argue-against-me

name	argue-against-me
description	Structured academic debate — Claude adopts an opposing position (from named historiographical schools or disciplinary counter-positions) and challenges your thesis through formal rounds with adjustable intensity. Produces a scorecard, revised thesis, and bibliography gap analysis. Use when stress-testing arguments, preparing for peer review, or strengthening a thesis before publication.
version	1.0.0
license	MIT
author	Kwok-leong Tang
contributors	[{"name":"Claude","type":"AI Assistant"}]

Argue Against Me

A structured devil's advocate for academic arguments. State your thesis, and Claude will systematically challenge it across formal debate rounds — drawing from historiographical traditions or disciplinary counter-positions — then deliver a scorecard, a stronger revised thesis, and suggestions for filling evidence gaps.

When to Use

• You're developing an argument and want to find its weak points before a reviewer does
• You want to test whether your evidence actually supports your claim
• You need to anticipate objections for a paper, dissertation chapter, or conference presentation
• You want to explore how your thesis looks from a specific intellectual tradition
• You're stuck and want an adversarial conversation to sharpen your thinking

Arguments

All arguments are optional.

Argument	Values	Default
intensity	gentle, firm, ruthless	firm
rounds	any integer	3
school	free text (e.g., "revisionist", "Marxist historiography", "postcolonial")	auto-selected: strongest opposing tradition for humanities; strongest disciplinary counter-position otherwise

Intensity Levels

gentle — Points out gaps and asks probing questions. Challenges weak evidence and logical gaps. Good for early-stage ideas you're still forming.

firm — Direct challenges with counterevidence. Challenges all claims, methodology, and source selection. The default — appropriate for arguments you believe are solid.

ruthless — Assumes nothing is established. Attacks everything including framing, definitions, periodization, and unstated assumptions. Use when preparing for hostile peer review or defending a controversial thesis.

How It Works

Starting a Debate

After invoking the skill, state your thesis. Optionally provide:

• Supporting evidence or sources
• The context (paper, dissertation, conference talk)
• Any specific concerns you want tested

Round Structure

Each round follows this cycle:

Round N of M
─────────────
1. YOU:    State or defend your position
2. CLAUDE: Attack (challenge the argument)
3. YOU:    Rebut
4. CLAUDE: Per-round assessment (brief verdict before next round)

Round 1 is special. Claude's opening attack will:

• Declare the opposing school or lens it's adopting, and why it's the strongest counter-position
• Identify the 2-3 strongest lines of attack against your thesis

Subsequent rounds build on the exchange. Claude escalates, introduces new counterevidence, or shifts angle when you successfully rebut a point. Claude does not repeat defeated arguments.

Mid-Debate Commands

At any point during the debate, you can say:

• extend — add more rounds beyond the original count
• resolve — end the debate early and skip to the closing output
• concede [point] — acknowledge a specific point to narrow the remaining debate

Closing Output

After the final round (or when you say resolve), Claude produces three things:

1. Scorecard

A table assessing each claim you made:

Claim	Held?	Strength (1-5)	Notes
...	✓/✗/△	...	...

• ✓ = held up under challenge
• ✗ = dismantled
• △ = partially held (survived with qualifications)

Strength is scored using the rubric in references/scoring_rubric.md.

Followed by an overall verdict paragraph.

2. Revised Thesis

A rewritten version of your original thesis that:

• Drops claims that were dismantled
• Adds qualifications where you made partial concessions
• Preserves what survived intact
• Notes what changed and why

3. Bibliography Gaps

Specific suggestions for strengthening your argument:

• Sources you'd need to address the counterarguments raised
• Primary sources that could fill evidentiary gaps
• Methodological frameworks that could shore up weak points
• Where relevant, suggests using other skills to find sources:
  • Library catalogs (Columbia CLIO, Harvard, HathiTrust, LOC, NLB Singapore)
  • Biographical databases (CBDB, JBDB)
  • Wikidata for linked data and identifiers
  • arXiv for relevant preprints
  • Zotero for managing found references

Best Practices

• State your thesis clearly up front. A vague thesis produces vague attacks. The more specific your claim, the more useful the challenge.
• Bring your evidence. If you have sources, cite them. Claude will attack the evidence, not just the logic — but only if you provide it.
• Don't concede too easily. The skill is most useful when you genuinely defend your position. Push back.
• Use ruthless before submission. If your argument survives ruthless intensity, it's ready for peer review.
• Try different schools. Run the same thesis against multiple traditions to find blind spots you didn't expect.

Example

User: /argue-against-me intensity=firm rounds=3

My thesis: The Song dynasty's commercial revolution (960-1279) was primarily
driven by state policy — particularly the monetization reforms and the
relaxation of market regulations — rather than by endogenous economic forces
or technological change.

I'm supporting this with evidence from the expansion of government-issued
currency, the abolition of the ward-market system, and state investment in
canal infrastructure.

Claude would:

1. Select an opposing lens (likely quantitative/cliometric or world-systems theory)
2. Attack the state-primacy framing — e.g., arguing that demographic growth, iron production advances, and Southeastern Asian trade networks preceded and enabled state reforms
3. Challenge the evidence — e.g., questioning whether currency expansion was state-led or a response to existing commercial demand
4. Run 3 rounds of structured debate
5. Produce scorecard, revised thesis, and bibliography gaps

**Step 2: Do NOT commit.** Just create the file. The controller will handle commits.