yolonot checks every shell command before your coding agent runs it. Safe commands run instantly; risky ones are blocked or sent to you.
- Your rules first: a git-tracked
.yolonot file decides 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.
Choose how cautious it is with four risk profiles (benchmark results).
Quick start
# Install via Homebrew (macOS / Linux) or Go
brew install sezaakgun/tap/yolonot
# go install github.com/sezaakgun/yolonot@latest
# Auto-detect installed harnesses and wire hooks, models & judge chain
yolonot setup
# Import existing permissions from Claude, Cursor, Codex, OpenCode, etc. into .yolonot
yolonot rules discover
Inspect or tune your active policy anytime:
yolonot status --all # active profile, judge chain, rules & hooks
yolonot check "git status" # dry-run a command through the full pipeline
yolonot profile use fast # switch between balanced, fast, strict, paranoid
yolonot suggest --list # promote recurring approvals into .yolonot rules
yolonot doctor && yolonot audit # verify hook health, latency & security rules
.yolonot policy
Drop a .yolonot file in your repository root (deny > ask > allow; evaluated in <1ms before any model call):
allow-cmd go test*
allow-path scripts/check.sh
ask-cmd git push*
deny-cmd *rm -rf /*
sensitive .env*
Safety scope: allow-cmd matches single commands only (chaining && or | voids static allow); critical risk never maps to allow; sensitive paths and ~/.ssh / ~/.aws are never attached to LLM prompts. yolonot gates shell tool calls—it is not an OS sandbox. See docs/security.md.
Docs
| Topic | File |
|---|
| How the pipeline decides — layers, hook ordering, pausing, dry-run | docs/how-it-works.md |
All CLI + /yolonot skill commands | docs/commands.md |
.yolonot rule files — format, walk-up, messages, redirects, sensitive files | docs/rules.md |
Risk profiles (fast / balanced / strict / paranoid) & risk tiers | docs/risk-profiles.md, docs/risk-tiers.md |
Judges (fast_allow, Dippy, LLM council) & model catalog / escalation | docs/judges.md, docs/models.md |
Eval suite & published benchmark results (N = 456 + 5,000 real-world calls) | docs/eval.md |
24 supported harnesses — install flags, --mod, per-host behavior | docs/harnesses.md |
| Configuration, schema, examples & prompt customization | docs/configuration.md, docs/prompt.md |
Analytics (log, stats, suggest) & architecture | docs/analytics.md, docs/architecture.md |
Uninstall
yolonot uninstall # removes hooks + skill, preserves ~/.yolonot/
Acknowledgements
The built-in fast_allow judge is inspired by Dippy by Lily Dayton (MIT-licensed) — ported to Go on mvdan.cc/sh with reject-to-LLM fallback.
License
Apache-2.0. See LICENSE.
