Skip to content
RedTeam

Chat-only review

Observed need and evidence

Section titled “Observed need and evidence”

AI assistants default to agreement. Users need structured adversarial review without installing anything or maintaining project files. This is the primary onboarding path and the fastest proof of value.

Desired user and business outcome

Section titled “Desired user and business outcome”

A user pastes a plan or decision into chat, runs /redteam challenge (or another command), and receives a structured review they can act on before committing.

Users and context

Section titled “Users and context”

Decision-makers, strategists, and AI power users who want immediate challenge output with no .redteam/ setup.

Current journey

Section titled “Current journey”
  1. User has a plan, memo, or decision draft
  2. User invokes /redteam <command> with the target in the message
  3. Model loads skill/SKILL.md and the command reference flow
  4. Model returns structured output in chat

Opportunity and hypothesis

Section titled “Opportunity and hypothesis”

If challenge output is structured (assumptions, risks, verdict, follow-ups) without any install step, users will adopt RedTeam before anchoring a project directory.

Intended behavior

Section titled “Intended behavior”

The harness loads the skill and reference flow; the model produces adversarial review output aligned with the command template.

Given / When / Then scenarios

Section titled “Given / When / Then scenarios”

  • Given the RedTeam skill is available in the harness and the user has not run /redteam init

  • When the user runs /redteam challenge <plan>

  • Then the model produces structured review output (assumptions surfaced, risks ranked, verdict, follow-ups) without requiring CONTEXT.md

  • Given CONTEXT.md is missing

  • When the user runs any chat-only command

  • Then the review proceeds using only the pasted target and skill reference — it must not error or block

Constraints and domain language

Section titled “Constraints and domain language”
  • Red teaming here means applied critical thinking, not cybersecurity penetration testing
  • Commands route through /redteam <command>; shortcuts may be pinned with /redteam pin

Success signals and telemetry

Section titled “Success signals and telemetry”
  • User completes a review in one session without setup
  • Output includes actionable follow-ups (qualitative review of command transcripts and .redteam/reviews/ when saved manually)

Open questions

Section titled “Open questions”
  • Which commands are most used in pure chat vs project-anchored mode? — maintainers
Section titled “Related requirements, tests, architecture, and ADRs”