castreader（castreader 音频）

CastReader — Read Web Pages & Books Aloud

Setup (once per session)

cd <skill-directory> && npm install

How to find target (chatId)

User messages look like: [Telegram username id:8716240840 ...] The number after id: is the target. You MUST use this number in every message tool call. Example: target is "8716240840".

Mode A: When user sends a URL

Step 1: Extract article

node scripts/read-url.js "<url>" 0

Returns: { title, language, totalParagraphs, totalCharacters, paragraphs[] }

Step 2: Show info + ask user to choose

Reply with this text:

📖 {title}
🌐 {language} · 📝 {totalParagraphs} paragraphs · 📊 {totalCharacters} chars

📋 Summary:
{write 2-3 sentence summary from paragraphs}

Reply a number to choose:
1️⃣ Listen to full article (~{totalCharacters} chars, ~{Math.ceil(totalCharacters / 200)} sec to generate)
2️⃣ Listen to summary only (~{summary_char_count} chars, ~{Math.ceil(summary_char_count / 200)} sec to generate)

STOP. Wait for user to reply 1 or 2.

Step 3a: User chose 1 (full article)

Reply: 🎙️ Generating full audio (~{totalCharacters} chars, ~{Math.ceil(totalCharacters / 200)} seconds)...

node scripts/read-url.js "<url>" all

Then send the audio file using the message tool:

{"action":"send", "target":"<chatId>", "channel":"telegram", "filePath":"<audioFile>", "caption":"🔊 {title}"}

Reply: ✅ Done!

Step 3b: User chose 2 (summary only)

Reply: 🎙️ Generating summary audio...

Save the SAME summary text you showed in Step 2 to a file and generate:

echo "<summary text>" > /tmp/castreader-summary.txt
node scripts/generate-text.js /tmp/castreader-summary.txt <language>

Then send the audio file using the message tool:

{"action":"send", "target":"<chatId>", "channel":"telegram", "filePath":"/tmp/castreader-summary.mp3", "caption":"📋 Summary: {title}"}

Reply: ✅ Done!

Mode B: When user asks to read a book (微信读书 / Kindle)

Books are synced from WeChat Reading or Kindle to ~/castreader-library/books/. Each book is stored in a folder like 书名-hashid (e.g. 儒林外史-dc532c705c6d3edc5503acc).

⚠️ CRITICAL: You MUST use sync-books.js --list to get the exact book folder ID. NEVER guess or construct the folder path yourself. The folder name includes a title prefix that you cannot predict.

Step 1: List available books

node scripts/sync-books.js --list

Returns: { books: [{ id, title, author, language, totalChapters, totalCharacters, source, syncedAt }] }

The id field is the exact folder name you must use in all subsequent commands. Example: "儒林外史-dc532c705c6d3edc5503acc".

Step 2: Show book list and let user choose

Reply with the book list:

📚 Your synced books:

1. 📖 {title} — {author}
   🌐 {language} · 📑 {totalChapters} chapters · 📊 {totalCharacters} chars · 📱 {source}

2. ...

Reply the number of the book you want to read.

STOP. Wait for user to choose a book.

Step 3: Show chapter list and let user choose

node scripts/sync-books.js --book "<id>"

Use the exact id from Step 1 output. Returns the book content with chapter list.

Reply:

📖 {title} — {author}
📑 {totalChapters} chapters · 📊 {totalCharacters} chars

📋 Chapters:
1. {chapter 1 title}
2. {chapter 2 title}
...

Reply a number to listen to a chapter, or "all" to listen to the full book.

STOP. Wait for user to choose.

Step 4a: User chose a chapter number

node scripts/sync-books.js --book "<id>" --chapter <num> --audio

Returns: { title, audioFile, fileSizeBytes }

Send the audio:

{"action":"send", "target":"<chatId>", "channel":"telegram", "filePath":"<audioFile>", "caption":"🔊 {bookTitle} — Chapter {num}"}

Step 4b: User chose "all" (full book)

node scripts/sync-books.js --book "<id>" --audio

Returns: { title, audioFile, fileSizeBytes }

Send the audio:

{"action":"send", "target":"<chatId>", "channel":"telegram", "filePath":"<audioFile>", "caption":"🔊 {bookTitle} (full)"}

Reading a chapter as text (no audio)

If the user wants to read (not listen), use without --audio:

node scripts/sync-books.js --book "<id>" --chapter <num>

Returns: { title, author, language, chapter: { number, title, text }, totalChapters }

Rules

• ALWAYS extract first (index=0 for URLs, --list for books), show info, wait for user choice. Never skip.
• ALWAYS send audio files using the message tool with target (numeric chatId) and channel ("telegram"). Never just print the file path.
• For books: ALWAYS run --list first and use the exact id from the output. NEVER construct book paths manually or use partial IDs.
• Do NOT use built-in TTS tools. ONLY use read-url.js, generate-text.js, and sync-books.js.
• Do NOT use web_fetch. ONLY use read-url.js.
• Do NOT use the read tool to directly access files in ~/castreader-library/. ONLY use sync-books.js.