The AI loop

The point of 知行 is not another todo list. It is an AI loop around the work you already do in Harness:

You work in Harness	AI does	You get
Chat sessions, notes, and files	Reads the day's material, keeps the sources, and drafts a structured daily report	A daily report to edit and save
Save the daily report	Turns “tomorrow I will…” into tomorrow's todos, and suggests goal progress from the evidence	Tomorrow's action list + updated goals
Pick a week or a month	Aggregates that period's daily reports into a weekly / monthly report in your voice	A report ready to share
Paste a paragraph or upload a file	Splits it into year → month → week goals with measures, and later updates progress from reports	A goal tree that keeps moving

The AI is optional. Without a configured model you can still write reports manually, manage todos and goals, and use the small lucky features. With a model, the loop above is the default way to work.

AI daily report

AI is reading the day's Harness sessions and turning them into a daily report draft.

• The editor collects the day's Harness sessions, your notes, completed todos and attachments. Each source is listed before generation, and you can uncheck what should not be used.
• 生成日报草稿 asks the configured DSH model to write the report. The result is a normal Markdown draft you can edit; nothing is saved until you confirm.
• The same flow can run without AI: write the report manually or start from the collected material.
• Saving the report writes 日报/YYYY-MM-DD.md and turns the “明天准备干什么” section into tomorrow's todos. The report also carries goal tags, so later reviews can find the relevant goals.
• After saving, AI can propose goal progress updates from the report's evidence; nothing changes until you accept.

AI weekly / monthly report

AI weekly report: pick the range, choose the audience, then edit the generated Markdown.

• Pick a week or a month; AI reads that period's daily reports and writes the weekly / monthly summary.
• Choose the audience (自己 / 团队 / 上级) and generate one or more versions in a single run. Each version stays editable.
• The generated report inherits the goal tags from the period, so the weekly review and the goal tree stay connected.
• Edit inline, save as 2026-W46 / 2026-11 in the same 日报/ folder, then copy, export to PDF / PNG / HTML, or reuse it in the next report.
• This is the “weekly review” part of the loop: daily work in, a shareable summary out.

AI goals

AI goal breakdown (top) and AI progress suggestions from a report (bottom).

• AI goal breakdown: paste a paragraph, a plan, or upload a .md / .docx / .xlsx / .pptx file. AI splits it into year → month → week goals with a measure for each, shows a preview, and only writes 目标.json after you confirm.
• A goal tree can be arbitrary depth: yearly goals, monthly goals under them, weekly goals under those. Child progress automatically rolls up; leaf goals can be updated manually or from AI.
• AI progress updates: after a report is saved, AI reads the evidence and suggests a new progress for the leaves it can justify. Each suggestion shows the old → new percentage and the evidence; check the ones to accept, then apply.
• Goal stall reminders, list / timeline views, and linked reports make the goals part of the daily routine, not a separate spreadsheet.
• This is the “life / work goals” part: AI generates the structure once, then keeps the progress current as you report.

AI assistant on Today

Paste a paragraph (or a URL / attachment); AI turns it into todos you can edit and add.

• A paragraph, a URL, or an attachment can be broken into multiple todos; preview them, adjust the dates, then add.
• Todos can be linked to goals; the calendar and the daily report share the same data.
• Morning / evening reminders and the daily quote keep the Today page useful even when you are not asking AI anything.

The four pages

Page	What it does
今日 / Today	Todos, calendar, AI assistant, reminders, and the two small lucky entries
日报 / Reports	AI daily / weekly / monthly reports, history, export, goal tags
目标 / Goals	AI goal breakdown, progress updates, hierarchy, timeline
设置 / Settings	Reminders, region / clock, theme, and the DSH model the plugin uses

A little luck

The Today banner also has two offline features: 今天吃什么 (a time-aware lunch picker) and 今日运势 (a daily sign, a tarot card, and a 12-sign horoscope). They use the built-in content and local storage only — no AI, no network. A small bonus, not the main loop.

Quick start

1. Install DeepSeek Harness Desktop 0.2.x and make sure plugins are enabled.
2. Install the plugin:
   • npm: dsh plugin --profile desktop add dsh-zhixing (installs 1.0.2), or
   • from source: clone https://github.com/pawvis/dsh-zhixing.git and select the cloned folder in Plugins → Add plugin.
3. Restart DSH and open 知行 in the sidebar.
4. Open 设置 to set reminders, region and clock. If you want AI, make sure a DSH model is configured; the model picker in the plugin header controls which one the plugin uses.
5. Try the loop: open a day's report, click 生成日报草稿, edit, save; then open 目标 and use 文字/文件 AI 拆解 on a plan.

The npm package is published; the latest version is 1.0.2. The GitHub source is for development or offline install. The plugin works without AI; the AI features simply stay unavailable.

Privacy & data

• Local data. Todos, goals, reports, imported files and settings live under ~/.dsh/data/知行/ (or $DSH_HOME/data/知行/). Reports are plain Markdown.
• No bundled keys. The plugin ships no API keys. AI features use the model already configured in DSH; only the content you send to an AI feature (report material, report draft, goal import, progress suggestions, assistant input) goes to that model.
• Sources stay visible. The report editor lists the sessions and materials used for a draft; you can uncheck them before generating.
• Weather is optional. The weather line makes a weather request for the configured region; turn it off in Settings. The daily quote is built in.
• Lucky features are offline. They use the built-in content and a few localStorage keys. No network, no AI.
• No telemetry. The plugin adds no analytics or remote logging.
• Screenshots in this README are illustrative and use demo content; real todo / report / goal content is never committed.

All screenshots above are captured from the plugin's UI with demo data.

Troubleshooting

• AI is unavailable: the banner says “AI 暂时不可用” when no model is configured or the model is unreachable. You can still write reports manually, manage todos / goals, and use the lucky features.
• No 知行 entry: check DSH Desktop 0.2.x, the plugin is enabled under Plugins, and DSH was restarted after installation.
• Where is my data? ~/.dsh/data/知行/: 日报/YYYY-MM-DD.md, weekly / monthly files in the same 日报/ folder (2026-W46.md, 2026-11.md), 待办.json, 目标.json, settings.json, 导入/ and log.md.
• A generated report looks off: edit it before saving; the AI draft is a draft, not an automatic commit.
• Goal progress did not change: progress changes only after you accept a suggestion (or edit it manually); child goals then recompute automatically.
• Updating: replace the plugin or pull the repository and restart DSH. The data lives outside the plugin folder, so it is kept.

Development

Requires Node.js 22.19+ for the checks; the plugin runs inside DSH.

npm run check

The check runs node scripts/check-components.mjs, node --check index.js and node --check client.js. There is no build step: DSH serves client.js directly, and the CSS is a template string inside the client.

index.js            # plugin entry / server routes
lib/                # store, routes, AI, reports, goals, reminders, ...
client.js           # the whole web UI (React h(), CSS)
locale/             # zh / en UI strings
cordis.patch.yml    # DSH bundle manifest
icon.svg
docs/DETAIL.md      # full feature notes (Chinese)
docs/readme/        # README assets
docs/lang/          # translations
scripts/            # checks

Full feature notes: docs/DETAIL.md (Chinese). Issues and ideas: GitHub Issues.

Languages

English · 简体中文

The plugin UI is Simplified Chinese; the Chinese README is the primary user guide.

License

MIT © dsh-zhixing contributors