by THUROI0787
Research and writing can be automated. Authorship cannot. Evidence list + agent skills for spotting AI-produced papers no human stood behind.
# Add to your Claude Code skills
git clone https://github.com/THUROI0787/absent-authorGuides for using ai agents skills like absent-author.
See how absent-author compares with popular alternatives.
absent-author is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by THUROI0787. Research and writing can be automated. Authorship cannot. Evidence list + agent skills for spotting AI-produced papers no human stood behind. It has 56 GitHub stars.
absent-author's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/THUROI0787/absent-author" and add it to your Claude Code skills directory (see the Installation section above).
absent-author is primarily written in Python. It is open-source under THUROI0787 on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh absent-author against similar tools.
No comments yet. Be the first to share your thoughts!
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
Reviewers now see papers that an agent produced end to end, with no human who steered, checked or owns them. This repository names the signs of that absence, with sources, and provides tools that turn them into evidence a reviewer can cite and an author can fix.
| You are | Use | You get |
|---|---|---|
| an author finishing an AI-assisted paper | paper-author-pass |
questions only you can answer, a claim–evidence map, verified references and numbers, then lighter prose; it ends with three questions to answer before you sign |
| a reviewer or AC facing a suspicious submission | paper-slop-screen |
W / R / Q grades, a quadrant, every finding with a location and a quote, a review paragraph and a note to the AC |
| anyone who wants a quick surface scan | tools/slop_lint.py |
calibrated counts of 28 surface traces, as candidates for a human to read |
| a maintainer of review norms | EVIDENCE.md |
83 sourced items with strengths, false-positive notes and paper-type exemptions |
git clone https://github.com/THUROI0787/absent-author.git && cd absent-author
./install.sh # both skills -> ~/.claude/skills (or --project, --codex, --dest DIR, --link)
Then, in Claude Code, Codex or any agent that reads SKILL.md:
Use paper-author-pass on paper/ in audit mode. I'll answer your questions.
Use paper-slop-screen to triage this arXiv preprint: <path or id>
The lint also runs on its own (Python 3.9+, no dependencies), and three helpers do the slow verification work:
python tools/slop_lint.py paper/ --source -o lint.md # surface traces; skips prompt dumps and checklists
python tools/ref_verify.py refs.bib --sample 5 --mailto you@example.org # every author, incl. the last
python tools/number_ledger.py paper/main.tex # same quantity stated with different values
python tools/pdf_hidden_text.py paper.pdf # hidden or remapped text (pip install pymupdf)
All four report candidates for a person to read, never verdicts. Optional system tools: poppler-utils (pdftotext, pdftoppm) and tesseract (OCR check for remapped text).
[!IMPORTANT] Before you run the screen on a submission under review, check your venue's reviewer LLM policy. ICML 2026 desk-rejected 497 papers linked to reviewers who broke the no-LLM policy they had chosen. The skill asks you to confirm this first.
- L-layer cluster: 5 of 6 L-cluster checks are above the human p90 (L03, L04, L05, L08, L10).
In calibration, 0% of human papers and 50% of AI-heavy papers reached at least this many.
Writing-polish signal only; 2026 Claude-based pipeline papers scored 0-1 here.
| ID | Check | Count | per 1k | Band | Human p50 / p90 / p99 | AUC |
| L03 | AI-associated lexicon cluster | 20 | 46 | high | 0.604 / 1.78 / 5.92 | 0.927 |
| L08 | Trailing -ing clauses | 3 | 6.9 | high | 0 / 0.202 / 0.809 | 0.961 |
| S01 | Defensive pre-emptive hedging | 6 | 13.8 | high | 0 / 0 / 0.485 | rare |
| S03 | Instruction / revision leakage | 6 | 13.8 | high | 0 / 0 / 0 | rare |
| P05 | Pipeline watermark / signature | 4 | 9.2 | high | 0 / 0 / 0 | 0.75 |
Every hit comes with a file, a line and a snippet. Hits are candidates; a clean report proves nothing. For what it is worth, this README scores 1 of 6 on the same cluster, with zero em dashes.
Writing (W) W1 clean, field-idiomatic prose; L-cluster 1/5; em dashes 6.3 per 1k words
Research (R) R3 R09 confirmed in three forms; R10; R15
Flags 2 P03: ref [13] cites "arXiv:2409.XXXXX" and cannot be found
R09: one configuration is 63.4 CIDEr in Tables III, IX, X and 57.2 in Table V
Quadrant C veneer: clean prose, unchecked research
Evidence
R09 ★★★ §III-B "we retain 2,077 pairs" 0.85 × 2,524 = 2,145, not 2,077
R09 ★★★ §VI-B "12.4% of test pairs" 12.4% of 207 = 25.7, not a whole number
Review-ready paragraph: substance only, no claims about provenance.
Note to the AC: questions the authors can answer (reconcile Table V with Tables III/IX/X; provide ref [13]).
Six worked examples, including the hardest call: worked-examples.md.
Reviewers already have intuitions about this. What has been missing is a shared vocabulary with evidence behind it: something a reviewer can cite in a review, an AC can act on, and an author can check before submitting. Sources for every claim are in docs/SOURCES.md.
Slop outsources the cost of verifying a paper to its reviewers. An author can be absent in three ways:
| Absence | What it looks like | Layer |
|---|---|---|
| Nobody polished it | The writing is AI-led and unowned: dash floods, "not X but Y", defensive caveats, coined terms, performative honesty | L, some S |
| Nobody steered it | No human judged which question was worth asking or what to do when the plan failed: pivots into audit papers, theoryslop, toy scale with grand claims | R, some S |
| Nobody checked it | The final artifact was never verified: fabricated references, chatbot residue, numbers that disagree between text and table, promised but missing analyses | P, R |
Two questions place a paper in one of four quadrants. Ordinary weaknesses go to a separate quality axis (Q), so a weak human paper is never called slop.
B (salvageable) and D (AI waste) are separated by the research axis alone. B gets a concrete rewrite list and is not rejected for its prose alone. D gets a reject-level review based only on defects anyone can verify, and a confidential evidence table for the AC written as questions the authors can answer. See docs/POSITIONING.md for the full reasoning, including why disclosed AI involvement may soon count in a paper's favour.