AI Paper-Mill & Research-Integrity Screener
First-pass screener that flags AI-generated, paper-mill, and fabricated scholarship using tortured-phrase, citation, statistical, and template detectors.
- Editorial triage of new journal submissions
- Pre-peer-review integrity screening
- Research-integrity office case intake
$12
· or 60 creditsSecure checkout via Stripe
Included in download
- Editorial triage of new journal submissions
- Pre-peer-review integrity screening
- file_read, browser, network automation included
Sample input
Screen this abstract for research-integrity red flags before I send it out for peer review: "In this work we apply counterfeit consciousness and profound learning to bosom malignancy classification. Across 10 patients, accuracy was 33.3% (p = 0.000). Our irregular woodland model bolsters the discoveries." Report any concerns with evidence and a concern level.
Sample output
Overall concern level: HIGH (triage signal, not a determination). A. Tortured phrases (STRONG): "counterfeit consciousness" (artificial intelligence), "profound learning" (deep learning), "bosom malignancy" (breast cancer), "irregular woodland" (random forest). Four synonym-swapped technical terms; no innocent stylistic origin. C. Statistical impossibility (STRONG): "33.3% of 10 patients" is not achievable — 3/10 = 30%, 4/10 = 40%; no integer count yields 33.3% at N=10. "p = 0.000" reported without a bound. Innocent explanation: transcription/rounding error, but combination is implausible. Convergence: two STRONG, independent signal classes (A + C) → HIGH concern. Recommended next steps for a human reviewer: request the raw data and exact counts; confirm sample size; verify any cited references online; route to the editor/research-integrity office per venue policy. Disclaimer: This is an automated first-pass triage. It identifies signals, not misconduct, and may produce false positives. No statement asserts that any person engaged in fraud.
AI Paper-Mill & Research-Integrity Screener
First-pass screener that flags AI-generated, paper-mill, and fabricated scholarship using tortured-phrase, citation, statistical, and template detectors.
$12
· or 60 creditsSecure checkout via Stripe
Included in download
- Editorial triage of new journal submissions
- Pre-peer-review integrity screening
- file_read, browser, network automation included
- Instant install
Sample input
Screen this abstract for research-integrity red flags before I send it out for peer review: "In this work we apply counterfeit consciousness and profound learning to bosom malignancy classification. Across 10 patients, accuracy was 33.3% (p = 0.000). Our irregular woodland model bolsters the discoveries." Report any concerns with evidence and a concern level.
Sample output
Overall concern level: HIGH (triage signal, not a determination). A. Tortured phrases (STRONG): "counterfeit consciousness" (artificial intelligence), "profound learning" (deep learning), "bosom malignancy" (breast cancer), "irregular woodland" (random forest). Four synonym-swapped technical terms; no innocent stylistic origin. C. Statistical impossibility (STRONG): "33.3% of 10 patients" is not achievable — 3/10 = 30%, 4/10 = 40%; no integer count yields 33.3% at N=10. "p = 0.000" reported without a bound. Innocent explanation: transcription/rounding error, but combination is implausible. Convergence: two STRONG, independent signal classes (A + C) → HIGH concern. Recommended next steps for a human reviewer: request the raw data and exact counts; confirm sample size; verify any cited references online; route to the editor/research-integrity office per venue policy. Disclaimer: This is an automated first-pass triage. It identifies signals, not misconduct, and may produce false positives. No statement asserts that any person engaged in fraud.
About This Skill
Journals, universities, and research-integrity offices are being flooded with AI-generated and paper-mill submissions, and the available screening tools are still immature. This skill encodes the red flags that experienced integrity investigators look for and turns any compatible agent into a structured, evidence-cited first-pass screener. It runs six detector passes over a manuscript, abstract, or review report: A. Tortured phrases & lexical fingerprints — established technical terms mangled by synonym-swapping paraphrase tools (e.g., "counterfeit consciousness" for artificial intelligence). B. Citation fabrication & manipulation — hallucinated references, DOI/title mismatches, citation-content mismatches, and coercive self-citation patterns. C. Statistical & numerical impossibilities — GRIM/GRIMMER inconsistencies, impossible percentages, p-value vs test-statistic conflicts, and out-of-range effect sizes. D. Template & paper-mill fingerprints — shared skeletons with swapped topic nouns, placeholder residue, and recycled figures. E. AI-generation stylistic markers — filler scaffolding and hedged uniformity (weighted low; only counts in clusters). F. Metadata & provenance anomalies — affiliation/email mismatches, brokered-authorship language, and version churn. Every flag is reported with the exact quoted span, its location, a confidence level, and at least one innocent explanation, then aggregated into a calibrated concern level (LOW / ELEVATED / HIGH) using a convergence-based scoring rubric. The skill is deliberately built as triage, not verdict: it never asserts misconduct, never screens on author identity or nationality, and always ends with recommended next steps for a human reviewer. What's included: a lean SKILL.md orchestrator plus a bundled REFERENCE.md containing the detailed detector playbooks, the convergence-based scoring rubric, and a ready-to-fill report template. Designed for editors, peer reviewers, and research-integrity officers who need a fast, defensible, human-in-the-loop first pass before deeper investigation.
Use Cases
- Editorial triage of new journal submissions
- Pre-peer-review integrity screening
- Research-integrity office case intake
- Checking a preprint or abstract for red flags
Known Limitations
This is a triage screener, not a determination of misconduct. It produces probabilistic signals that require human review and can yield false positives (e.g., non-native English, legitimate methods boilerplate, niche terminology). Citation verification requires web access; without it, references are reported as unverifiable rather than confirmed. It cannot inspect figures/images in image-only PDFs, perform pixel-level image forensics, or detect signals not present in the supplied text. It does not screen on author identity or nationality by design.
How to Install
mkdir -p ~/.claude/skills && curl -sL https://www.agensi.io/api/install/ai-paper-mill-research-integrity-screener -o /tmp/ai-paper-mill-research-integrity-screener.zip && unzip -o /tmp/ai-paper-mill-research-integrity-screener.zip -d ~/.claude/skills && rm /tmp/ai-paper-mill-research-integrity-screener.zipFree skills install directly. Paid skills require purchase - use the download button above after buying.
Reviews
No reviews yet - be the first to share your experience.
Only users who have downloaded or purchased this skill can leave a review.
Early access skill
Be the first to review this skill.
Only users who have downloaded or purchased this skill can leave a review.
Security Scanned
Passed automated security review
Permissions
Allowed Hosts
File Scopes
Read Files is needed to ingest the manuscript/abstract being screened. Browser and Network Access are optional and used only to verify DOIs and citations against public registries (e.g., Crossref, doi.org); when unavailable, the skill reports references as unverifiable rather than guessing. The skill does not write files, run shell commands, or read environment variables.
Universal: works with any agent that supports the open SKILL.md standard. No runtime dependencies and no required external services. Best results when the agent has web access to verify DOIs and citations against public registries; without web access, the skill explicitly reports references as unverifiable rather than guessing.