Command guardrails for AI coding agents

Stop clicking “Approve” on every command without going full --yolo. yolonot runs safe commands instantly and stops the risky ones.

$brew install sezaakgun/tap/yolonot && yolonot setup
Finds your agents and wires the hooks. Free and open source · Undo with yolonot uninstall

How it works

Your rules first

A .yolonot file in your repo says what is always fine and what never is.

Instant local check

A built-in bash parser clears read-only commands in microseconds, with no AI call.

AI only when unsure

The rest goes to a local or cloud model. Critical commands are never auto-allowed.

Try it: pick a command and see which step decides.

Command
Profile
$git status --short && git diff HEAD~1 | grep -E "TODO|FIXME"
ALLOW · Local AST (fast_allow) · <1ms (~1.6µs)

    Fewer risky commands slip through

    456 commands from 3 open benchmark suites. Lower is better.

    PROFILE / MODE 1. Risky Missed (FNR · 0–20%) 2. Safe Blocked (FPR · 0–20%) BORDERLINE POLICY yolonot [balanced] 4.9% 0.0% 1.4% ask (cached) yolonot [fast] 4.9% 1.4% 0.0% ask (headless) yolonot [strict] 1.6% 0.0% 19.4% ask (low/mod) yolonot [paranoid] 0.0% 8.4% 91.6% ask (lockdown) Claude Auto Mode‡ 17.0% 0.4% (8.5% S1) no ask state 0% 10% 20% 0% 10% 20% Unclassified baselines (off 0–20% scale): Manual Prompts ask on 100% of commands (93% habitual click); --yolo misses 100% of risks.
    ‡ Claude Auto Mode figures are from Anthropic's March 2026 report (52 real overeager actions), not from these suites. Methodology and full results

    Why not manual prompts or Auto Mode?

    Manual prompts Claude Auto Mode yolonot
    Repo rules
    ◐Per tool, by hand
    ✕Drops project allowlists
    ✓One .yolonot file
    Reads scripts before running
    ◐You open them
    ✕Command text only
    ✓Reads and pins them
    Remembers approvals
    ✕Asks every time
    ✕Stateless
    ✓Remembers, suggests rules
    Choice of model
    ✕You are the model
    ◐Anthropic only
    ✓Local or any cloud
    Works with
    ◐Each agent separately
    ✕Claude Code only
    ✓24 agents

    Works with 24 agents and counting, including Claude Code, Codex, Cursor, OpenCode, Antigravity and Copilot CLI. See all agents · Missing yours? Ask for it.

    FAQ

    Do I need an API key?
    No. Your rules and the local parser handle more than half of everyday commands with no API calls. For the rest, plug in a local model (Ollama) or any cloud provider.
    What if the model is slow or offline?
    yolonot never fails open. A timeout falls back to your agent's normal permission prompt, or to deny when no one is watching.
    Can I try it safely, or turn it off?
    yolonot check "rm -rf build/" shows the decision without running anything. yolonot pause turns protection off for a while, and yolonot uninstall removes the hooks and restores your settings.