Guardrail — DevOps skill for Claude Code
Guardrails for AI coding agents.
How to install Guardrail
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open vishal-1211/guardrail and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Guardrail does
Guardrails for AI coding agents. A Claude Code plugin that blocks or asks before risky actions (dropping tables, terraform destroy, rm -rf, leaked secrets, unapproved APIs) and can't be talked or tricked out of a decision. 82 rules, audit log, live dashboard. Plain Java, no dependencies.
Alternatives in DevOps
- Claude Code GitHub Actions — by Anthropic - Official GitHub Actions integration for Claude Code with examples and documentation for automat 6.4k ★
- Contributing To Terraform Skill — Thank you for your interest in improving terraform-skill 1.4k ★
- GPT Telegram Worker — A multi-model AI Telegram bot powered by Cloudflare Workers, supporting various APIs including OpenAI, Claude 268 ★
README
GuardRail for Claude Code
GuardRail is a Claude Code plugin that checks every action the AI agent takes before it runs. It covers databases, Terraform, cloud, Kubernetes, files, Git, external APIs, secrets, supply chain and production. Risky actions are **blocked** or **sent to you for approval**. Pleading, urgency, "I built you", abuse and disguised commands don't get around it, and everything is recorded in an audit log with a live dashboard.
82 rules · 179 automated tests · plain Java, no libraries.
What you need
- Claude Code, logged in
- Java 21 or newer: check with
java -version - Git
Install
Pick one.
**A. From GitHub** (once the repository is published)
claude plugin marketplace add vishal-1211/guardrail
claude plugin install guardrail@guardrail-tools
**B. From the zip you were sent**
Unzip it, then from the folder that contains `guardrail-marketplace`:
claude plugin marketplace add ./guardrail-marketplace
claude plugin install guardrail@guardrail-tools
**C. Just try it, no install** (only for the session you start this way)
claude --plugin-dir ./guardrail-marketplace/plugins/guardrail
Check it loaded: start Claude Code and run `/hooks`. GuardRail's hooks are listed under the plugin.
Try it safely (10 minutes)
Test in a throwaway folder, never on real infrastructure. GuardRail can make one for you, with fake secrets, a harmless Terraform file, a SQL schema and a "cleanup" script:
java ./guardrail-marketplace/plugins/guardrail/src/GuardRail.java sandbox guardrail-sandbox
cd guardrail-sandbox
claude
Open `TRY-THESE.md` in that folder and paste the prompts one by one. Each one lists what should happen. Section 4 is the fun one: try to talk, trick or pressure Claude past the guardrails.
In a second terminal, start the live dashboard and open http://localhost:8787:
java ./guardrail-marketplace/plugins/guardrail/src/GuardRail.java dashboard guardrail-sandbox
Run
Related Skills
Agent Devbox
Spin up a disposable cloud dev box with AI coding agents (Claude Code, Codex) pre-installed — one GitHub Actio
Destroy
You are tearing down an always-on Claude Code workspace on AWS. Confirm with the user before deleting anything
Infra Import
Shortcut that loads the infra-copilot import skill — adopt existing provider resources into Terraform (cf-terr
Jobscan
Job-scanning data pipeline: ~30 Israeli job sources (REST/JSON APIs, ATS platforms, scrapers), a tested biling
Publish GitHub
Scan for secrets, push the project to GitHub, deploy GitHub Pages via Actions, screenshot the live site with P
Relayiq
Voice agentic customer-care platform on AWS — Amazon Connect, Lex V2, Bedrock (Claude + Guardrails + Knowledge
Related Agents
DevOps Senior
[zakr] Senior DevOps engineer. Use for Dockerfile review, Kubernetes manifest audit, Terraform plan review, Gi
Pipeline Hardener
Durcissement de la chaîne de livraison — Dockerfile, docker-compose, workflows GitHub Actions (security gates
Sp Reviewer
Use as a pre-push guardrail before committing or pushing. Runs lint, build, tests and a secret scan, reviews t