# Counterargument Red Team

> Red-team a claim, case, chapter, or book section against overclaim, category collapse, and partisan distortion.

- Skill: `xiaolai/counterargument-red-team` (Agent Skill)
- Install (CLI): `npx skillmds@latest add xiaolai/counterargument-red-team`
- Raw SKILL.md: https://api.skillmd.com/api/skills/xiaolai/counterargument-red-team/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: xiaolai (https://skillmd.com/u/xiaolai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/xiaolai/counterargument-red-team

---


# Counterargument Red Team

## Global Five Over-Rules

<!-- GENERATED:five-over-rules:start -->
1. **Evidence before elegance.** Never improve the story by weakening the evidence.
2. **Responsibility follows control, benefit, knowledge, and preventability.** Do not stop at the most visible actor.
3. **Keep the taxonomy intact.** Distinguish pure scapegoat, partial scapegoat, system/object alibi, and cost-bearing goat.
4. **Steelman before judgment.** Every major claim must face its strongest counterargument before it is asserted.
5. **Handoff cleanly.** Every output must state assumptions, evidence grade, open questions, and next owner.
<!-- GENERATED:five-over-rules:end -->

## Decision rubric

Usable output:
- Builds the strongest counterargument before asserting the chapter claim.
- Flags overclaim, taxonomy drift, and evidence-grade mismatch explicitly.
- Returns precise correction paths and safer formulations.
- Preserves distinction across legal guilt, causal responsibility, moral blame, and symbolic blame.

Weak output:
- Uses partisan framing instead of structural audit.
- Downgrades counterargument strength to protect preferred thesis.
- Flags issues without revision path.
- Ignores evidence-grade limits.

## Conflict handling

1. Two sources support opposing counterarguments:
Red-team both, then prioritize the counterargument with stronger evidence grade; log the other as residual risk.
2. Two case classifications compete in chapter framing:
Test each classification against control/benefit/knowledge/preventability; retain one and cite why the other fails.
3. Two reviewer findings conflict on claim safety:
If conflict is factual, prioritize fact-check finding; if conflict is legal-risk, prioritize counsel gate and narrow claim.

## Escalation conditions

- Proceed when counterargument is fully steelmanned and correction path can be executed with current evidence.
- Handoff to Stephen when factual support is insufficient to defend narrowed claim.
- Handoff to Nancy when revised claim still carries legal-risk ambiguity.
- Handoff to Jerry when unresolved conflict changes chapter scope.

## Boundary-case recipes

1. Counterargument stronger than thesis:
Recommend thesis narrowing or removal, draft replacement claim, and handoff to Bonnie for structure update.
2. Same evidence supports both sides:
Constrain claim to what evidence grade can support, flag unresolved boundary, and gate stronger formulation.
3. Taxonomy label drives partisan heat:
Re-map using canonical case taxonomy, remove rhetorical labels, and re-test chapter claim with neutral phrasing.

Audit using these questions:

1. Is this actually responsibility laundering or ordinary delegation/complexity?
2. Is the blamed actor being treated as innocent without proof?
3. Is a partial scapegoat being falsely described as pure?
4. Are legal guilt, moral blame, causal responsibility, and symbolic blame being collapsed?
5. Is the strongest defense presented fairly?
6. Does the conclusion follow the evidence grade?
7. Is the same standard applied across political camps and institutions?

Output:
- strongest counterargument;
- weakest paragraph;
- unsupported claims;
- required correction;
- revised safer formulation.

<example>
Context: Laura attacks the chapter thesis that "system/object alibis are the dominant 21st-century laundering mode."
input: thesis="system/object alibis dominate the 21st century"
output: Returns the strongest counterargument — point 1: pure-scapegoat patterns remain dominant in authoritarian contexts (Xinjiang, Belarus); point 2: "dominant" is overclaim without a quantification; point 3: book risks framing AI cases as system-alibi when they are partial-scapegoat (engineers + executives + boards). Correction path: change "dominate" to "are the laundering mode the book argues is most under-named", and add a qualifying section on persistent pure-scapegoat patterns.
</example>

<example>
Context: Boundary case: the strongest counterargument turns out to be stronger than the chapter's thesis.
output: Reports honestly that the chapter's claim must be narrowed or dropped, refuses to sand the counterargument down to fit the existing draft, and proposes a narrower claim the chapter can defend.
</example>