Puzle Read Skill

Puzle Read Skill

Save web articles, documents and files to your personal reading library powered by Puzle. AI analyzes the full text and enables semantic search across everything you've read.

First-time Setup (show this on skill install)

When this skill is first activated, immediately greet the user and guide them through setup:

1. Check if a token is already configured:

   from puzle_reading import PuzleReadingClient
   if PuzleReadingClient.token_is_configured():
       # Token exists, ready to go

   If yes, tell the user: "Puzle Read Skill is ready! You can send me any URL, file, or text and I'll save it to your reading library. You can also search across everything you've read."

2. If no token is configured, walk the user through it:

   • Tell the user: "To get started, I need your Puzle token. Here's how to get it:"
   • Send the get_token.gif file (located at <this-skill-directory>/get_token.gif) to the user as an attachment so they can see the visual guide.
     • In OpenClaw: send via channel as a file attachment
     • In other environments: provide the file path for the user to open
   • Tell the user to open https://read-web-test.puzle.com.cn, log in, then follow the GIF to copy their token and paste it back.
   • Once received, save it with PuzleReadingClient.save_token(token) and confirm.

3. After setup is complete, briefly explain what the skill can do:

   • "You can now send me any URL, article, PDF, or text and I'll save it to your Puzle reading library. I can also summarize articles, search across your readings, and more. Just try sending me a link!"

Prerequisites

The bundled Client SDK lives at scripts/puzle_reading.py relative to this skill directory. It requires the requests library.

import sys
sys.path.insert(0, "<this-skill-directory>/scripts")
from puzle_reading import PuzleReadingClient

Token Management

Token is stored in ~/.config/puzle/config.json with file permission 0o600 (owner read/write only). The SDK looks for a token in the following priority order:

1. Constructor argument PuzleReadingClient(token="...")
2. Environment variable PUZLE_TOKEN
3. Config file ~/.config/puzle/config.json

User sends a token directly

If the user sends a token in the conversation (typically a long string starting with eyJ...), save it immediately:

PuzleReadingClient.save_token(token)
client = PuzleReadingClient()  # auto-loads from config file

After saving, tell the user: "Token saved to ~/.config/puzle/config.json. You won't need to provide it again in future sessions."

User does not have a token

Guide them step by step:

1. Tell the user to open Puzle Read: https://read-web-test.puzle.com.cn and log in
2. Send the tutorial GIF to the user as an attachment so they can see how to get the token. The GIF is located at get_token.gif relative to this skill directory.
   • In OpenClaw: send it via the channel as a file attachment
   • In other environments: provide the file path for the user to open
3. The GIF shows: open browser DevTools → Application → Cookies → copy the JWT token value
4. Ask the user to paste the token back to you

Once received, save it using the steps above.

Token already configured (most common case)

Once a token has been saved, all subsequent calls work automatically — no need to ask the user again:

client = PuzleReadingClient()  # auto-reads from ~/.config/puzle/config.json

You can check beforehand:

if not PuzleReadingClient.token_is_configured():
    # Guide the user to provide a token
    ...

Token expired (401 error)

The token is valid for approximately 7 days. When you receive PuzleAPIError(code=401):

1. Tell the user their token has expired
2. Ask them to obtain a new token from Puzle
3. Save the new token: PuzleReadingClient.save_token(new_token)

Environment variable method (optional)

Users can also configure via environment variable, which takes precedence over the config file:

export PUZLE_TOKEN="eyJhbG..."

Two Modes of Use

Mode A: Save for later ("Read It Later")

When the user just wants to save content — no need to wait for processing. Create the reading, give user the web link, done.

result = client.create_reading_from_url("https://example.com/article")
# result.web_url → "https://read.puzle.com.cn/read/42"
# Tell the user: "Saved! You can view it here: {result.web_url}"

Use this mode when:

• "save this" / "bookmark" / "read it later" / "store this link"
• "save all of these" — batch-saving multiple links
• User uploads a file but doesn't ask for any analysis — "just store this PDF"
• User shares a link in passing without asking a question about its content
• "keep it for later"

Mode B: Analyze now (background processing)

When the user wants to understand, summarize, or work with the content, processing takes 30–90 seconds. To avoid blocking the conversation:

1. Create the reading and immediately give user the web link
2. Spawn a background task / subagent to run wait_for_reading() + analysis
3. When the background task completes, present the result to the user

# Step 1: Create and immediately respond to user
result = client.create_reading_from_url("https://example.com/article")
# Tell user: "Processing now. You can preview it here: {result.web_url}"

# Step 2: In a background task / subagent, wait and analyze
detail = client.wait_for_reading(result.reading_id, result.resource_type)
content = detail["data"]["content"]
# Generate summary / answer questions based on content

# Step 3: When done, present results to user

Why background processing? wait_for_reading() blocks for 30–90 seconds. Running it in a subagent or background process keeps the conversation responsive. The user gets the web link instantly and can start reading there, while the analysis runs in parallel.

Use this mode when:

• "what does this article say" / "summarize this" / "give me a summary"
• "what are the key findings in this PDF" — need to read and analyze
• "compare the viewpoints of these two articles" — need content from multiple readings
• User asks a specific question about the article content
• You need to search across readings afterwards

When to Use Which Method

User shares a URL — Decision Chain

When the user gives you a link, do NOT call create_reading_from_url() directly. Try the following approaches in order, falling through to the next step only on failure:

User sends a URL
  │
  ▼
┌─────────────────────────────────────────────────┐
│ Step 1: Fetch the page yourself                 │
│                                                 │
│ Use WebFetch, browser, or any available tool    │
│ to retrieve the page content.                   │
│                                                 │
│ Success (got meaningful content)?               │
│   → Extract using readability (or use the       │
│     tool's formatted output directly)           │
│   → Extract: title, body HTML, plain text,      │
│     author, site name                           │
│   → Call create_reading_from_html()             │
│   ✅ Done                                       │
│                                                 │
│ Benefits of this approach:                      │
│ - Skips Puzle's fetching phase, faster          │
│ - Handles JS-rendered dynamic pages             │
│ - Handles pages with anti-scraping measures     │
│ - Higher content quality (you can preprocess)   │
└────────────┬────────────────────────────────────┘
             │ Failed: empty content, 401/403,
             │ auth wall, login required, timeout
             ▼
┌─────────────────────────────────────────────────┐
│ Step 2: Check for internal/specialized tools    │
│                                                 │
│ The URL might be an internal or auth-protected  │
│ resource. Look in the environment for tools     │
│ that can access it:                             │
│ - MCP servers for internal docs (Confluence,    │
│   Notion, Google Docs, etc.)                    │
│ - API endpoints that can fetch internal content │
│ - Authenticated browser sessions                │
│                                                 │
│ Found a tool and got content?                   │
│   → Save content as a local file (.html / .md)  │
│   → Call create_reading_from_file()             │
│   ✅ Done                                       │
└────────────┬────────────────────────────────────┘
             │ No suitable tool, or still failed
             ▼
┌─────────────────────────────────────────────────┐
│ Step 3: Fallback — let Puzle server fetch it    │
│                                                 │
│ Call create_reading_from_url(url)               │
│ Puzle backend handles fetching and parsing.     │
│ This may still fail for auth-protected pages,   │
│ but it's the last resort.                       │
│ ✅ Done                                         │
└─────────────────────────────────────────────────┘

Step 1 readability extraction example:

# Assume you've fetched the page HTML using WebFetch or similar tools
raw_html = fetch_result  # full HTML

# Extract body content (use any available readability tool or extract manually)
# You need: title, content (HTML), text_content (plain text)

result = client.create_reading_from_html(
    url=original_url,
    title=extracted_title,
    content=extracted_html,       # body HTML
    text_content=extracted_text,  # plain text version
    byline=author,               # optional
    site_name=site_name,          # optional
)

When to skip directly to Step 3 (no need to attempt Step 1/2):

• User explicitly says "let Puzle fetch it" or "use the URL method"
• Batch-saving multiple links (efficiency over quality — no need to fetch each one individually)

User provides content directly — create_reading_from_html()