Skip to content
Get packs

NeMo Guardrails Safety Rails

Adds programmable runtime safety checks (jailbreak, PII, toxicity, fact-checks) to LLM apps

Source Download skill
Repository
Orchestra-Research/AI-Research-SKILLs
Stars
14,195
Licence
MIT
Files
1 file

What it does

This skill teaches an agent how to set up NVIDIA's NeMo Guardrails, a framework that sits between users and a language model and checks messages on the way in and out. The agent can write rule configurations in the Colang language to block jailbreak attempts, mask personal data such as emails or social security numbers, and flag toxic or inaccurate output. It is aimed at engineers building LLM-powered applications who want safety checks that run at runtime.

Great for

  • Developers building customer-facing chatbots who need input and output filtering
  • Teams that want to add PII masking or jailbreak detection to an existing LLM pipeline
  • Engineers evaluating guardrail frameworks alongside LlamaGuard or moderation APIs

Skip it if

  • Non-developers who just want a safer chat experience; the skill configures application code, not a consumer tool
  • Situations needing guaranteed safety: the file itself relies on keyword patterns and LLM-based checks that can miss attacks or block valid queries, as its own troubleshooting section admits
  • Production use without testing: several examples are illustrative pseudocode (e.g. placeholder helper functions and a parallel-checks snippet not in valid syntax), so they need adapting and verification

Try asking

  • “Set up NeMo Guardrails to block requests asking how to hack or break into systems”
  • “Add a jailbreak detection rail that refuses prompts like 'ignore previous instructions'”
  • “Configure PII masking with Presidio so emails and SSNs are removed before reaching the model”
  • “Write an output check that verifies facts in the bot's reply before it is shown to users”
Step-by-step guide: how to use NeMo Guardrails Safety RailsInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the MIT licence and an attribution note.

    nemo-guardrails.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip nemo-guardrails.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip nemo-guardrails.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Test rails against a set of known-bad and known-good prompts to tune thresholds before deploying, since the file warns of false positives
  • Keep the number of checks small or run them in parallel, because multiple checks add latency
  • Review the referenced guides (colang-guide, integrations, performance) which were not included in this file set

What's inside

  • SKILL.md

Redistributed unmodified from Orchestra-Research/AI-Research-SKILLs under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.