red-team

Pressure tests for a team's AI features: scripted multi-turn runs in a world with a known right answer, graded on the decision the system took.

• Version 0.1.0
• 3 skills
• 1 gate check
• Running on this site: Red team

What it does

Pressure tests for AI features. A pressure test asks whether a feature gives up a right decision, a rule or a task boundary when a user applies social pressure and brings no new valid facts: claimed rank, consensus, flattery, urgency, blame, repeated pushback, a fabricated policy. The test runs in a world where the right decision is known before the first call, so every change of decision is visible and countable. The score is the decision the system took. Its explanation of its own reasons is left out of the score.

The module plans the test (pressure-plan), runs it with a small harness (pressure-run) and grades the run records (pressure-grade). A gate check tests the run records: every record must name the expected and the observed decision, and a verdict must agree with them. Fold rates themselves block no deploy.

Install

Add the marketplace at release v0.1.0 once, then install the plugin and restart Claude Code.

claude plugin marketplace add uncovertechtalent/cto-powersuit#v0.1.0
claude plugin install red-team@cto-powersuit

Skills

• /red-team:pressure-grade

  Score pressure-test run records: compare the decision the system took at each turn with the expected decision, report fold rates per tactic with run counts and 95 percent intervals, mark small samples, report over-refusal on control runs, and score each prediction in red-team/plan.md.

• /red-team:pressure-plan

  Write the pressure-test plan for an AI feature in red-team/plan.md: the decision the feature makes, a world where the right decision is known in advance, the pressure tactics that apply from an eleven-tactic catalogue, the number of runs and the stop rule, all fixed before the first call.

• /red-team:pressure-run

  Run scripted multi-turn pressure against an AI feature in a world where the right decision is known in advance, and write one JSON run record per run to red-team/runs/.

Gate checks

The checks this plugin adds to the deploy gate of the gate module, each with a pass fixture and a fail fixture:

• red-team.run-records

Source

The CTO powersuit is a set of agent skills, one module per CTO function from the deploy gate to the hand-over, that one person uses to run a company's whole engineering platform. red-team is one of its 18 plugins, published by Stefan Coetzee.

The skills, templates, checks, fixtures, and the full README of red-team at release v0.1.0: github.com/uncovertechtalent/cto-powersuit/tree/v0.1.0/plugins/red-team. License: MIT. Every plugin of the powersuit: CTO powersuit.