em paragraph gap —— designed after a printed book's rhythm, easy on the eyes.
• 🔤 A⁻ / A⁺ font sizing —— zoom between 16 and 20px, reflows automatically.
• 📊 Progress bar + chapter browser —— a thin bar shows where you are in the current chapter; the "Next chapter" button at the end of the chapter never auto-jumps (avoids accidental paging); first and last chapters grey out.
• 💾 Scroll position persistence —— close the tab and reopen it, you continue exactly where you stopped.
• 🔔 New-content badge —— when the AI delivers a new draft and the tab is unfocused, the tab title shows a "New" badge without stealing focus.

Install

dsh plugin --profile web add dsh-reading-pad

After installing, restart dsh web, then open the right sidebar → add the "📖 Reading Pad" tab from the "+" menu.

Requires DSH client packages >=0.1.5-rc.2. Uses the official native right-sidebar keyed slots (sidebar.right.pane.tab), no dependency on dsh-better-sidebar.

You can also install directly from GitHub (skipping npm):

dsh plugin --profile web add github:shengyvself/dsh-reading-pad

What you can do with it

After installing, just talk to the agent:

• "Push this draft to the reading pad" —— the AI calls reading_pad_send to deliver the typeset Markdown
• "Read chapter one" —— switch to the "Manuscript" mode and open the first chapter of your current writing project
• "Switch to Night theme" —— flip to the dark low-blue-light mode
• "Bump the font up a bit" —— press A⁺ to enlarge

The panel never acts on its own; it only shows content when you or the AI trigger it.

Content sources

Outline / lore / drafting-pad drafts are phase-2 placeholders (greyed out).

Compatibility

Installing the plugin does not upgrade the host DSH. The peerDependencies declare a minimum client-package version of >=0.1.5-rc.2; verified working on 0.1.5-rc.2.

Security

The reading pad is a read-only consumer, with explicit boundaries:

• Writes only the local plugin state file (~/.dsh/<state>.json, session display text + history ≤20 FIFO); never touches canon / creative files.
• No heavy third-party Markdown renderer. Rendering is a pure front-end M1 subset (headings / paragraphs / bold-italic / strikethrough / inline code / links-images / single-level lists / blockquote / hr / fenced code blocks / GFM tables); HTML is escaped, URLs restricted to http / https / relative paths.
• No network calls. Except for reusing the host's existing /api/writing/* HTTP RPC, there are no outbound requests.

Full policy: SECURITY.md.

Dependency notes

Server-side dependencies (@deepseek-ai/dsh-tools, @deepseek-ai/dsh-typert-protocol) are declared in peerDependencies —— provided by the host DSH, never written to dependencies (avoids manifest shadowing the host version). Client-side packages also go through peerDependencies (cordis / dsh-client-* / react).

Since v0.1.0 the panel is decoupled from dsh-better-sidebar: it registers to the official native keyed slots (sidebar.right.pane.tab + .title) via ctx.sidebarRightTabs.register (two-step registration).

Build

node scripts/build.mjs   # server-side ESM copy + client esbuild to browser CJS (ModuleLoader wrapper, id=dsh-reading-pad)

Development

See CONTRIBUTING.md, ARCHITECTURE.md, CHANGELOG.md.

License

Apache License 2.0 © shengyvself