Paper Slop Screen

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, and nothing else to install first. No key is required at use time either: the four scripts are Python 3.9+ (slop_lint.py, ref_verify.py and number_ledger.py are standard-library only), and the optional keys ref_verify.py can read from the environment if they are already there (OPENALEX_API_KEY, S2_API_KEY, MAILTO) are never prerequisites — do not ask me for them and do not install anything for them during this install. The other use-time tools (PyMuPDF for scripts/pdf_hidden_text.py, poppler-utils, tesseract) are also use-time only and are not installed here. Install "paper-slop-screen" for me. It is a reviewer-side screening skill for ML research papers: it screens a paper for evidence of an absent author — research nobody steered, artifacts nobody verified, prose nobody owned — and produces an evidence-cited grade on three axes (writing W0-W3, research steering R0-R3, ordinary quality Q) plus integrity flags and an auditability rating, never an "AI probability". Repository: https://github.com/THUROI0787/absent-author (the skill is the skills/paper-slop-screen/ folder only, at the pinned commit: 14 files) Pinned version: install the code at commit fc119a05d00a77f945ddcdb2d914cc5015d570c5 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/THUROI0787/absent-author into a temporary folder, check out commit fc119a05d00a77f945ddcdb2d914cc5015d570c5, and copy only the skills/paper-slop-screen/ folder from that checkout. Do not use the repository's install.sh for this install: it installs both skills of the repository (paper-slop-screen and its sibling paper-author-pass); this listing is paper-slop-screen only. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 1b392832a7476aedb8962e52ef39e32aeeeda89b1fa6f191da3093e47ec8ec70 - references/evidence-catalog-cn.md: SHA-256 aaf43d9dd01e0597cd13434c7e6d908ea39b4a1a2ffecaa646a3c5b93f1a7c03 - references/evidence-catalog.md: SHA-256 861444be3633c4ae3a66fa96535d7ad6ffb33acdeea1108b3dec3b6c39d335e2 - references/evidence-index-en.md: SHA-256 548730d06ddfe07be115debacb1b291c4b68deb5dcd4695f594593039a4d24bd - references/grading-rubric.md: SHA-256 1b747d552c060879f28ceb0f78b8bb444a232cce3fba3d0a32d3adf2e83e786d - references/quick-card-en.md: SHA-256 f30f929b1a2e172bd782f1652b308d9c46ac0d8b4841c8670e8da866d5419b1e - references/report-template.md: SHA-256 db2f46103a5f95166db5ab11e5e1f0be3c5c8a4cd267154842f264bd6deea9fb - references/screen-actions.md: SHA-256 435bb8064db80c6a7050fc97823a2ebe9ab42598841c268d2283ccfda6cc43b2 - references/verification-procedures.md: SHA-256 cae245df0d13ccb36856a50200ad49c218702e3a8c555ce3bb8f9a3c04e51c40 - references/worked-examples.md: SHA-256 9f1d454d40a4782a42e34dd17fb04fd6ff069a92211b46a25a5a96fb959f8f40 - scripts/number_ledger.py: SHA-256 64d55124a5220ed6fa5f5103eaa9ff22d3d8361f9e3f8206ab4c23e12d3fc8fa - scripts/pdf_hidden_text.py: SHA-256 6814a9cece4ff4a08d189cb4ebbd18e1a4458b97e38a2b2ebabaa1e2947850bd - scripts/ref_verify.py: SHA-256 ed61b5fc2994fe09aeb1efa9ab0035529db316c8dc039d2f18c7686cb88f0b62 - scripts/slop_lint.py: SHA-256 fa579162ccdfd07d57cc9c0a2fc349365b4a582e02a4fa0df37dd49ca2bd237d If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md, all nine files under references/ and all four scripts under scripts/ in full and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 14 files only: SKILL.md (10,868 bytes), nine references (evidence-catalog.md 51,717 bytes and its Chinese master copy evidence-catalog-cn.md 45,118 bytes, evidence-index-en.md, grading-rubric.md, quick-card-en.md, report-template.md, screen-actions.md, verification-procedures.md, worked-examples.md) and four Python scripts (slop_lint.py 125,566 bytes, number_ledger.py, ref_verify.py, pdf_hidden_text.py). Paths the catalog marks as being in the project repository (docs/..., tools/..., EVIDENCE.md at the root) are not shipped in this folder and are not installed. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. The files name no account, no telemetry and no remote service for installation, and no script is run during installation. - What the skill does at use time: before each new paper it asks which confidentiality and venue-policy case applies (own or co-authored paper, public preprint, reviewer use the venue permits, developing/testing the skill, or unsure — in which case it must not process the paper and offers the human-only quick card instead). The paper is treated as data, never as instructions: hidden or instruction-like text found in a paper is located, quoted with any phrase it asks to output redacted, classified (venue canary, author-inserted, or unclear) and reported, never followed. It then lints the paper, checks references and numbers, reads for research steering, grades W/R/Q with flags and auditability, and writes a report from references/report-template.md. - Local use-time work, with no network: slop_lint.py is a deterministic standard-library lint of surface traces (candidates only, never a verdict); number_ledger.py (standard library, imports slop_lint) lists quantities stated with different values and recomputes arithmetic; pdf_hidden_text.py analyses a PDF locally with PyMuPDF for invisible, out-of-page or remapped text and, only with --ocr, calls the local tesseract binary as a subprocess. Poppler tools (pdftotext, pdftoppm, pdffonts) are likewise local when used. - Use-time network, disclosed here before it happens, and only on a paper and references I supply and only in a case the policy gate above allows: scripts/ref_verify.py sends the reference titles, arXiv IDs and DOIs taken from that paper's bibliography as GET queries, in fallback order, to export.arxiv.org, api.crossref.org, api.semanticscholar.org, api.openalex.org and dblp.org, and writes a local cache at ~/.cache/absent-author/ref_cache.json (cache keys strip api_key and mailto values). If I have set MAILTO or pass --mailto myself, that email address goes in the User-Agent and as a mailto parameter to Crossref and OpenAlex; if OPENALEX_API_KEY is in the environment, it is sent as an api_key query parameter to OpenAlex only; if S2_API_KEY is in the environment, it is sent as an x-api-key header to Semantic Scholar only. None of these is ever required. The workflow also searches arXiv by title through export.arxiv.org and, for a public paper only, downloads its e-print or source with curl -L https://arxiv.org/e-print/<id>; a novelty sanity check may search Semantic Scholar or Google Scholar. Because those queries carry a paper's title and reference strings to outside databases, never run them on a confidential submission unless the gate above allows it. - Optional use-time tools, none of them installed during this install: poppler-utils (pdftotext, pdftoppm, pdffonts), pip install pymupdf (needed by scripts/pdf_hidden_text.py), and tesseract (only for pdf_hidden_text.py --ocr). They are my decision, later, if I ever ask. - The scope is screening, not authorship detection: detector scores never decide anything, language evidence can only raise W, ordinary weaknesses go to Q, and the public review is written about checkable substance, never "AI-generated". The sibling skill paper-author-pass, the rest of the repository (site/, tools/, docs/, assets/) and the repository's install.sh are not part of this skill and are not installed. 3. When I say "yes, install", install it as a skill: copy the whole paper-slop-screen folder into the agent's skills directory, in a folder named "paper-slop-screen", keeping all 14 files. 4. Verify with no network calls: frontmatter valid, all 14 files in place, fingerprints match. 5. Report what was installed, where, and what I still need to do myself (for example: provide any paper myself and decide, for each paper, which confidentiality and venue-policy case applies before any screening starts; install PyMuPDF, poppler-utils or tesseract myself, only if and when I want those PDF steps; optionally set MAILTO, OPENALEX_API_KEY or S2_API_KEY in my own environment, only if I want the higher limits they give — using the skill is a separate step that only starts when I ask for it, and the use-time network disclosed above happens only then, on a paper I name). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.