Lark/Feishu Sheets & Cloud File Download (with PDF extraction)

Read, write and manage Lark/Feishu Sheets, and download Lark/Feishu cloud files, by calling the official OpenAPI from local scripts.

Prerequisites

• python3 on PATH
• Feishu/Lark app credentials configured in ~/.openclaw/openclaw.json under channels.feishu:

```json

{

"channels": {

"feishu": {

"appId": "cli_xxx",

"appSecret": "xxx",

"domain": "feishu"

}

}

}

```

• The Feishu/Lark app must have Sheets read & write permissions and Drive file download permissions enabled in the developer console.
• The target spreadsheet/file must be shared with the app/bot identity.

Quick Start

Get spreadsheet token from the URL

Example URL:

https://.../sheets/YOUR_SPREADSHEET_TOKEN?sheet=SHEET_ID

• spreadsheet_token = YOUR_SPREADSHEET_TOKEN
• sheet query param (often a sheetId) = SHEET_ID

Read / Export

# Export a single range to CSV
python3 {baseDir}/scripts/sheets_export.py \
  --token YOUR_SPREADSHEET_TOKEN \
  --range 'SHEET_ID!A1:Z200' \
  --csv /tmp/sheet.csv

# Or export to JSON (recommended for multi-range)
python3 {baseDir}/scripts/sheets_export.py \
  --url 'https://xxx.larksuite.com/sheets/YOUR_SPREADSHEET_TOKEN?sheet=SHEET_ID' \
  --range 'SHEET_ID!A1:Z200' \
  --json /tmp/sheet.json

Then load /tmp/sheet.csv or /tmp/sheet.json and continue with analysis/summarization.

Write / Update

# List all sheet tabs
python3 {baseDir}/scripts/sheets_write.py \
  --token YOUR_SPREADSHEET_TOKEN list-sheets

# Write values to a single range
python3 {baseDir}/scripts/sheets_write.py \
  --token YOUR_SPREADSHEET_TOKEN \
  write --range 'SheetId!A1:C2' --values '[["a","b","c"],["d","e","f"]]'

# Write values from a JSON file
python3 {baseDir}/scripts/sheets_write.py \
  --token YOUR_SPREADSHEET_TOKEN \
  write --range 'SheetId!A1:C2' --values-file /tmp/data.json

# Batch write multiple ranges at once
python3 {baseDir}/scripts/sheets_write.py \
  --token YOUR_SPREADSHEET_TOKEN \
  batch-write --batch '[{"range":"Sheet1!A1:B1","values":[["x","y"]]},{"range":"Sheet1!A2:B2","values":[["1","2"]]}]'

# Add a new sheet tab
python3 {baseDir}/scripts/sheets_write.py \
  --token YOUR_SPREADSHEET_TOKEN \
  add-sheet --title 'NewSheet'

# Clone an existing sheet's values into a new tab
python3 {baseDir}/scripts/sheets_write.py \
  --token YOUR_SPREADSHEET_TOKEN \
  clone-sheet --source-sheet-id abc123 --title 'ClonedSheet' --clone-range 'A1:Z200'

Using a URL instead of --token

Both scripts accept --url to auto-extract the spreadsheet token:

python3 {baseDir}/scripts/sheets_write.py \
  --url 'https://xxx.larksuite.com/sheets/YOUR_SPREADSHEET_TOKEN?sheet=SHEET_ID' \
  write --range 'SHEET_ID!A1:B1' --values '[["hello","world"]]'

File Download

Download cloud files (PDF, documents, etc.) from Lark/Feishu Drive.

Get file token from the URL

Example URL:

https://.../file/YOUR_FILE_TOKEN

• file_token = YOUR_FILE_TOKEN

Download a file

# Download by URL (PDF files auto-extract text to .txt)
python3 {baseDir}/scripts/file_download.py \
  --url "https://.../file/YOUR_FILE_TOKEN" \
  --out /tmp/report.pdf

# Download by file token directly
python3 {baseDir}/scripts/file_download.py \
  --file-token YOUR_FILE_TOKEN \
  --out /tmp/report.pdf

# Force text extraction for non-.pdf files
python3 {baseDir}/scripts/file_download.py \
  --file-token YOUR_FILE_TOKEN \
  --out /tmp/document.bin --extract-text

Reading downloaded PDF content

When --out ends with .pdf, the script automatically:

1. Extracts text to a .txt file (e.g. /tmp/report.pdf → /tmp/report.txt)
2. Extracts embedded images to a _images/ directory (e.g. /tmp/report_images/img-000.png, ...)
3. If text is garbled/unreadable, renders each page as a PNG image to _pages/ directory for visual reading

Text extraction priority: pdfplumber → pypdf → pdftotext (poppler). All Python packages are auto-installed via pip on first use. Includes garbled-text detection — if extracted text is unreadable (e.g. scanned PDF, special fonts), pages are rendered to images automatically.

Image extraction priority: pypdf → pdfimages (poppler).

Page rendering (garbled fallback): pymupdf → pdf2image.

The typical workflow is:

1. Run the download script
2. If text is readable → read /tmp/report.txt with the Read tool
3. If text is garbled → read page images in /tmp/report_pages/ with the Read tool (AI vision)
4. Read embedded images in /tmp/report_images/ for charts, diagrams, etc.
5. Summarize / analyze the content

For non-PDF files, use --extract-text to force extraction.

Write Subcommands Reference

All subcommands support --dry-run to preview without executing.

Notes / Gotchas

• Range format: the API accepts "{sheetId}!A1:Z200" or "{sheetTitle}!A1:Z200".
• If you don't know the sheet title, use list-sheets first, or start with the sheet= value from the URL.
• Large sheets: export only the needed columns/rows first; widen the range iteratively.
• Values format: must be a JSON array of arrays (rows x columns), e.g. [["a","b"],["c","d"]].
• Secrets: the script reads appId/appSecret from ~/.openclaw/openclaw.json. Do not print or paste those credentials into chat.

Troubleshooting

• 403 / permission errors:
• Confirm the sheet/file has been shared with the app/bot identity.
• Confirm the Lark/Feishu app has the required permissions (Sheets read & write, Drive file download) enabled in the developer console.
• values_batch_get failed / values_batch_update failed with non-zero code:
• Most often a bad range string. Try a smaller range or verify the sheetId/title via list-sheets.
• addSheet failed:
• The title may already exist. Sheet titles must be unique within a spreadsheet.

External Endpoints

This skill makes outbound requests to the following Lark/Feishu OpenAPI endpoints only:

For Lark (international) users, the base URL is https://open.larksuite.com instead.

Security & Privacy

• Credentials are local. appId/appSecret are read from ~/.openclaw/openclaw.json and only sent to the official Feishu/Lark OpenAPI for token exchange.
• No data leaves the Feishu ecosystem. All read/write operations go through the official Lark OpenAPI.
• Scripts are sandboxed. They only access the OpenClaw config file and the target spreadsheet. No other files or environment variables are read.