dsh-session-chat-share

English | 中文

Share a selected range of chat messages as Markdown, HTML, TXT, or PNG — a community plugin for DeepSeek Harness (tagged dsh-plugin and listed in the awesome-dsh-plugin market registry).

This repository is the standalone distribution of the plugin: it ships both halves prebuilt (lib/), installable with dsh plugin add and from the Plugin Market. The reference implementation lives in the harness repository as packages/session-query/session-chat-share, where both halves are built and tested.

Compatibility: DeepSeek Harness 0.1.6-alpha.2 or later. The plugin reads session data through the host's session-query service and one payload route, and contributes to the Session Header's utilities slot. Harness 0.1.0-rc.x needs the previous dsh-chat-share package instead (its last compatible release is dsh-chat-share@1.3.0).

What it does

• Registers the Web /share slash command — plain /share opens the dialog, /share txt saves the whole chat as one .txt, /share last <n> saves only the newest n messages (combine: /share txt last 10).
• The browser half adds a Share action to the Session Header and, on a harness whose ui-workspace carries the sessionRowMenu contribution registry, Share and Save TXT rows to each session's sidebar ... menu. The dialog lists the session's shareable messages (append-origin user/message and assistant/message text), lets you pick an inclusive range via From/To selects or by clicking message rows — or switch to multi-select mode to export the union of chosen rows — choose Markdown, HTML, TXT, or PNG, preview the rendered artifact (GFM), then copy it to the clipboard or download the file (.md / .html / .txt / .png). Nothing is uploaded: the recipient opens the artifact directly.
• Options: redact sensitive info (credential shapes and local absolute/home paths, on by default), include tool calls (bounded tool-call rows, off by default), and include subagent conversations (child sessions appended with section headers, off by default).
• The HTML artifact is a self-contained page with GFM-lite rendering (headings, lists, tables, blockquotes, links, fenced code, inline code/emphasis) and session images embedded as data URIs; artifacts follow the active UI locale. PNG is the HTML artifact rasterized as one long image.
• Optional host-side auto-save: with autoSaveDir configured on the plugin row, one TXT per session is written after every completed turn.
• Session data is read cold-safely through the host's session-query service, so an export covers the whole session no matter what the browser has paged into the transcript — no persistence changes and no model involvement. The command stays on the human-command plane with zero token effect.

Install

npm (preferred — the Plugin Market prefers npm sources):

dsh plugin --profile demo add dsh-session-chat-share

GitHub (alternative; ships the same prebuilt artifacts):

dsh plugin --profile demo add github:chrisx9z/dsh-session-chat-share#v1.4.4

The package ships prebuilt artifacts (lib/ — host and browser halves), so neither install needs a build step. Releases are tagged v1.x.y; pin a tag or commit for reproducible installs.

Then use it in any session of that profile:

/share

Plugin Market

The plugin is listed in the awesome-dsh-plugin registry, so it appears in Settings → Plugin Market — browse, one-click install, and updates once the catalog refreshes (usually within a day of a registry change). The npm source above makes market installs resolve to the published package.

Browser-half requirements

The installed package's browser half is picked up by the host's dsh.client scan, so the Header button and its dialog work on hosts whose composition includes it. Its imports are limited to platform modules every web build provides (@deepseek-ai/dsh-client-store, @deepseek-ai/dsh-client-ui-primitives, React), plus a bundled copy of html-to-image used for PNG export.

For the official distribution path (the harness repository's own web bundle), integrate the package as packages/session-query/session-chat-share and compose the chat-share row in packages/bundle/web-app/cordis.patch.yml.

How it works

• Host half (src/index.ts): registers /share on the human-command plane and serves GET /api/session.share?sessionId=<id>&includeSubagents=<bool>. The route observes the session through sessionQuery (live or cold), folds durable events into shareable messages (user/message, assistant/message, append-origin tool/call), appends direct subagent children when asked, and inlines referenced images as base64. With autoSaveDir configured it also writes one TXT per session after each completed turn.
• Browser half (src/client/): a controller fetches that payload once per session, keeps per-session dialog state, filters rows by the dialog options, and renders the chosen range with pure renderers (render.ts) into Markdown (verbatim text under role headers), a self-contained HTML page (GFM-lite), plain text, or a PNG rasterization.
• The dialog lists at most the newest 300 shareable rows and says so; direct saves (/share txt) always export the whole chat.

Development and tests

The package's test suite (command and payload-route behaviour, controller state, renderers, dialog, header action, and a real Loader composition) runs inside a deepseek-harness checkout where the @deepseek-ai/* workspace dependencies resolve:

pnpm exec vitest run packages/session-query/session-chat-share

Limitations

• The dialog lists up to 300 shareable rows; older messages stay reachable through direct saves, which always export the whole chat.
• Sharing is a copy/download artifact, not a hosted link: nothing is uploaded to a server.
• Redaction is best-effort pattern matching, not a guarantee; review the artifact before sharing.
• Message text is shared as rendered on the surface; reasoning text and tool results are not included (tool calls only, opt-in).
• The sidebar ... menu rows need a harness whose ui-workspace carries the sessionRowMenu contribution registry. On a stock harness the plugin still mounts and the Header action plus /share remain the entry points.

License

MIT