kill-argument
Two-thread adversarial review: a fresh reviewer constructs the strongest 200-word rejection memo, then a second fresh reviewer defends the paper point-by-point and surfaces still-unresolved critical issues. Use when user says \"kill argument\", \"adversarial review\", \"hostile review\", \"rebuttal preparation\", \"reviewer-2 simulation\", or before submitting a theory paper that has already passed standard review rounds.
npx skills add majiayu000/claude-skill-registry --skill kill-argument --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
The skill creates a two-thread adversarial review process: the first thread generates the strongest 200-word rejection memo for a given paper, and the second thread defends it point-by-point, classifying unresolved issues. It is designed to be used when users say phrases like "kill argument", "adversarial review", or similar, or before submitting a theory paper that has already passed standard review rounds. It is explicitly described as a verdict-bearing exercise producing an attack memo and a defense with unresolved issues surfaced for later action.
How it works
- Step 1: Discover paper files: locate the paper directory, find the LaTeX entry, and gather source files (tex, bib, figures).
- Step 2: Attack memo (Thread 1, fresh codex): invoke mcp__codex__codex with model gpt-5.5 and reasoning effort xhigh in a read-only sandbox, cwd set to the paper directory, and a prompt that instructs simulating a hostile reviewer to construct a single strongest rejection paragraph of approximately 200 words. The prompt emphasizes selecting one damaging line of attack from axes like theorem validity, assumption-vs-claim mismatch, missing proof obligations, limit-order ambiguity, claim-vs-evidence gap, or scope overclaim, and to cite specific file:line locations when accusing. Output is the rejection memo only, about 200 words, with a single argument.
- Step 2.5 (optional beast effort): if enabled, perform six axis probes as separate fresh-codex prompts to gather evidence breadth, then synthesize a single 200-word verdict by committing on the strongest axis.
- Step 3: Adjudication memo (Thread 2): invoke a second mcp__codex__codex call to read the paper files and the verbatim attack memo, decompose into 3-7 rejection points, and for each point assign one of: answered_by_current_text, partially_answered, still_unresolved, with structured fields including Attack claim, Verdict, Evidence, Severity, and If unresolved, recommended fix. Then provide a Net assessment, Top action items, and a concise summary.
- Step 4: Write KILL_ARGUMENT.md and KILL_ARGUMENT.json containing the human-readable report and a machine-readable artifact per the specified format.
When to use it
- After 1-2 rounds of /auto-paper-improvement-loop settled at a stable score, but before submission
- During rebuttal preparation to predict the strongest objection
- For theory papers with a high-level title that may oversimplify the actual theorem
- For papers where a reviewer might attack scope, assumption-vs-claim mismatch, missing proof obligations, or evidence-vs-headline gaps
What it can touch
The workflow relies on Bash-like orchestration and the tools listed: Bash, Read, Write, Edit, Grep, Glob, mcp__codex__codex. It reads paper files and writes KILL_ARGUMENT.md and KILL_ARGUMENT.json to the paper directory.
Caveats
- Declares constants such as REVIEWER_MODEL =
gpt-5.5, CONTEXT_POLICY =fresh, ATTACK_LENGTH ≈ 200 words, DEFENSE_DECOMPOSITION = 3-7, OUTPUT =KILL_ARGUMENT.md+KILL_ARGUMENT.json, and RENDER_HTML = true. - Requires the paper directory structure and specified source files to exist; all steps rely on current paper files only and do not consult prior reviews.
- The process produces a single strongest rejection paragraph and a defense, not a list of multiple independent points.
# Kill Argument Exercise: Adversarial Attack-Defense Review > 🔒 **Do not wrap this skill in `/loop`, `/schedule`, or `CronCreate`.** It is > verdict-bearing — it produces an adversarial accept/reject verdict (attack → > adjudication). Re-firing it on a wall-clock timer adds no new signal (the > attack changes only when the *paper* changes). Schedule the *external wait > that precedes it* — draft stable → then run this **once** before submission. > See > [`shared-references/external-cadence.md`](../shared-references/external-cadence.md). Stress-test the headline claims of a paper against the strongest possible rejection argument: **$ARGUMENTS** ## Why This Exists Standard score-based reviews (`/research-review`, `/auto-paper-improvement-loop`) tend to produce **balanced** weakness lists. Each weakness gets ~equal attention, ranked CRITICAL > MAJOR > MINOR. Empirically, this misses one specific failure mode: the **single most damaging argument** a reviewer would write in a rejection paragraph — the one sentence that, if a senior area chair reads it, kills the paper. A balanced reviewer might list "scope-overclaim risk" as MAJOR alongside 3-5 other MAJORs, never quite committing. An
- Why This Exists
- How This Differs From Other Review Skills
- When To Use
- Constants
- Workflow
- Step 1: Discover paper files
- Step 2: Attack memo (Thread 1, fresh codex)
- Step 2.5 (optional, beast effort): multi-axis attack fan-out
- Step 3: Adjudication memo (Thread 2, fresh codex with attack + paper)
- Step 4: Write KILLARGUMENT.md and KILLARGUMENT.json
- Step 5: Print summary
- Output Contract
- Key Rules
- When NOT to Use
cd "$PAPER_DIR" Find the LaTeX entry point echo "Entry: $ENTRY" Find all source files codex should read find . -name "*.tex" -not -path "./.git/*" 2>/dev/null find . -name "*.bib" -not -path "./.git/*" 2>/dev/null find figures/ -name "*.pdf" -o -name "*.png" 2>/dev/null ls -la *.pdf 2>/dev/null # compiled PDF
What does the kill-argument skill do?
Two-thread adversarial review: a fresh reviewer constructs the strongest 200-word rejection memo, then a second fresh reviewer defends the paper point-by-point and surfaces still-unresolved critical issues. Use when user says \"kill argument\", \"adversarial review\", \"hostile review\", \"rebuttal preparation\", \"reviewer-2 simulation\", or before submitting a theory paper that has already passed standard review rounds.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill kill-argument --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.