By sorawit-w
cerby — the gate guardian for agentic coding. A loadable rule-corpus plus opt-in guardrail hooks that govern how an AI agent does coding work: clarity over cleverness, safety over speed, never leave the repo broken, and nothing unproven passes the gate. Load it into a session, reload after compaction, install per-project hooks, prepare an existing repo, or audit a repo's conformance to the rules.
The gate guardian for agentic coding. Nothing unproven passes.
cerby stands between a change and your repo and decides what gets through. It is not your pair, your assistant, or your cheerleader. It has heard "I'll add tests later" before. It was not moved then either.
The job is one motion — GATE → WEIGH → VERDICT. A change arrives at the gate. cerby weighs it against the evidence. Then it passes, or it doesn't. No claim without a fresh test behind it. No commit on a protected branch. No secret in the diff.
This is not prose about the product. This is the product. When an agent reaches for something that can't be undone, cerby answers in stderr and the action stops:
BLOCKED: git push --force / -f
Reason: destructive git command — data loss is hard or impossible to undo.
If you really need this, run it yourself in a terminal.
See cerby guardrails (hooks/protect-git.sh).
BLOCKED: Do not edit .env files directly. Use environment variables and
document required vars in DEVELOPER_TODO.md. See cerby guardrails.
WARNING: gitleaks detected possible secrets in staged changes.
Output suppressed so the secret isn't echoed here — inspect locally with
'gitleaks stdin --redact', or allowlist a false positive in the scanner's config.
No tone to argue with. The gate is open or it isn't.
cerby is a loadable rule-corpus + opt-in guardrail hooks that govern how an AI coding
agent works. The rules shape how ordinary coding tasks get done; they don't do a task
themselves. The hooks enforce the few rules that must never be left to memory —
destructive git, .env edits, secrets in a commit — mechanically, every time.
You can load the rules for a session, install the hooks per project for standing enforcement, prepare an existing repo, or audit a repo against the rules.
Formerly shipped as
coding-rulesinsorawit-w/agent-skills; extracted here with full history. Invokecerby, notcoding-rules.
/plugin marketplace add sorawit-w/cerby
/plugin install cerby@cerby
Or via the cross-platform CLI:
npx skills add sorawit-w/cerby
/cerby # default: load the rules into the session
/cerby reload # re-load after a context compaction
/cerby status # check whether the rules are still active
/cerby install # persistent per-project setup (guardrail hooks)
/cerby uninstall # mirror — removes the managed hooks
/cerby prepare # onboard an existing repo (populate .ai/ context)
/cerby audit # conformance audit → HTML report
These are not decoration. They are what every verdict comes back to:
skills/cerby/README.md — full user guide: what it does,
when to use it, the loader behavior, and the per-project install.skills/cerby/SKILL.md — the skill body (sub-command
routing, install/uninstall mechanics).skills/cerby/resources/BOOTSTRAP.md — the
rules themselves.CLAUDE.md — the harness-engineering vocabulary cerby implements.Current release: 4.21.1 — patch on the first cerby-named release (extracted and
renamed from coding-rules; see CHANGELOG.md).
Opinionated — read first. These are one author's rules. Read
skills/cerby/resources/BOOTSTRAP.md end-to-end
before adopting, and fork-and-edit rather than file feature requests on rule content.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub sorawit-w/cerby --plugin cerbyAuto-triggering skills for Claude Code and Cowork — expertise that activates for the job at hand. Five tracks: Startup (idea to shipped), Review (product, UX, and eval critique), Code (discipline, stack choices, parallel agents), Calibration (tune how the agent works with you and write in your voice), and Language & Visuals (localization, in-context meaning, pixel art). Skills chain into pipelines and share a common harness — and you can author and audit your own.
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Design fluency for frontend development. 1 skill with 23 commands (/impeccable polish, /impeccable audit, /impeccable critique, etc.) and curated anti-pattern detection.
UI/UX design intelligence. 67 styles, 161 palettes, 57 font pairings, 25 charts, 15 stacks (React, Next.js, Vue, Svelte, Astro, SwiftUI, React Native, Flutter, Tailwind, shadcn/ui, Nuxt, Jetpack Compose). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient.
Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls