What it does
ClawX is a desktop app that provides a graphical interface for OpenClaw AI agents. It turns CLI-based AI orchestration into a desktop experience without using the terminal. China website is https://clawx.com.cn.
- Connects Instruction files the AI reads, with steps that call an outside service
- Account or key Some steps need a key for GH_TOKEN (per the README)
- Category Automation and agents
For makers: add a badge to your README
If you made this, paste this code into your README. The badge links back to this page.
[](https://mcp.popnx.com/en/skills/valuecell-ai-clawx)Similar Skills
Same kind of work, ordered like the Skills list
1 skills (instruction sets) by ValueCell-ai that an AI reads and follows · not an MCP server
How a Skill differs from MCP
MCP = the hands
Connects the AI to an outside service or data, such as reading Notion or creating an invoice. A program or server has to be running.
Skill = the manual
A text file that teaches the AI how to do a task. It connects to nothing.
Some overlap: a few Skills run scripts that call tools, and a few MCP servers only send knowledge. Start from the task you want done.
Works with
Cursor
From the maker’s README and the folders in the repo
Install
Is it safe?
- Who made it
GitHub account ValueCell-ai (organization), 1 years old
- Use and upkeep
★ 7,596 stars · code last changed Nov 16, 2026
- Any scripts?
No, plain text files the AI reads and follows
- Uses the internet?
No scripts, so it does not go online by itself
- Changes machine-wide settings?
No hooks or writes to machine settings found
Checked automatically from the repo files at commit bc89906 (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
First: which AI do you use?
Pick your app and this page shows only how to install it there
Option 1Upload to claude.ai (web and app)
Uploading to claude.ai would fail: the description is over 200 characters (202). Use "Let the AI install it" instead
Option 1Let Claude Code install it
Copy this sentence and paste it into a Claude Code chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Restart Claude Code, then type /computer-use or ask for a matching task and Claude picks it up
Checked against official docs Nov 12, 2026 · Source: code.claude.com
For Terminal users
A Skill is instructions the AI will follow. Read the files before installing, and install only from makers you trust.
Open Terminal
Mac: press Command+Space and type Terminal · Windows: the command below does not work in PowerShell; use “Let AI install it” above instead
Copy the skill folder to ~/.claude/skills
This downloads the repo to a temporary folder and copies only the folders that contain a SKILL.md
git clone --depth 1 https://github.com/valuecell-ai/clawx.git /tmp/popnx-skill-valuecell-ai-clawx-bc89906 && mkdir -p ~/.claude/skills && cp -R '/tmp/popnx-skill-valuecell-ai-clawx-bc89906/resources/skills/computer-use' ~/.claude/skills/computer-useRestart Claude Code, then type /computer-use or ask for a matching task and Claude picks it up
ChatGPT desktop app
The ChatGPT desktop app has Codex built in, and Codex reads skills from ~/.agents/skills
Option 1Let ChatGPT install it
Open Codex in the ChatGPT desktop app and paste this sentence into the chat
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Option 2Put the folder there yourself
Put the computer-use folder at ~/.agents/skills/computer-use
~ is your home folder ($HOME)
For Terminal users
A Skill is instructions the AI will follow. Read the files before installing, and install only from makers you trust.
Open Terminal
Mac: press Command+Space and type Terminal · Windows: the command below does not work in PowerShell; use “Let AI install it” above instead
Copy the skill folder to ~/.agents/skills
This downloads the repo to a temporary folder and copies only the folders that contain a SKILL.md
git clone --depth 1 https://github.com/valuecell-ai/clawx.git /tmp/popnx-skill-valuecell-ai-clawx-bc89906 && mkdir -p ~/.agents/skills && cp -R '/tmp/popnx-skill-valuecell-ai-clawx-bc89906/resources/skills/computer-use' ~/.agents/skills/computer-use
After installing
- ChatGPT desktop app: open Skills in the sidebar to see that the skill is there
- To use it: in ChatGPT type @ and pick the skill · in Codex CLI or the IDE extension run /skills or type $ and the skill name
- Codex picks up newly installed skills by itself; if one does not show up, restart Codex
Checked against official docs Nov 15, 2026 · Source: learn.chatgpt.com
ChatGPT on the web and mobile
Skills you install yourself do not work here; only skills that come with a plugin from the directory
To use this skill with ChatGPT, use the ChatGPT desktop app as above
Checked against official docs Nov 15, 2026 · Source: learn.chatgpt.com
Option 1Let Codex install it
Copy this sentence and paste it into a Codex chat · Works in Codex inside the ChatGPT desktop app, Codex CLI and the IDE extension
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex picks up newly installed skills by itself; if one does not show up, restart Codex · To use it: in ChatGPT type @ and pick the skill · in Codex CLI or the IDE extension run /skills or type $ and the skill name
Checked against official docs Nov 15, 2026 · Source: learn.chatgpt.com
For Terminal users
A Skill is instructions the AI will follow. Read the files before installing, and install only from makers you trust.
Open Terminal
Mac: press Command+Space and type Terminal · Windows: the command below does not work in PowerShell; use “Let AI install it” above instead
Copy the skill folder to ~/.agents/skills
This downloads the repo to a temporary folder and copies only the folders that contain a SKILL.md
git clone --depth 1 https://github.com/valuecell-ai/clawx.git /tmp/popnx-skill-valuecell-ai-clawx-bc89906 && mkdir -p ~/.agents/skills && cp -R '/tmp/popnx-skill-valuecell-ai-clawx-bc89906/resources/skills/computer-use' ~/.agents/skills/computer-useRestart Codex, then ask for a matching task; Codex lists the skill
Option 1Let Cursor install it
Copy this sentence and paste it into a Cursor chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
In Cursor’s Agent chat, type / and pick computer-use
Checked against official docs Nov 12, 2026 · Source: cursor.com
For Terminal users
A Skill is instructions the AI will follow. Read the files before installing, and install only from makers you trust.
Open Terminal
Mac: press Command+Space and type Terminal · Windows: the command below does not work in PowerShell; use “Let AI install it” above instead
Copy the skill folder to ~/.cursor/skills
This downloads the repo to a temporary folder and copies only the folders that contain a SKILL.md
git clone --depth 1 https://github.com/valuecell-ai/clawx.git /tmp/popnx-skill-valuecell-ai-clawx-bc89906 && mkdir -p ~/.cursor/skills && cp -R '/tmp/popnx-skill-valuecell-ai-clawx-bc89906/resources/skills/computer-use' ~/.cursor/skills/computer-useIn Cursor’s Agent chat, type / and pick computer-use
Option 1Let the AI install it
For other apps that read skills. The sentence gives the paths for Claude Code, Codex and Cursor so the AI picks the one for your app · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs Nov 12, 2026 · Source: code.claude.com, learn.chatgpt.com, cursor.com
For Terminal users
Other apps that support Agent Skills (such as Gemini CLI or GitHub Copilot) use different folders. Check that app’s guide, then copy the folder with the SKILL.md there.
Skill folders checked against the official Claude Code, Codex and Cursor guides on Nov 11, 2026
✓ Commands checked Nov 17, 2026
View the original SKILL.md
The text below comes from the maker’s repo, not from PopNx MCP
Showing the first 8,000 characters · Open the full file on GitHub
---
name: computer-use
description: Operate a local app or desktop UI through ClawX's bundled native CUA CLI, or when the user selects /computer-use. Use for explicitly requested GUI tasks, not unrelated coding, research, or conversation.
---
# Computer Use in ClawX
Based on the official CUA 0.37.0 accompanying Skill. **This entrypoint takes precedence
over every upstream document and example**, including [UPSTREAM-SKILL.md](UPSTREAM-SKILL.md).
## Host Boundary
- Electron Main owns the 0.37.0 SDK daemon, restarts, OS grants, permission-policy and
telemetry configuration. Selecting this Skill does not enable Computer Use. Never
install/update a driver, run `serve`, daemon `stop`, `autostart`, `permissions grant`,
MCP setup, config/telemetry changes, or `--permission-mode`/`--capability-manifest` flags.
Do not run standalone diagnostics or remote/Fleet endpoints; use only the host descriptor.
- If unavailable, ask the user to enable Developer Mode in Settings > Advanced, open
Computer Use in the sidebar, review status, and opt in themselves. On macOS they must
explicitly use Request Permissions and review Accessibility/Screen Recording in
System Settings; a ClawX restart may be needed. Never operate OS grant dialogs.
- Use existing local `exec` and image-capable `read`, not the old `computer` tool,
an action wrapper, MCP server, or another UI automation channel. Do not loosen exec
approvals, sandbox, workspace-only access, or global image settings. If exec cannot
reach the host binary/endpoint or read cannot access host images, report the limitation.
- Shipped platforms: macOS 13+ Intel/Apple silicon and Windows 10+ x64 only.
[LINUX.md](LINUX.md), [EMBEDDING.md](EMBEDDING.md), and [README.md](README.md) are
preserved upstream references, not additional ClawX support or setup promises.
## Bootstrap Once Per Workflow
Read the live file named by `CLAWX_CUA_CONNECTION_FILE` at task start, not before every action.
Its shape is `{"v":2,"generation":"550e8400-e29b-41d4-a716-446655440000","driverVersion":"0.37.0","binaryPath":"...","socketPath":"..."}`.
Require v=2, driverVersion=0.37.0, a UUID generation, absolute bundled executable and explicit
absolute socket path (Windows: named pipe). Missing/malformed/incompatible descriptors block
the workflow. Never guess paths, use PATH's driver, or source/eval this data as shell code.
POSIX (macOS): print and inspect, then use the decoded paths. No jq, Node, or Python needed:
```sh
: "${CLAWX_CUA_CONNECTION_FILE:?Computer Use connection unavailable; check ClawX}"
cat < "$CLAWX_CUA_CONNECTION_FILE"
```
Replace the two placeholder paths below with the descriptor's values, shell-quoted as
single arguments (escape any embedded apostrophe). Choose a unique public workflow label,
not `default`, and repeat the same label on every call that accepts `session`, including
observations, actions, verification, and cleanup. In 0.37.0 `cli.rs` uses `cli-explicit` for
named CLI sessions across separate exec calls; anonymous calls are disposable. Keep this
host-owned CLI path; do not switch to persistent MCP. Labels confer no authorization
and cannot adopt another daemon generation or transport kind. Never set reserved session fields.
```sh
'/absolute/path/from/binaryPath' --socket '/absolute/path/from/socketPath' call list_windows '{"session":"clawx-review-20260909-01"}'
'/absolute/path/from/binaryPath' --socket '/absolute/path/from/socketPath' describe get_window_state
```
PowerShell: parse, validate, invoke via `&`, then inspect discovery before choosing a window:
```powershell
$ErrorActionPreference = 'Stop'
if (-not $env:CLAWX_CUA_CONNECTION_FILE) { throw 'Computer Use connection unavailable; check ClawX' }
$c = Get-Content -LiteralPath $env:CLAWX_CUA_CONNECTION_FILE -Raw | ConvertFrom-Json
if ($c.v -ne 2 -or $c.driverVersion -ne '0.37.0' -or $c.generation -notmatch '^[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$' -or $c.binaryPath -notmatch '^[A-Za-z]:\\' -or $c.socketPath -notlike '\\.\pipe\*') { throw 'Invalid ClawX CUA descriptor' }
if (-not (Test-Path -LiteralPath $c.binaryPath -PathType Leaf)) { throw 'Bundled CUA executable unavailable' }
$OutputEncoding = [System.Text.UTF8Encoding]::new($false)
@{ session = 'clawx-review-20260909-01' } | ConvertTo-Json -Compress -Depth 10 | & $c.binaryPath --socket $c.socketPath call list_windows
```
PowerShell 5.1 can mangle quoted JSON argv and encode piped Unicode as ASCII: use UTF-8
JSON on stdin as above and **omit the positional JSON argument**, including `{}` (it wins
over stdin). Use `ConvertTo-Json -Depth 10` for nested targets/predicates, not hashtable text.
Exec may use new shells: restore inspected literal paths/session or repeat bootstrap;
do not assume `$c` survives. After unavailability/restart, reread the descriptor and invalidate
observations, tokens, browser refs and session assumptions even if the endpoint is unchanged.
On Windows, `exec` may already run PowerShell. Never nest `powershell -Command "..."`
around `$env:...` or `$c...`: the outer shell expands those variables and produces parser
errors. Write a UTF-8 `.ps1` or JSON file and invoke it with `powershell -File`; keep request
JSON on stdin with no positional JSON argument. Bound observation commands (30s is enough for
state/screenshot; reserve 120s for bounded scripts). For Chromium/Electron screenshot-only
checks use `include_accessibility_tree:false` and a small `max_elements`; if
`parse_visual_regions` returns `code: "not_installed"`, use screenshot plus `read` or UIA once,
then stop instead of retrying an unavailable extension.
## Native Observe, Act, Verify
- Name the app, goal and postcondition. Read [UPSTREAM-SKILL.md](UPSTREAM-SKILL.md) and
[MACOS.md](MACOS.md) or [WINDOWS.md](WINDOWS.md) on demand, not every reference at once.
- The full native tool surface is available subject to host permissions, not a ClawX
action subset: use `list-tools`/`describe TOOL` on the descriptor binary for exact schemas.
Translate upstream calls to that absolute binary with `--socket` and your named session.
Select exact `(pid, window_id)` from `list_windows`/`launch_app`; get fresh `get_window_state`.
Prefer snapshot-bound `element_token`, native menus (`invoke_menu`) and exact geometry.
Foreground delivery needs user authorization; even menus may change focus on some platforms.
- Select a window or desktop `target` per action. `capture_scope` is retired; there is no
`deescalate_session`. The standalone `screenshot` was removed:
use `get_window_state` or explicitly authorized `get_desktop_state`, then read its image.
Cursor themes no longer accept `cursor_id`/legacy artwork; leave host configuration alone.
- For browser page tasks only, load [BROWSER.md](BROWSER.md): bind the exact native window,
prepare only when needed/authorized, use `get_browser_state` and typed page operations,
and refresh stale refs. `end_session` now restores owned Chromium debugging/consent state;
inspect cleanup failures rather than assume restoration. Never enable recording or replay automatically.
- Recording is unavailable through ClawX's one-shot CLI workflow: in 0.37.0
`get_recording_state` and `stop_recording` do not accept a public `session` and need
the owning persistent connection. Do not use `recording start/status/stop` or start
recording through `call start_recording`; report this limitation instead of adding MCP.
[RECORDING.md](RECORDING.md) is reference only, not a supported ClawX recording recipe.
- Only for continuation/recall, use advertised `history_status` then a bounded `history_query`;
denied/absent/unhealthy history is optional, never a reason to change host settings.
- Use bounded trees and `verify_state` predicates; schema-check filters and read large results from files.
`effect`, `route`, `evidence`, and `escalation` describe action facts, not task success.
Verify `satisfied` against the intended postcondition; `unsatisDiscussion
Ask, share a tip or report that it does not work. Sorted by what people marked helpful.
No posts yet. Be the first to ask or share.
Posts here are written by users, not the PopNx team.
From GitHub
- GitHub stars
- 7,596
- Maker
- ValueCell-ai
- License
- MIT
- Last code change
- Nov 16, 2026
- SKILL.md files in the repo (including translated copies and templates)
- 1
- Topics
- agent, agentic-ai, agents, ai, clawdbot, moltbot, openclaw, skill