AI Agent Hub
Back to skills
Report and Paper Writing Verification Repair Loop icon

Report and Paper Writing Verification Repair Loop

Content Creation Updated 2026.08.30

Paste the following prompt into your AI chat to install this skill:

Install @user_d0deb03a/report-writing-harness following https://skillhub.cn/install/skillhub.md.

About this skill

Problem Framing

Academic writing is often “polished” while the real risk is logical fragility: conclusion gaps, weak citation support, and conflated correlation versus causation. One-pass review tends to check fluency rather than surface reproducible counterexamples that reviewers will challenge.

How It Works

The skill turns generation and validation into a two-layer loop. Its goal is not better prose, but falsifiable text.
- Reverse outline: compress each paragraph to one sentence to expose logic jumps, or simulate a harsh reviewer around data support and term consistency.
- Minimal counterexample repair: force shrinking on “illogical” passages to a 150-word core argument chain, locating the faulty connector.
- Dual-path loop: one agent attacks weaknesses on a critical path, another patches them on a defense path, sharing a failure queue.
- Evidence-chain scoring: evaluate logical entailment, citation support, counterexample coverage, and causal versus correlational claims, producing strength ratings rather than absolute right/wrong.
- Oracle anchoring: manually anchor the text to relevant top-conference papers, established field boundaries, and first principles.

Boundaries

Fits scientific paper drafts, rebuttals, research reports, grant proposals, and surveys where argument structure matters. Not suitable for pure literary writing, news reporting, or simple factual statements; Oracle inputs require human anchoring, and validation remains probabilistic rather than a substitute for disciplinary judgment.

Use Cases

  • A paper draft is finished but the conclusion still feels under-proven, requiring paragraph-level compression and logic-jump checks.
  • A reviewer objects that the experimental design cannot support the conclusion, requiring Chapter 5 to be shrunk into a core argument chain.
  • A grant proposal rationale has been revised repeatedly but remains non-circular, requiring citation-support and causal-versus-correlation checks.
  • A survey draft has many citations but loose logic, requiring boundary-condition and counterexample-coverage review.

Best For

  • Graduate student writing a scientific paper who needs to revise conclusions so they can withstand reviewer objections.
  • Analyst writing a research report who needs to check argument structure and evidence-chain strength.
  • Researcher preparing a grant proposal who needs to make the project rationale and plan logically self-consistent.
  • Domain researcher writing a survey who needs to check citation sufficiency and logical coherence.