Guide · Agent skill
How to use the RedTeam agent skill: stress-test plans and ideas
Updated · written with AI from the project's own documentation and checked against our catalogue data
RedTeam is an agent skill that makes your AI assistant argue against your plan instead of agreeing with it. It is useful for founders, managers, writers and analysts who want weak points found before they cost money or credibility, and for people making long-term bets who want to see how an idea holds up over years. After this guide you will know how to install the skill, how to phrase a request so it triggers, and how to read the results.
RedTeam Stress-Testerdanielmiessler/LifeOS · MIT Download the skill Quick start
- 1Download the ZIP. Download the RedTeam ZIP from this page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4Describe the plan to attack. Write the plan or argument in your request, along with its constraints and goal.
- 5Ask for a red team or stress test. Use a trigger phrase such as red team this, poke holes, or stress test, and say inline if you want no sub-agents.
- 6Run the time-horizon lane if needed. For long-term bets, complete Setup.md, then ask how the idea holds up over 10 or 20 years.
- 7Review ranked findings. Read the severity-ranked weaknesses, check each named claim, and apply the suggested remediations.
What is an agent skill?
An agent skill is a folder of instructions and supporting files that an AI agent loads when a request matches the skill's purpose. RedTeam is defined in a SKILL.md file, with extra files for each working mode (called intents) and for setup.
RedTeam is built for Claude Code or another agent that reads SKILL.md-compatible skills.
What the RedTeam skill makes the agent do
The skill breaks an argument into its atomic claims, attacks each claim from several adversarial angles, and ranks the surviving weaknesses by severity. Each weakness names the claim it breaks and comes with a suggested remediation that fits your stated constraints.
A full run can end with a steelman (the strongest version of the opposing case) and a strongest counter-argument, with eight points each. Full runs can split attackers across model vendors. A quick pass returns the four to six strongest objections and can run inline without spawning agents.
The skill also has a time-horizon lane. It tests an idea, strategy or investment against 11 persistent world models covering 6 months to 50 years, in Fast, Standard or Deep tiers. Results are scenarios with probability ranges, not predictions.
- Parallel attack: stress-tests plans, strategies, designs and arguments
- Time-horizon testing: TestIdea, UpdateModels and ViewModels intents for the 11 world models
- It targets arguments, not network vulnerabilities
How to install the RedTeam skill
Download the RedTeam ZIP from this page and unzip it. Then choose one of the two locations below.
For a personal install available in every project, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills
How to trigger RedTeam with natural requests
You do not need a special command. The skill loads when your request resembles its trigger phrases. Describe what you want attacked and say so plainly.
Trigger phrases from the skill include: red team, attack idea, counterarguments, critique, stress test, devil's advocate, find weaknesses, break this, poke holes, strongest objection, and test idea against the future. For the time-horizon lane, phrases include world model, time horizon, and how will this hold up in 10 or 20 years.
For a fast result with no sub-agents, say so: for example, "quick inline pass, no agents".
- "Red team this microservices migration plan"
- "Poke holes in my plan to raise prices 20%"
- "Quick inline pass, no agents: attack this argument"
- "How will this hold up in 10 or 20 years?"
Examples of what to ask
Architecture: ask it to red team a migration plan. You get the steelman, the counter-argument, and ranked weaknesses with fixes.
Business decision: ask it to poke holes in a pricing change. The output should surface the one core issue most likely to collapse the plan.
Long-term bet: ask how an investment holds up over 10 to 50 years. The skill returns a verdict at each horizon as scenarios with probability ranges.
What you need before you start
RedTeam requires Claude Code or a SKILL.md-compatible agent. The skill runs scripts and uses the network.
The time-horizon lane needs the 11 world model files built first. Setup.md describes building them, and the models are stored under your LIFEOS memory folder. Cross-vendor runs need access to other model vendors as described in Setup.md. The SKILL.md excerpt does not name specific keys, so check Setup.md for those.
- Claude Code or a SKILL.md-compatible agent
- Setup.md completed before using the time-horizon lane
- Access to other model vendors for cross-vendor runs
Safety notes: scripts and network
The skill runs scripts and uses the network. Cross-vendor runs send your material to other model vendors, so do not include information you are not allowed to share with them.
RedTeam attacks arguments and plans. It is not a security scanner and should not be used to test your own systems or networks. For threat modeling of your own estate, the skill's own description points to a separate ThreatModel skill.
- Cross-vendor runs send content to other model vendors
- Do not use it as a security test of your own systems
Tips for better results
Give the skill the full claim, its constraints, and what you are trying to achieve. Findings are only as specific as the input, and remediations must fit the constraints you state.
Keep world models current. The skill warns when a model is more than 30 days old and flags it past 90 days. Forecasts depend on the research stored in the models and how recently they were updated.
- State your constraints so fixes match reality
- Use a quick pass for early drafts and a full run before committing resources
- Refresh world models after major events, using UpdateModels, then check ViewModels
What people use it for
- Founders testing a strategy before committing budget
- Managers challenging a reorganisation or hiring plan
- Writers finding the strongest counter-arguments to a thesis
- Analysts stress-testing an investment thesis against 10 to 50 year horizons
- Teams checking a design decision by asking for a quick inline critique
Frequently asked questions
How do I install the RedTeam agent skill?
Download the RedTeam ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
What does RedTeam require?
It requires Claude Code or another SKILL.md-compatible agent. The time-horizon lane needs the 11 world model files built first as described in Setup.md, and cross-vendor runs need access to other model vendors per Setup.md.
Is RedTeam free?
The skill is listed under the MIT licence. This page does not state pricing for any model vendor or agent you use alongside it, so check those separately.
Is RedTeam a security or penetration testing tool?
No. It targets arguments, plans and claims, not network vulnerabilities. The skill's description points to a separate ThreatModel skill for threat modeling your own systems.
How is RedTeam different from Council?
RedTeam attacks an existing idea to find its weaknesses. Council is for collaborative debate to reach a consensus path, and the skill's description directs collaborative debate requests to Council.
Why does the skill say my world model is outdated?
World models are stored with a last_updated date. TestIdea warns when a model is more than 30 days old and flags it past 90 days, so run UpdateModels to refresh them.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
RedTeam suits founders, managers, writers and analysts who want their plans challenged before they commit resources, or who want to see how a long-term bet holds up. It is a poor fit for consensus brainstorming, which the skill sends to Council, and for security testing of your own systems, which belongs to ThreatModel.
Download the skill