360 Secure SkillHub
360 安全技能中心

Remarkable

该 Skill 用 ddvk rmapi 从 reMarkable 平板云端拉取手写笔记、草图和绘图(PDF/PNG),按内容类型分流:手写文字交由视觉模型解读后写入 memory 或日志,草图用 AI 图像编辑工具精修后返回,并支持按文件夹、标签或星标批量取件。

coolmanns
coolmanns
热度 92.08
最近更新 2026-05-17 11:25:00
分类:知识依赖:代码执行 / 读写文件 / 第三方 API / 内部调用大模型应用场景:笔记整理 / 图像生成 / 工作流 / 文件转换

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add remarkable-tablet
Metadata
name
remarkable
description
Fetch handwritten notes, sketches, and drawings from a reMarkable tablet via Cloud API (rmapi). Process content by refining artwork with AI image generation, extracting handwritten text to memory/journal, or using sketches as input for other workflows. Use when working with reMarkable tablet content, syncing handwritten notes, processing sketches, or integrating tablet drawings into projects.

reMarkable Tablet Integration (rmapi)

Fetch handwritten notes, sketches, and drawings from a reMarkable tablet via Cloud API. Process them — refine artwork with AI image generation, extract text to memory/journal, or use as input for other workflows.

Typical Use Cases

  1. Journal entries — User writes thoughts on reMarkable → fetch → OCR/interpret → append to memory/YYYY-MM-DD.md or a dedicated journal file
  2. Sketch refinement — User draws a rough graphic → fetch → enhance with nano-banana-pro (AI image editing) → return polished version
  3. Brainstorming/notes — User jots down ideas, lists, diagrams → fetch → extract structure → add to project docs or memory
  4. Illustrations — User creates hand-drawn art → fetch → optionally stylize → use in blog posts, social media, etc.

Processing Pipeline

reMarkable tablet → Cloud sync → rmapi fetch → PDF/PNG
                                                  ↓
                                    ┌─────────────┴─────────────┐
                                    │                           │
                              Text content?               Visual/sketch?
                                    │                           │
                              OCR / interpret            nano-banana-pro
                                    │                     (AI enhance)
                                    │                           │
                              Add to memory/            Return refined
                              journal/project            image to user

Setup

  • Tool: rmapi (ddvk fork) v0.0.32
  • Binary: ~/bin/rmapi
  • Config: ~/.rmapi (device token after auth)
  • Sync folder: ~/clawd/remarkable-sync/

Authentication (ONE-TIME)

  1. User goes to https://my.remarkable.com/connect/desktop
  2. Logs in, gets 8-character code
  3. Run rmapi and enter the code
  4. Token saved to ~/.rmapi — future runs are automatic

Commands

# List files/folders
rmapi ls
rmapi ls --json

# Navigate
rmapi cd "folder name"

# Find by tag / starred / regex
rmapi find --tag="share-with-gandalf" /
rmapi find --starred /
rmapi find / ".*sketch.*"

# Download (PDF)
rmapi get "filename"

# Download with annotations rendered (best for sketches)
rmapi geta "filename"

# Bulk download folder
rmapi mget -o ~/clawd/remarkable-sync/ "/Shared with Gandalf"

Sharing Workflows

Option A: Dedicated Folder

User creates "Shared with Gandalf" folder on reMarkable → moves items there → agent fetches with rmapi mget

Option B: Tag-Based

User tags documents with share-with-gandalf → agent discovers with rmapi find --tag

Option C: Starred Items

User stars items → agent fetches with rmapi find --starred

Fetch Script

# Fetch from shared folder
~/clawd/scripts/remarkable-fetch.sh

# Fetch starred items
~/clawd/scripts/remarkable-fetch.sh --starred

# Fetch by tag
~/clawd/scripts/remarkable-fetch.sh --tag="share-with-gandalf"

Notes

  • Tablet must cloud-sync before files are available
  • geta renders annotations into PDF (preferred for handwritten content)
  • Use convert (ImageMagick) to go from PDF → PNG for image processing
  • For text extraction, interpret the handwriting visually (vision model) rather than traditional OCR