A Claude Code skill that pulls Claude out of "yes-man" mode and forces it to stress-test your idea before you build it — instead of approving everything you say.
When you're about to build something or green-light a plan, /red-team convenes a council of six historical figures. Five attack the idea from a fixed angle each; a judge then weighs every verdict and rules.
| Member | Role | Mandate |
|---|---|---|
| Diogenes | The Contrarian | Finds the fatal flaws — what actually kills this, ranked by lethality. |
| Alexander | The Expansionist | Finds the biggest honest upside — the 10x version you're too cautious to see. |
| Descartes | The First-Principles Thinker | Gets only the bare idea, no framing. Rebuilds the logic from zero. |
| Bacon | The Deep Researcher | Leaves the room and returns with real market data, competitor pricing, and prior art. |
| Hetty Green | The Buyer | Plays the actual customer and says straight up whether she'd pay. |
| Solomon | The Judge | Reads all five verdicts and renders one honest ruling — leading with the worst news. |
The five attackers run as parallel subagents so their verdicts don't contaminate each other. Solomon then synthesizes one verdict: the call (build / don't build / fix-X-first), where the council converged, the fatal flaws, a reality check from the data, the bigger swing, and the one thing to change first.
Uncritical agreement is worthless to you and expensive later. This skill exists to defeat the default tendency to find merit in whatever it's shown and approve it. A real flaw caught now is the most valuable thing the model can hand you; praise you didn't earn is the least.
Clone into your Claude Code skills directory:
git clone https://github.com/Tarikv1/red-team ~/.claude/skills/red-teamThen, in any Claude Code session:
/red-team
It also triggers on its own before Claude is about to approve a non-trivial idea or plan, or when you ask it to challenge, poke holes in, pressure-test, or "roast" your thinking.
Inspired by this Nate Herk YouTube video.