← Back to Skills Marketplace
claw-shell
by
imaginelogo
· GitHub ↗
· v1.0.0
13572
Downloads
8
Stars
304
Active Installs
1
Versions
Install in OpenClaw
/install claw-shell
Description
Runs shell commands inside a dedicated tmux session named claw, captures, and returns the output, with safety checks for destructive commands.
Usage Guidance
Install only if you intentionally want the agent to have local shell access. Use it in a disposable or sandboxed environment, approve commands carefully, avoid displaying secrets in the tmux session, and inspect or kill the claw tmux session when finished.
Capability Analysis
Type: OpenClaw Skill
Name: claw-shell
Version: 1.0.0
This skill is classified as suspicious due to its core functionality of executing arbitrary shell commands via `tmux send-keys` in `handler.js`. While the `SKILL.md` and `handler.js` include basic client-side checks for 'dangerous' commands (e.g., `sudo`, `rm`, `reboot`) and instruct the AI agent to seek user confirmation, these checks are easily bypassable (e.g., by using `sh -c 'rm -rf /'`). The ability to run any shell command grants broad system access, enabling potential data exfiltration, persistence, or other malicious activities, even if not explicitly coded into the skill itself.
Capability Assessment
Purpose & Capability
The purpose is coherent and disclosed: run shell commands in a dedicated tmux session and return output. However, that is broad local command execution with the user's privileges, including the ability to read or modify files, access network tools, and start processes.
Instruction Scope
SKILL.md says dangerous commands require user confirmation, but handler.js only uses a narrow substring denylist and has no robust approval boundary or allowlist for mutating commands.
Install Mechanism
The artifact has only SKILL.md and handler.js, with no install script or hidden setup. It depends on tmux, but the package metadata does not clearly declare that dependency.
Credentials
A shell runner naturally needs command execution, but this implementation is not sandboxed, not limited to a working directory, and sends user-controlled command text through execSync-based shell interpolation before tmux receives it.
Persistence & Privilege
The skill creates or reuses a persistent tmux session named claw and captures the last 200 pane lines, so long-running processes and prior terminal output can persist across tool calls.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install claw-shell - After installation, invoke the skill by name or use
/claw-shell - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
claw-shell 1.0.0 initial release:
- Provides an interface to run shell commands exclusively inside a tmux session named claw.
- Always creates or attaches to the session, ensuring isolation from other tmux sessions.
- Captures and returns command output to the agent.
- Includes safety checks: blocks potentially dangerous commands (e.g., sudo, rm, shutdown) unless user confirms.
- Simple, single-tool interface (`claw_shell_run`) with clear command input and safety guidance.
Metadata
Frequently Asked Questions
What is claw-shell?
Runs shell commands inside a dedicated tmux session named claw, captures, and returns the output, with safety checks for destructive commands. It is an AI Agent Skill for Claude Code / OpenClaw, with 13572 downloads so far.
How do I install claw-shell?
Run "/install claw-shell" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is claw-shell free?
Yes, claw-shell is completely free (open-source). You can download, install and use it at no cost.
Which platforms does claw-shell support?
claw-shell is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created claw-shell?
It is built and maintained by imaginelogo (@imaginelogo); the current version is v1.0.0.
More Skills