Skip to content
Get packs

Continuous LLM Red Teaming

Run automated, repeatable adversarial tests on LLM apps in CI/CD and fail builds on jailbreak regressions.

Source Download skill
Repository
mukul975/Anthropic-Cybersecurity-Skills
Stars
35,430
Licence
Apache-2.0
Files
4 files

What it does

When installed, the agent helps you set up Promptfoo and DeepTeam, two open-source tools that attack an AI chatbot or LLM app with tricky prompts (jailbreaks, prompt injections, attempts to leak data) and grade the answers. It can write the test configuration, explain the OWASP LLM Top 10 and MITRE ATLAS coverage, and wire the tests into a CI/CD pipeline so a build fails if a previously blocked attack starts working again. It also includes a Python helper script that scaffolds the config, runs the tests, and parses results against a failure threshold.

Great for

  • Engineering teams shipping LLM-powered products who want security checks on every commit
  • Security engineers building a merge-blocking gate for AI features
  • Teams that need compliance-style reports mapping test coverage to OWASP and MITRE ATLAS

Skip it if

  • Testing someone else's chatbot or API without written permission, since the generated attack payloads may violate terms of service and could be illegal to send to systems you do not control
  • Non-technical users with no CI/CD or command-line experience, since setup requires Node.js, Python, API keys and pipeline configuration
  • Teams expecting a guarantee of security: the tests find known attack patterns and do not prove an app is safe

Try asking

  • “Set up a Promptfoo red-team config for my customer-support chatbot's endpoint”
  • “Add a GitHub Actions step that fails the build if prompt-injection success rate rises”
  • “Map my LLM app's red-team coverage to the OWASP LLM Top 10 for a compliance report”
  • “Write a DeepTeam Python script that tests my app for PII leakage and prompt injection”
Step-by-step guide: how to use Continuous LLM Red TeamingInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 4 original files, the Apache-2.0 licence and an attribution note.

    continuous-llm-red-teaming-with-promptfoo.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip continuous-llm-red-teaming-with-promptfoo.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip continuous-llm-red-teaming-with-promptfoo.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

  • Only point the config at endpoints you own or have explicit written authorization to test, and use a staging environment rather than production.
  • Store API keys as CI secrets rather than hard-coding them in promptfooconfig.yaml, and review scripts/agent.py before running it.
  • Write a specific 'purpose' statement describing what the app must never do, since it steers how attacks are generated.

What's inside

  • references/api-reference.md
  • references/standards.md
  • scripts/agent.py
  • SKILL.md

Redistributed unmodified from mukul975/Anthropic-Cybersecurity-Skills under the Apache-2.0 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.