Commands without asking
Approvals switched off, or a rule that pre-approves every shell command.
OWASP LLM06
Read the guide →NeuroShield guides
What your AI coding agents can do without asking
Claude Code, Cursor, Codex and Copilot read your files, run commands and load MCP servers, hooks and skills. Here's where each one keeps its settings, what in them lets it act on its own, and how to close it.
One guide per agent
CCClaude Code
Permission modes, allow rules, MCP servers, hooks, skills and the managed policy.
Read the guide →CUCursor
The CLI's allow rules, MCP servers, hooks, skills and extensions.
Read the guide →CXCodex CLI
Approval policy, sandbox mode, skills and profiles in config.toml.
Read the guide →COGitHub Copilot
Agent mode, tool auto-approval and the Copilot CLI.
Read the guide →VSVS Code
Chat tool approval, terminal auto-approve and MCP servers.
Read the guide →CDClaude Desktop
The MCP servers it starts and what each one can reach.
Read the guide →Seven ways an agent gets more reach than you meant
Each one is a fixed, published rule in NeuroShield, tied to the standard it comes from.
Keys within reach
API keys in .env files, MCP configs or folders the agent may read.
CWE-798OWASP LLM02
Read the guide →Reaches other machines
Pre-approved ssh or remote commands to your servers.
OWASP LLM06
Read the guide →Reads and sends
An add-on that reads a secret and sends it over the network.
CWE-200OWASP LLM02
Read the guide →Runs code it downloads
A hook or script that pipes a download straight into a shell.
CWE-494OWASP LLM03
Read the guide →Hidden instructions
Invisible characters in rules files, skills or CLAUDE.md.
CWE-1427OWASP LLM01
Read the guide →Known risky add-ons
An MCP server, skill or extension on our signed list of known ones.
OWASP LLM03
Read the guide →Or check all of it at once
NeuroShield, part of DriveScape One's Shields, reads every agent on your computer and each project they work in.
Read only
It reads settings files and never runs your agents, their MCP servers or their scripts.
Nothing leaves this computer
The check runs locally. No account, no upload, no AI inside: fixed, published rules.
Nothing changes without your OK
Where a safe fix exists, Fix It for Me makes it after you confirm, with a backup and Undo.
Questions
Is NeuroShield an AI?
No. It reads files and applies fixed rules we publish, so the same settings always give the same answer. Nothing is sent to a model or a server.
Does it work on Windows and Linux?
Yes. NeuroShield reads the same agents on macOS, Windows and Linux. GhostShield, the other half of Shields, is macOS only.
Will it change my settings?
Only when you press Fix It for Me on one item and confirm. It keeps a copy first, and Undo puts the file back.
See your own agents
The check is free: how many agents you have and what they can do. Details and fixes come with Shields.