claw-shell

Routes shell commands through a dedicated, guarded tmux session with output capture.

78|3|Updated Mar 25, 2026
One-click install
npx skills add https://github.com/CUHK-AIM-Group/NeuroClaw --skill claw-shell
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: claw-shell
Source: https://github.com/CUHK-AIM-Group/NeuroClaw/tree/main/skills/claw-shell
Command: npx skills add https://github.com/CUHK-AIM-Group/NeuroClaw --skill claw-shell

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Running system-level shell commands directly inside AI agents is risky and brittle. claw-shell provides a centralized, safe gateway that routes all shell operations through a dedicated tmux session, captures output, and enforces safety rules.

Core Features & Use Cases

  • Central tmux session named claw to isolate and persist shell interactions.
  • Safety guardrails that require explicit confirmation for dangerous commands and prevent cross-session interference.
  • Output capture and return to the calling skill or agent for reliable feedback.
  • Supports environment setup steps and long-running tasks without freezing the main workflow.
  • Widely used by NeuroClaw components such as freesurfer-tool, dependency-planner, and git-essentials.

Quick Start

Run a shell command through claw-shell to execute inside the claw tmux session and receive the output.

Frequently Asked Questions about claw-shell

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I safely execute shell commands through a tmux session in an AI workflow?

Executing shell commands safely through a tmux session requires a centralized gateway that isolates interactions, captures output, and enforces safety rules. This approach routes operations through a dedicated, persistent tmux session to ensure controlled, auditable execution.

Why does running system-level shell commands directly inside AI agents cause issues?

Running system-level shell commands directly inside AI agents is risky and brittle because it lacks isolation and safety guardrails. A centralized gateway prevents cross-session interference, gates dangerous commands requiring explicit confirmation, and captures output for reliable feedback.

Can I run long-running processes and environment setup steps without freezing the main workflow?

Yes, you can run long-running processes and environment setup steps without freezing the main workflow by routing them through a dedicated tmux session. The session persists shell interactions independently, allowing the main workflow to continue operating.

What's the best way to prevent cross-session interference when automating devops tasks?

Preventing cross-session interference when automating devops tasks is best achieved by routing all shell operations through a centralized, permission-guarded tmux session. This isolates interactions, enforces safety guardrails, and prevents concurrent processes from disrupting each other.

Do I need explicit confirmation to execute dangerous shell commands in an automated environment?

Yes, explicit confirmation is required to execute dangerous shell commands in an automated environment. Safety guardrails gate dangerous commands, preventing accidental execution while still allowing standard file operations and environment setup to proceed normally.

How does output capture work when running shell commands through a dedicated tmux session?

Output capture works by routing shell commands through a dedicated tmux session that records the execution results and returns them to the calling skill or agent. This ensures reliable feedback for subsequent operations without losing terminal output.