Execute a shell command and return stdout, stderr, and exit code. Much faster than using Terminal+clipboard. Runs in /bin/zsh by default. Supports working directory (cwd), timeout, and stdin input. Use for file operations, system info, script execution, and any shell automation task.
AI agents invoke run_shell to trigger actions in MacWright. What it does depends on the arguments the agent supplies, and its effects often reach beyond the immediate call — builds kicked off, notifications sent, workflows started.
This tool allows execution of arbitrary shell commands in zsh, which can perform any operation a user could do at the terminal. This is Execute (not Destructive alone) because the destructiveness depends on the command argument, but the severity is critical due to potential for complete system compromise, data exfiltration, malware installation, or destructive operations like 'rm -rf /', given an AI agent could…
From the tool's definition Tool executes shell commands with support for working directory, stdin input, and timeout. Description explicitly states 'Execute a shell command' and 'script execution'. Capable of arbitrary shell automation including file operations and system-level actions.
Attacks that exploit this kind of access
Execute a shell command and return stdout, stderr, and exit code. Much faster than using Terminal+clipboard. Runs in /bin/zsh by default. Supports working directory (cwd), timeout, and stdin input. Use for file operations, system info, script execution, and any shell automation task. It is categorised as a Execute tool in the MacWright MCP Server, which means it can trigger actions or run processes. Use rate limits and argument validation.
Register the MacWright MCP server in PolicyLayer and add a rule for run_shell: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches MacWright. Nothing to install.
run_shell is a Execute tool with high risk. Execute tools should be rate-limited and have argument validation enabled.
Yes. Add a rate_limit block to the run_shell rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for run_shell. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
run_shell is provided by the MacWright MCP server (ruchit-p/macwright). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Every MCP server has a record like this.
Type a name, get the same breakdown: verified identity, auth posture, risk grade, capabilities, recommended policy.
Teams ship this data inside their own products. See what a licence covers →