iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A coding agent should run freely only inside a boundary you have chosen. Set both its approval policy—when it must pause—and its execution boundary—which files, network resources, and credentials it can reach. For routine work, keep access limited to the project and require approval when a task crosses that boundary. Use stronger isolation for untrusted code.
Approval prompts and sandboxing control different things
An approval policy governs the agent’s workflow: whether a command runs automatically, prompts you, or is denied. A sandbox governs what the command can access, even after it runs. OpenAI summarizes the distinction this way: “The sandbox defines which files and network resources ChatGPT can access.” OpenAI’s Codex sandbox documentation describes the sandbox boundary; its approval documentation covers when the agent pauses.
Neither a prompt telling the model to be careful nor an allowlist by itself is a substitute for an enforced boundary. Check the active settings for your particular product, client, operating system, version, and organization: available modes and defaults can differ.
Free tools Windows power users keep installed
One-click scans. No signup required.
Choose a practical baseline for routine project work
For ordinary work in a repository you trust, a useful starting point is to let the agent read and modify only the project, while requiring approval when it needs to act beyond that scope. This preserves useful autonomy for edits and routine commands without giving every command unrestricted access to your machine or network.
#1 Best Overall
- POWERFUL SECURITY KEY: The Security Key C NFC is the essential physical passkey for protecting your digital life from phishing attacks. It ensures only you can access your accounts.
- WORKS WITH 1000+ ACCOUNTS: Compatible with Google, Microsoft, and Apple. A single Security Key C NFC secures 100 of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your Security Key C NFC via USB-C and tap it, or tap it against your phone (NFC) to authenticate. No batteries, no internet connection, and no extra fees required.
- TRUSTED PASSKEY TECHNOLOGY: Uses the latest passkey standards (FIDO2/WebAuthn & FIDO U2F) but does not support One-Time Passwords. For complex needs, check out the YubiKey 5 Series.
- BUILT TO LAST: Made from tough, waterproof, and crush-resistant materials. Manufactured in Sweden and programmed in the USA with the highest security standards.
- Keep writable scope narrow. Prefer the project directory over your home directory or the whole filesystem.
- Make network access deliberate. Decide whether the agent needs outbound access and, where supported, limit destinations.
- Keep credentials out of reach where possible. Do not expose raw secrets to untrusted code simply to let it authenticate.
- Review changes and escalation requests. Inspect diffs and consider why a task needs access beyond its normal scope.
Escalate only for a specific need, such as installing a dependency or accessing a resource outside the project, and return to the narrower configuration afterward. For untrusted repositories or generated code, use an isolated environment rather than relying on prompts alone.
How Codex and Claude Code express permissions
These products use their own controls; similarly named concepts do not necessarily behave identically. The table summarizes the documented options, not every client’s defaults. Check the selected mode in your own environment.
Rank #2
- POWERFUL SECURITY KEY: The YubiKey 5 NFC is the most versatile physical passkey, protecting your digital life from phishing attacks. It ensures only you can access your accounts
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5 NFC secures 100+ of your favorite accounts, including email, password managers, and more
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5 NFC via USB and tap it, or tap it against your phone (NFC), to authenticate. No batteries, no internet connection, and no extra fees required
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it
- PRIMARY & SPARE KEYS: Just like having a spare house key, we recommend buying two YubiKeys - one for daily use and one as a spare. That way you’ll never get locked out of your accounts
| Product | Documented controls | Practical interpretation |
|---|---|---|
| Codex | Sandbox modes include read-only, workspace-write, and danger-full-access. Approval policies include on-request and never. |
OpenAI documents workspace-write plus on-request as a lower-risk local automation preset. danger-full-access together with never removes filesystem and network sandbox boundaries. Availability depends on client, execution environment, local configuration, and organization requirements. OpenAI Codex sandbox documentation; OpenAI Codex approval documentation. |
| Claude Code | Modes include default/manual, acceptEdits, plan, auto, dontAsk, and bypassPermissions. Rules can allow, ask, or deny actions. |
In manual mode, Bash commands generally prompt except for built-in read-only commands; reads in configured working directories do not. Rule precedence is deny, then ask, then allow. Anthropic says, “Permission rules are enforced by Claude Code, not by the model.” It recommends bypassPermissions only in isolated containers or VMs. Anthropic Claude Code permissions documentation. |
Use the product’s own documentation and interface to determine which setting is active. A mode label alone does not tell you the full filesystem, network, or credential boundary.
Why an allowlist is not a complete execution boundary
An allowlist can narrow which commands appear to run without approval, but it does not necessarily contain every way a process can influence later actions. In an article published August 18, 2026, Docker described CVE-2026-22708 as a Cursor Auto-Run allowlist bypass: shell built-ins including export, typeset, and declare could alter environment variables before a later approved command ran. Docker says Pillar Security disclosed the issue on January 14, 2026, that Cursor rated it High, and that it was patched in version 2.3. These are details reported by Docker, a vendor with a commercial interest in sandboxing, not an independent assessment of all allowlist implementations. Treat the patch information as time-sensitive. Docker’s account of CVE-2026-22708.
Rank #3
- POWERFUL SECURITY KEY: The YubiKey 5C NFC is the most versatile physical passkey, protecting your digital life from phishing attacks. It ensures only you can access your accounts
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5C NFC secures 100+ of your favorite accounts, including email, password managers, and more
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5C NFC via USB and tap it, or tap it against your phone (NFC), to authenticate. No batteries, no internet connection, and no extra fees required
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it
- PRIMARY & SPARE KEYS: Just like having a spare house key, we recommend buying two YubiKeys - one for daily use and one as a spare. That way you’ll never get locked out of your accounts
The practical lesson is to assess the boundary around the process, not just the command names in a permission rule. Consider what environment variables, files, sockets, and follow-on processes are accessible, and keep the tool patched.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What isolation changes—and what it does not
Docker’s documentation describes Sandboxes as microVMs with a separate kernel and a network policy boundary. That can provide a stronger separation than a command allowlist, but the actual mounted resources and network rules still matter. Docker Sandboxes documentation describes the configuration and its caveats.
Rank #4
- POWERFUL SECURITY KEY: The Security Key NFC is the essential physical passkey for protecting your digital life from phishing attacks. It ensures only you can access your accounts.
- WORKS WITH 1000+ ACCOUNTS: Compatible with Google, Microsoft, and Apple. A single Security Key NFC secures 100 of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your Security Key NFC via USB-A and tap it, or tap it against your phone (NFC) to authenticate. No batteries, no internet connection, and no extra fees required.
- TRUSTED PASSKEY TECHNOLOGY: Uses the latest passkey standards (FIDO2/WebAuthn & FIDO U2F) but does not support One-Time Passwords. For complex needs, check out the YubiKey 5 Series.
- BUILT TO LAST: Made from tough, waterproof, and crush-resistant materials. Manufactured in Sweden and programmed in the USA with the highest security standards.
- Workspace: A direct workspace mount is read-write, so changes are visible on the host. Clone mode instead uses a private clone.
- Network: Outbound TCP follows network policy; do not assume that every protocol or destination is handled the same way.
- Credentials: A forwarded SSH-agent socket can let processes request signatures without receiving private keys, but it still gives them a route to request authentication.
- Shared resources: Supported agents may share a skills store. Anything shared across the boundary can affect what the agent can read or change.
Isolation reduces exposure only to the extent that its configuration does. Review mounts, network policy, forwarded sockets, and shared stores before running untrusted work.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Quick Recap
Checklist before letting an agent run unattended
- Confirm the active approval mode and sandbox in the actual client or environment.
- Limit writable directories to the project or a disposable clone.
- Restrict outbound network access to what the task requires.
- Avoid exposing host credentials to untrusted repositories or commands.
- Check which rules apply, their scope, and whether organization policy constrains local settings.
- Review diffs and inspect hooks, shared files, sockets, and other resources that cross the boundary.
- Use stronger isolation for untrusted code, and ensure you can discard or restore the working environment.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
