adversarial-brainstorm

Adversarial Brainstorming

Overview

Pressure-test ideas and designs through adversarial analysis before implementation.

Find what can fail, how it fails, and what safeguards are needed. Prioritize discovering hidden assumptions, edge cases, abuse paths, operational risks, and rollback gaps.

Anti-Pattern: "Happy Path Means Safe"

Treat all designs as vulnerable until challenged. "Simple" changes still need failure-mode analysis because small assumptions cause large regressions.

Checklist

You MUST create a task for each of these items and complete them in order:

1. Explore project context deeply — check files, docs, recent commits
2. Clarify assumptions — ask one question at a time about constraints, trust boundaries, and success criteria
3. Generate challenge models — propose 2-3 adversarial lenses with trade-offs and recommendation
4. Stress-test the design — identify edge cases, failure modes, abuse paths, and operational hazards
5. Present adversarial review — show findings and mitigations in sections scaled to complexity, get user approval after each section
6. Write adversarial design doc — save to <agent-settings-directory(.claude|.agents|.codex)>/plans/YYYY-MM-DD-<topic>-adversarial-design.md and commit
7. Transition to implementation planning — invoke plan skill

Process Flow

digraph adversarial_brainstorming {
    "Explore project context" [shape=box];
    "Clarify assumptions" [shape=box];
    "Propose challenge models" [shape=box];
    "Stress-test design" [shape=box];
    "Present findings + mitigations" [shape=box];
    "User approves mitigations?" [shape=diamond];
    "Write adversarial design doc" [shape=box];
    "Invoke plan skill" [shape=doublecircle];

    "Explore project context" -> "Clarify assumptions";
    "Clarify assumptions" -> "Propose challenge models";
    "Propose challenge models" -> "Stress-test design";
    "Stress-test design" -> "Present findings + mitigations";
    "Present findings + mitigations" -> "User approves mitigations?";
    "User approves mitigations?" -> "Present findings + mitigations" [label="no, revise"];
    "User approves mitigations?" -> "Write adversarial design doc" [label="yes"];
    "Write adversarial design doc" -> "Invoke plan skill";
}