← Back to Skills Marketplace
glass2claw
by
Jonathan Jing
· GitHub ↗
· v2.3.3
679
Downloads
1
Stars
0
Active Installs
21
Versions
Install in OpenClaw
/install glass2claw
Description
Ray-Ban glasses → voice command → WhatsApp → OpenClaw auto-routes your photo into the right database. Hands-free life logging.
README (SKILL.md)
glass2claw: From Your Eyes to Your Database — Instantly
🛠️ Installation
1. Ask OpenClaw (Recommended)
Tell OpenClaw: "Install the glass2claw skill." The agent will handle the installation and configuration automatically.
2. Manual Installation (CLI)
If you prefer the terminal, run:
clawhub install glass2claw
You're wearing your Meta Ray-Ban glasses. You see a wine label, a business card, a tea tin. You say:
"Hey Meta, take a picture and send this to myself on WhatsApp."
That's it. OpenClaw does the rest.
The photo lands in your WhatsApp. OpenClaw's Vision Router picks it up, classifies what it is, and writes a structured entry into the right database — wine cellar, contacts, tea collection, whatever you've set up.
No typing. No app switching. No friction.
📸 How It Works
Meta Ray-Ban glasses
→ "Hey Meta, take a picture and send this to myself on WhatsApp"
→ Meta AI delivers the photo to your WhatsApp
→ OpenClaw (WhatsApp session) receives the image
→ classifies intent: Wine | Tea | Contacts | Cigar | ...
→ routes to the matching specialist agent
→ writes structured entry to your database
Your only action is the voice command. Everything downstream is automatic.
🔧 What You Need to Set Up
This skill is a routing protocol — it defines the pattern, not the specific implementation. You bring your own:
- Meta AI + WhatsApp connection — enable Meta AI on your Ray-Ban glasses and link it to WhatsApp (one-time setup in the Meta View app)
- OpenClaw with WhatsApp channel — your OpenClaw instance needs a WhatsApp session to receive the incoming images
- Destination databases — connect whichever databases you want: Notion, Airtable, a local file, a Discord channel. The skill routes to wherever you configure it
- Database credentials — set up API access for your chosen database yourself (Notion API key, Airtable token, etc.)
The skill templates in this package show one reference implementation using Notion + Discord. Adapt them to your own stack.
🔒 Privacy
This skill processes photos from your personal camera. Images flow from WhatsApp → your OpenClaw instance → your configured destination. Any external services you connect (Notion, Discord, etc.) are governed by their own privacy policies. All routing logic runs on your own OpenClaw instance.
📦 What's Included
SAMPLE_AGENT.md— reference routing logic for the hub agentSAMPLE_SOUL_WINE.md— reference persona for a wine specialist agent
Use these as starting points. Customize for your own categories and destinations.
Created by JonathanJing | AI Reliability Architect
Usage Guidance
This package is an instruction-only routing template and appears coherent, but before installing: (1) confirm you control the OpenClaw instance and the WhatsApp session used to receive images (do not point it at someone else’s account); (2) only configure session keys that map to trusted destinations — the skill will forward image URLs to whatever session keys you supply; (3) store and supply destination DB credentials yourself and keep them secret (the skill does not request them, but downstream agents will need them); (4) review SAMPLE_AGENT.md and SAMPLE_SOUL_WINE.md and test with non‑sensitive images first to verify routing behavior; (5) be mindful that images may contain PII — only forward to services you trust (Notion, Discord, Airtable, etc.). If you need stronger assurance, ask the author for runnable code, a homepage, or an explicit privacy/security audit of how your OpenClaw instance handles forwarded media.
Capability Analysis
Type: OpenClaw Skill
Name: glass2claw
Version: 2.3.3
The skill bundle is designed for hands-free image logging and routing. Analysis of all files, including the agent instructions in `SKILL.md`, `SAMPLE_AGENT.md`, and `SAMPLE_SOUL_WINE.md`, reveals no evidence of malicious intent. The skill utilizes standard OpenClaw tools (`sessions_send`, `message`) for inter-agent communication and media handling. While `SAMPLE_AGENT.md` uses prompt chaining to instruct a specialist agent, the instructions are benign and directly related to the skill's stated purpose (e.g., 'Post the image to the channel, then analyze and log to your database'). The skill explicitly states that users must configure their own database credentials and that all routing logic runs on the user's OpenClaw instance, indicating no attempt at credential theft or unauthorized data exfiltration. No suspicious external calls, obfuscation, or persistence mechanisms were found.
Capability Assessment
Purpose & Capability
Name/description (vision → WhatsApp → OpenClaw → destination DB) matches the provided materials. The package contains only templates and instructions; it asks the user to provide the Meta/WhatsApp link, an OpenClaw WhatsApp session, and destination DB credentials—these are appropriate and expected for the stated routing functionality.
Instruction Scope
SKILL.md and sample files instruct the agent to detect incoming WhatsApp image messages, classify intent, and forward image URLs to configured specialist sessions using sessions_send and message tools. The scope stays within the described routing task, but it explicitly forwards image URLs to whatever session keys the user configures — so correct session-key configuration is critical. Minor clarity issue: the hub file says 'Do NOT describe or analyze the image yourself' while also instructing to 'classify the image intent from the URL context or any accompanying text' (this implies classification from metadata/text rather than pixel analysis).
Install Mechanism
No install spec or third‑party downloads are included; this is instruction-only. The README suggests using the platform's installer (clawhub) which is expected. Nothing in the package pulls arbitrary code or writes files.
Credentials
The skill declares no required environment variables or credentials. It reasonably expects the user to provide external connections (Meta/WhatsApp, OpenClaw session, destination DB API keys) but does not request or attempt to capture them itself. This is proportionate to the skill's purpose.
Persistence & Privilege
always is false and disable-model-invocation is default (agent may invoke autonomously, which is normal). The skill does not request elevated or permanent system presence and does not modify other skills or system‑wide configs in the provided files.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install glass2claw - After installation, invoke the skill by name or use
/glass2claw - Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.3.3
Forced re-publish with version bump to v2.3.3 to ensure UI refresh.
v2.3.2
Added simplified installation instructions to SKILL.md and README.md.
v2.3.1
Shorten description to fit preview
v2.3.0
Add missing SAMPLE_AGENT.md (hub routing logic) and SAMPLE_SOUL_WINE.md (wine specialist persona) — fixing package inconsistency flagged in audit
v2.2.0
Simplify: skill is now a protocol description, not a prescriptive implementation. Remove hardcoded config paths and env requirements. Users bring their own databases and credentials.
v2.1.0
Fix audit findings: clarify data flow (Discord+Notion are external services), remove misleading 'no data leaves infrastructure' claim, add Privacy & Data Handling section
v2.0.1
Fix voice command: full one-shot command is 'Hey Meta, take a picture and send this to myself on WhatsApp'
v2.0.0
Clarified Meta Ray-Ban glasses as primary use case; added voice command flow and Meta AI WhatsApp setup instructions
v1.5.0
Fix VT Code Insight flags: declare configPaths in metadata, explicit data flow diagram, replace vague 'private messenger context' with concrete Discord session routing, no code changes
v1.4.5
Retrigger VirusTotal scan — no code changes, instruction-only skill
v1.4.4
glass2claw v1.4.4
- No file changes detected in this version.
- Functionality and documentation remain unchanged from the previous release.
v1.4.3
glass2claw 1.4.3 refines setup instructions and simplifies requirements for OpenClaw integrations.
- Clarified the focus on logic-based routing, with updated, streamlined protocol description.
- Simplified environmental requirements (removed mandatory config file check in metadata).
- Updated configuration steps for better security and ease of use, emphasizing a single dedicated database ID file.
- Enhanced documentation to stress privacy, context-binding, and least privilege practices.
- Removed references to required files in metadata for greater flexibility.
v1.4.2
Version 1.4.2 – No code or documentation changes detected.
- No visible updates or modifications in this release.
- Skill behavior and configuration remain unchanged.
v1.4.1
glass2claw v1.4.1
- Added requirement for the presence of the `configs/vision_router.md` file in the OpenClaw integration metadata.
- Clarified the security guardrails section to reflect that enforcement is by agent instruction, and recommended using OpenClaw's `allowedDirectories` and `allowedTools` policies for enhanced security.
- Minor improvements to phrasing in the security documentation for better clarity.
v1.4.0
glass2claw 1.4.0
- Clarified that the hub channel uses a specialized task tag to identify intent.
- Updated privacy guidance to recommend securing captures with private servers/groups.
- Refined configuration instructions, emphasizing creation and naming of the dedicated config file.
- Limited the Agent’s context binding to only `configs/vision_router.md` for enhanced security.
- Streamlined instructions to focus on copying relevant templates for agent prompts.
v1.3.0
glass2claw 1.3.0 changelog:
- Refactored as a logic-only protocol, removing WhatsApp and Discord as formal requirements—works with any OpenClaw messenger integration.
- Updated setup instructions to focus on local configuration and privacy, simplifying and clarifying the process.
- Security guardrails streamlined: emphasized API-only communications and strict local context binding.
- Outdated implementation steps were replaced with configuration and template guidance for broader compatibility.
- Metadata updated: WhatsApp and Discord no longer explicitly required, only NOTION_API_KEY is needed.
v1.2.0
Summary: Improved security, privacy, and operational guardrails for glass2claw.
- Enhanced privacy instructions: Enforce use of a private Discord server and specify encrypted channel data flow.
- Added requirement to store media temporarily in the local `~/.openclaw/media` directory before processing.
- Updated security guardrails: External calls restricted solely to `api.notion.com`; internal messaging is limited to authorized OpenClaw sessions.
- Introduced "context locking," strictly forbidding file access outside the workspace and the approved config file.
v1.1.1
**glass2claw v1.1.1 Changelog**
- Enhanced security guidance: Added sections on least privilege Notion integration and running in a private Discord server.
- Updated configuration: Agent now reads Notion database IDs strictly from a local `configs/vision_router.md` file, instead of broad environment search.
- Protocol clarification: Simplified architecture steps and reworded instructions for clarity.
- Expanded metadata: Now explicitly documents required WhatsApp and Discord channels.
- Introduced explicit security guardrails limiting Agent actions (API communication, database insertion protocols, local processing only).
- Minor description and documentation refinements for conciseness and practical setup.
v1.1.0
glass2claw 1.1.0 introduces a flatter, more voice-driven setup and new template structure.
- Renamed core sample files to all-caps (`SAMPLE_AGENT.md`, `SAMPLE_SOUL_WINE.md`) for clarity and standardization.
- Updated setup steps for a simplified, flat configuration—apply provided templates directly to Discord and agent prompts.
- Expanded documentation to explicitly support Meta Ray-Ban Smart Glasses and hands-free voice commands.
- Architecture section now includes a clear mermaid diagram and improved, step-by-step overview.
- Enhanced description and rationale, with stronger emphasis on customization via Markdown and context-awareness.
- Deprecated previous sample files (`sample_agent.md`, `sample_soul.md`).
v1.0.1
glass2claw v1.0.1
- Major redesign: migrated from script-based router to a system-centric, documentation-driven workflow.
- Added setup and architecture guide for integrating WhatsApp, Discord channels, and Notion databases using markdown configuration files.
- Introduced Discord channel roles (Ingress/Hub/Specialists) for modular image processing and logging.
- Sample agent and specialist configuration markdown files included for rapid onboarding.
- Environment now requires only the NOTION_API_KEY.
Metadata
Frequently Asked Questions
What is glass2claw?
Ray-Ban glasses → voice command → WhatsApp → OpenClaw auto-routes your photo into the right database. Hands-free life logging. It is an AI Agent Skill for Claude Code / OpenClaw, with 679 downloads so far.
How do I install glass2claw?
Run "/install glass2claw" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is glass2claw free?
Yes, glass2claw is completely free (open-source). You can download, install and use it at no cost.
Which platforms does glass2claw support?
glass2claw is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created glass2claw?
It is built and maintained by Jonathan Jing (@jonathanjing); the current version is v2.3.3.
More Skills