Open-source guardrail
Switching off an agent's approvals or sandbox
AgentTrail Guard asks you about this by default: you decide before the call runs. Where an agent cannot ask, the call is refused instead.
- Default action
- Ask
- Severity
- High severity
- Library version
- 0.2.1 · Sep 29, 2026
What it does
What it catches, and what it misses
Written into the rule itself, next to what it matches, so you can judge it before you trust it.
Holds the flags that make a coding agent act without asking: Claude Code's --dangerously-skip-permissions, --allow-dangerously-skip-permissions and --permission-mode bypassPermissions; Codex's --dangerously-bypass-approvals-and-sandbox, --yolo, --full-auto, -a never or --ask-for-approval never, --sandbox danger-full-access, and the same two settings passed as approval_policy or sandbox_mode config values; Gemini's --yolo, -y and --approval-mode yolo; cursor-agent --force, -f or --yolo; and Aider's --yes-always, its accepted short form --yes, and AIDER_YES_ALWAYS. Deliberately NOT matched: the narrower modes (--permission-mode plan or acceptEdits, --sandbox workspace-write, -a on-request, --approval-mode auto_edit), and --auto-approve or -y on any other command — terraform apply --auto-approve and apt-get install -y are not agents. Misses Cursor's CLI under its primary name agent, which is too generic to match on, and a setting written to an agent's config file with a file tool, which fs.agent-self-config holds for Claude Code and Codex. A quoted MENTION is not a use: a search, a git commit -m message, an echo or a curl --data body that only names this command is left alone, as long as every shell metacharacter stays inside the quotes.
Examples
Tested on every build
The rule must match every command on the left and none on the right, or the library does not build. Catching the real thing is easy; staying quiet on the near-miss is the hard part.
Catches (14)
- claude --dangerously-skip-permissions
- claude -p "fix the build" --permission-mode bypassPermissions
- codex exec --dangerously-bypass-approvals-and-sandbox "migrate the schema"
- codex --yolo
- codex --full-auto
- codex -a never
- codex exec --sandbox danger-full-access "run the migrations"
- codex exec -c approval_policy=never "run the migrations"
- gemini --yolo -p "clean up the repo"
- gemini --approval-mode=yolo
- cursor-agent -p --force "update the dependencies"
- aider --yes-always --message "apply the refactor"
- export AIDER_YES_ALWAYS=true
- PowerShellclaude --dangerously-skip-permissions
Stays quiet on (15)
- git commit -m "docs: explain claude --dangerously-skip-permissions"
- grep -rn "claude --dangerously-skip-permissions" docs/
- echo "never run claude --dangerously-skip-permissions"
- curl --data "we ran claude --dangerously-skip-permissions" https://api.example.com/comments
- claude --permission-mode plan
- claude --permission-mode acceptEdits
- codex --sandbox workspace-write
- codex -a on-request
- gemini --approval-mode auto_edit
- terraform apply --auto-approve
- apt-get install -y curl
- aider --no-auto-commits src/app.py
- cursor-agent --help
- git push --force-with-lease origin feature/x
- pnpm run lint
In your agent
What "ask" means in each app
The guard runs as a hook in each app, and each app gives a hook different powers. Here is what this rule's default action does in each one.
- Claude Code
- You are asked before the call runs. If your settings.json already allows the tool outright, Claude Code runs it with no prompt.
- Cursor
- A terminal command that reaches the guard gets Cursor's approval card. Any other call that needs approval, such as a file edit or a read, is refused, because Cursor cannot ask there.
- Codex CLI
- Codex cannot ask, so the call is refused, and the reason says a person has to approve it.
Each app hands the guard a different set of calls: in a Cursor sandbox run mode, for one, some terminal commands run without reaching the guard at all. Read the notes for Cursor and for Codex CLI before you rely on a rule there.
Make it yours
Change what it does in one command
Turn it off, change its action, or silence it on one command shape. The narrow one is allow: the rule keeps catching everything else.
Run it locally
Put these guardrails in front of your agent.
AgentTrail Guard is free and open source. It checks every command and file change against the whole library before your agent runs it, on your machine, with no account.
npm i -g @agenttrail/guard