Research integrity guardrail

Skill zhoy0409-debug/scholarflow-skills/skills/research-integrity-guardrail

ScholarFlow Skills by Zhoy: reusable Agent Skills for literature review, academic writing, figures, bioinformatics, and research deliverables.

Install

npx -y skills add zhoy0409-debug/scholarflow-skills --skill research-integrity-guardrail

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when a researcher needs manuscripts, reports, or study materials audited for unsupported claims, inconsistent numbers, citation gaps, reproducibility omissions, or overstatement.

SKILL.md

3.0 KB, as published. Nobody here has run it

Research Integrity Guardrail

Protect the evidence chain before polishing language. This skill audits a research artifact; it does not invent evidence, conceal uncertainty, or make unsupported work sound more convincing.

Establish the audit boundary

Identify the artifact, its audience, the claimed contribution, and the materials available for verification: manuscript, figures, tables, reference list, raw calculations, methods, code, protocol, or supplementary files. Mark any missing material as unverified rather than inferring it.

Build an evidence ledger

For each consequential claim or result, record:

Claim or result	Evidence location	Verification status	Required action
	Table, figure, source, calculation, method, or code	Supported / unclear / inconsistent / unsupported

Do not treat a citation as verification unless it actually supports the claim being made.

Audit in this order

Numbers and cross-references: reconcile repeated counts, percentages, ratios, P values, figure labels, table values, and appendix references.
Claims and citations: locate each major claim's support; flag orphaned citations, unused references, missing close comparators, and statements stronger than their evidence.
Methods and reproducibility: check that datasets, preprocessing, exclusions, controls, parameters, statistics, code, and evaluation splits are sufficiently described for the stated result.
Interpretation: distinguish observed findings from mechanism, causality, clinical meaning, novelty, and generalization. Require limitations where they materially affect interpretation.
Presentation integrity: identify selective reporting, missing negative cases, unexplained visual choices, and claims that do not trace to a figure, table, calculation, or source.

Return findings before revisions

Lead with material risks, ordered by severity. Separate facts from judgment:

Blocked: demonstrably inconsistent, unsupported, or missing evidence.
Needs verification: cannot be checked from the supplied materials.
Conservative rewrite needed: evidence exists but the wording overstates it.
Editorial: clarity or structure issue with no integrity implication.

Use this format:

## Integrity Findings
1. [Blocked] Location - issue - evidence - required fix

## Evidence Ledger
| Claim or result | Evidence location | Verification status | Required action |
| --- | --- | --- | --- |

## Release Decision
- Ready items:
- Blocking items:
- Items requiring author confirmation:

Only after the user resolves the material risks should this skill help revise wording. For journal formatting, use journal-submission-normalizer; for manuscript-journal fit, use journal-selection-advisor; for figure QA, use polish-sci-figures.