ChromeManager Skill

Author: Devilflasher

Other languages: 中文

Disclaimer

1. This project is open source and intended for learning and communication only. Closed-source commercial use is prohibited.
2. Users must comply with local laws and regulations. Any illegal use is prohibited.
3. The developer is not liable for any direct or indirect loss caused by using this project.
4. Using this project means you have read and agreed to this disclaimer.

Overview

chromemanager-skill connects ChromeManager to external AI agents (such as OpenClaw and Hermes), so agents can call local APIs for window management and run webpage automation actions through debugPort.

This repository includes two standalone folders:

• chromemanager-skill-openclaw
• chromemanager-skill-hermes

Both skills provide the same core capabilities. The main differences are integration and trigger behavior.

Core Capabilities

• Use ChromeManager to batch open/import/select/navigate/close windows
• Execute webpage actions via debugPort: click/fill/press/extract/screenshot
• Support multi-window task orchestration and concurrency control
• Support multi-step run-plan workflows with automatic progress reporting

Shared base actions:

• prepare: open and prepare target windows
• snapshot: capture page structure and refs
• click: click a page element
• fill: fill text into an input
• press: press a keyboard key
• extract: extract text from the page
• screenshot: capture page screenshots
• desktop-screenshot: capture the full desktop
• report-progress: report status with optional screenshot
• validate-stage: verify page state
• recover: reload or re-navigate failed windows
• run-plan: run a multi-step workflow

Installation

Tip

Recommended for beginners: let your local AI agent install it

Send the following prompt to an AI agent installed on your computer that has permission to access local files:

Please install ChromeManager Skill:
https://github.com/devilflasher/chromemanager-skill

Read the repository README first, then follow the instructions that apply to your current environment to install the skill, install dependencies, and configure config.json.

Ask me for the ChromeManager local API token. Do not guess, generate, upload, or expose the token.

When finished, tell me the actual installation directory, available trigger command, and whether I need to restart or start a new session.

For both OpenClaw and Hermes editions:

1. Place the target skill folder under the agent skills directory
2. Enter that skill directory
3. Run:

npm install

4. Edit config.json and set:
   • api_token: API token generated in ChromeManager
   • api_host: API host address (usually 127.0.0.1)
   • api_port: API port (default 18923)
   • software_path: local install path of ChromeManager executable (Windows example: D:/***/***/ChromeManager.exe; macOS example: /Applications/***/ChromeManager.app)
   • autoStart: whether to auto-launch ChromeManager when not running (true/false)
5. Enable local API in ChromeManager and make sure the token matches api_token

OpenClaw and Hermes

OpenClaw

• Recommended: install into the active workspace with openclaw skills install <path-to-chromemanager-skill-openclaw>. OpenClaw installs there by default.
• Active workspace skill directory: <active-workspace>/skills/
  • With the conventional default workspace, this is C:\Users\<your-username>\.openclaw\workspace\skills\ on Windows and ~/.openclaw/workspace/skills/ on macOS or WSL.
• Shared global skill directory: C:\Users\<your-username>\.openclaw\skills\ on Windows, or ~/.openclaw/skills/ on macOS or WSL. Use openclaw skills install --global <path-to-chromemanager-skill-openclaw> to install there.
• Trigger style: use the automatically registered /cm command, or ask OpenClaw to use the ChromeManager skill in natural language.

Hermes

• Windows (local Hermes):
  • C:\Users\<your-username>\AppData\Local\Hermes\skills\
• Windows (Hermes running in WSL):
  • ~/.hermes/skills/
• macOS:
  • ~/.hermes/skills/
• For an external skill directory, add the directory to skills.external_dirs in Hermes config.yaml. This is %LOCALAPPDATA%\Hermes\config.yaml on Windows and ~/.hermes/config.yaml on macOS or WSL.
• Trigger style: use the automatically registered /cm command. Natural-language requests can also load the skill by intent.

Command Examples

You can ask for concise usage help directly:

/cm help
/cm 帮助

Important

For first-time setup:

• OpenClaw:
  1. Install chromemanager-skill-openclaw and use its automatically registered /cm command.
• Hermes:
  1. Install chromemanager-skill-hermes and start a new session if the skill is not yet listed. Hermes automatically registers /cm.

Single-line task example:

/cm prepare 1-5, open https://example.com, click "Check in", then close all pages

You can also split into steps:

/cm prepare 1-5
/cm open https://example.com
/cm click "Check in"
/cm close all pages

Concurrency Control

If users need to control concurrency, they can describe it directly in natural language:

• "run in batches of 2 windows"
• "run in batches of 3 windows"
• "run all in parallel"
• "run serially"

Recommendations:

• Increase concurrency for "open page" and "close page"
• Keep "click" and "input" more conservative to reduce page-state divergence

run-plan

Use run-plan for longer workflows, especially when you need:

• multi-step chains (for example: prepare -> navigate -> fill -> submit -> validate)
• step-to-step result reuse
• conditional branches, failure policies, and concurrency control

Default behavior:

• run-plan appends an automatic report-progress at the end
• autoReportOnPlanEnd default: true
• autoReportCapture default: desktop (set none to disable screenshots)
• autoReportStrict default: false (auto-report failure does not fail the whole plan)
• autoReportFailedQuestion lets you customize the follow-up question when blocked

Recommended practice:

• send one report-progress after each key stage
• when validation/recovery fails or wallet confirmation is blocked, screenshot first and report
• include: current stage, affected windows, result status, and next-step question

Join the NoBiggie XChat group for discussion: