Readwise Article Saver#

Save articles to Readwise Reader with LLM-powered tagging from a controlled taxonomy. WeChat articles are fetched server-side to avoid Readwise's parsing failures.

Workflow#

When the user sends a message containing one or more URLs, execute these steps immediately without asking for confirmation.

Step 1 — Fetch and save the article#

Use exec to run the bundled Python script. The script handles:

• WeChat detection and server-side fetching with MicroMessenger UA
• HTML content validation (empty-page detection)
• Title and author extraction
• Calling the Readwise Save API

bash

python3 ~/.openclaw/workspace/skills/readwise_article_saver/save_article.py "THE_URL"

The script outputs JSON to stdout:

json

{
  "status": "ok",
  "title": "Article Title",
  "author": "Author Name",
  "domain": "mp.weixin.qq.com",
  "text_preview": "First 8000 characters of article body text...",
  "is_wechat": true,
  "fetch_method": "server_fetch",
  "readwise_status": 201
}

Or on failure:

json

{
  "status": "error",
  "error": "Description of what went wrong",
  "fallback_saved": true
}

If status is "error" and fallback_saved is false, inform the user that manual saving is needed (open in WeChat → share to Readwise).

Step 2 — Generate tags with llm-task#

If Step 1 returned status: "ok" and includes a text_preview, use the llm-task tool to classify the article. Pass the full taxonomy as the prompt and the article metadata as input.

Call llm-task with:

json

{
  "prompt": "You are a document classifier. Read the document and return 2-5 tags as a JSON array of strings. PREFER tags from the taxonomy. If the content's central subject is not covered, create a new specific tag (1-3 words, same specificity as existing tags). Never create broad tags like 'Technology' or 'Finance'. Never assign 'favorite' or 'shortlist'.\n\nTAXONOMY:\n- AI agent: AI agents, autonomous systems, agentic workflows, tool-use architectures\n- Chips: semiconductors, chip design, GPU/TPU, NVIDIA/AMD/TSMC, export controls\n- AI 上下文: context windows, RAG, prompt engineering, foundation models, broader AI landscape\n- VC: venture capital, fund mechanics, early-stage investments, seed/pre-A/series-A\n- PE: private equity, series-B/C/D, buyouts, LBO mechanics\n- Fundraising: LP/GP dynamics, new LP allocation trends\n- Private Credit: direct lending, BDCs, unitranche, mezzanine, private debt\n- Equity: public equities, stock analysis, earnings, equity research, trading ideas\n- M&A: mergers, acquisitions, deal-making, corporate restructuring\n- Market: broad market conditions, macro outlook, cross-asset dynamics\n- Family Office: family office structures, ultra-HNW wealth management\n- Launching Fund: starting a fund, emerging manager playbooks, GP fundraising\n- Politics: domestic politics, elections, government policy (single country; NOT cross-border)\n- IR: international relations, foreign policy, diplomacy, geopolitics, great-power competition\n- Economics: macroeconomics, monetary/fiscal policy, trade economics\n- infra: infrastructure investment, physical/digital infrastructure\n- Consumer: consumer markets, retail, CPG, consumption-driven analysis\n- Startup Growth: startup scaling, growth strategies, go-to-market, PMF\n- Founder: founder-centric advice, founder stories, lessons from building\n- China, US, Europe, Middle East: apply when region is primary focus\n- Ray Dalio, Paul Graham, Howard Marks, 黄铮, Trump: apply ONLY if person is central subject (>50% content)\n- YC, XVC, Space X, Anthropic, Cursor: apply ONLY if company is primary subject\n- Career: career strategy, job transitions, professional development\n- Personal Development: mindset, self-improvement, habits, mental models\n- Mindset: psychological frameworks, resilience, cognitive biases\n- Guide: practical how-to, tutorials, step-by-step guides\n\nRULES:\n1. Always separate Politics from IR.\n2. For finance, choose the most specific tag. Never use generic 'Finance'.\n3. Key Thinker/Company tags only if central subject, not passing mention.\n4. Geographic tags only when region is primary focus.\n5. Return a JSON array of 2-5 strings. Nothing else.",
  "input": {
    "title": "<title from Step 1>",
    "author": "<author from Step 1>",
    "domain": "<domain from Step 1>",
    "text": "<text_preview from Step 1>"
  },
  "schema": {
    "type": "array",
    "items": { "type": "string" },
    "minItems": 1,
    "maxItems": 5
  }
}

The llm-task tool returns a JSON array of tag strings, e.g. ["AI agent", "China", "Guide"].

Step 3 — Apply tags to the saved article#

Use exec to call the Readwise API to update the article's tags:

bash

python3 ~/.openclaw/workspace/skills/readwise_article_saver/update_tags.py "THE_URL" "tag1" "tag2" "tag3"

Step 4 — Report to user#

Combine the results and report concisely:

• ✅ Success: ✅ 「Article Title」已保存到 Readwise Reader。标签: tag1, tag2, tag3
• ⚠️ Partial: ⚠️ 文章已保存但标签生成失败。标签: openclaw
• ❌ Failure: ❌ 无法保存此文章。建议在微信中打开后手动保存。

Do NOT add unnecessary commentary. Report the result and move on.

Handling multiple URLs#

If the user sends multiple URLs in one message, process each URL through Steps 1-3 sequentially, then present a summary table of all results.

Troubleshooting#

ErrorCauseSolutionREADWISE_TOKEN not setEnv var missingSet in openclaw.json under skills.entries``Readwise API 401Token expiredRegenerate at readwise.io/access_tokenServer fetch failed (WeChat)Link expired or anti-botUser should save manually from WeChatllm-task returns errorLLM provider issueArticle is still saved; tags fallback to "openclaw"