rseng-fact-checking

Solid

Covers verifying facts and sources at the content level: checking that a cited source actually contains and supports the claim it is cited for (claim-source alignment), assessing source trustworthiness (peer-review status, venue reputation, predatory-publishing signals, primary vs secondary), and flagging fabricated or misattributed support. Use PROACTIVELY whenever the agent itself asserts checkable facts or attaches sources to claims, and when the user asks to fact-check a document, verify that references support their claims, or assess whether a source is trustworthy. (rseng-citation-hygiene verifies references exist and are unretracted; rseng-research-integrity checks a document's own numbers.)

AI & Automation 14 stars 2 forks Updated 4 days ago MIT

Install

View on GitHub

Quality Score: 83/100

Stars 20%
39
Recency 20%
100
Frontmatter 20%
70
Documentation 15%
100
Issue Health 10%
80
License 10%
100
Description 5%
100

Skill Content

# Fact-checking and source verification A reference can exist, resolve and be unretracted - and still not say what it is cited for. That last mile is where fabrication actually lives: real papers attached to claims they do not support, secondary sources laundered as primary, numbers that drifted in retelling. rseng-citation-hygiene verifies that references EXIST and are current; this skill verifies that they SUPPORT their claims and deserve trust. The duty is self-referential and stated plainly: an agent that asserts a fact or attaches a source must be prepared to show where in the source the support lives - and must say "I could not verify this" when it cannot. Confidence is not evidence. ## Claim-source alignment: the core check For each claim-citation pair: 1. Extract what the claim actually asserts: the specific finding, number, method or position attributed to the source - not the general topic. Most misattribution hides in specificity ("X causes Y" cited to a paper showing correlation; "widely used" cited to a paper that merely mentions). 2. Open the source and FIND the support: the abstract often suffices for existence of a finding (OpenAlex serves abstracts); exact numbers, conditions and caveats need the relevant section. Record WHERE the support is (section, table) - unfindable support is the finding. 3. Grade the alignment honestly: supports as stated / supports with caveats the text drops / related but does not support / contradic...

Details

Author
fdiblen
Repository
fdiblen/rseng-agent-skills
Created
4 days ago
Last Updated
4 days ago
Language
Python
License
MIT

Similar Skills

Semantically similar based on skill content — not just same category

Data & Documents Solid

rseng-citation-hygiene

Covers verifying that every citation is real, correct and current: checking references in manuscripts, READMEs, references files and code metadata against Crossref and OpenAlex, screening cited DOIs against the Retraction Watch database, catching fabricated or mis-attributed citations (a documented AI failure mode), and keeping bibliographies and CITATION.cff files resolvable. Use PROACTIVELY before any bibliography, reference list or citation metadata is finalized or published, whenever the agent itself has produced citations, and when the user asks to check references, mentions broken DOIs, retracted papers or citation verification, or prepares a manuscript, README or software paper with references. (Making your own software citable is rseng-citation-metadata.)

14 Updated 4 days ago
fdiblen
AI & Automation Listed

fact-checking

Audit claims, citations, recency, and confidence. Use when facts need verification.

2 Updated 1 months ago
kirilenkofedir-max
Data & Documents Listed

factcheck

Adversarial citation and claim audit for any document — "do these sources actually say what the text claims?" Takes a markdown file, URL, or pasted draft (including this repo's own pulse/verdict briefs), extracts every checkable claim, then fans out verifier sub-agents instructed to refute, not confirm: does the cited link resolve AND assert the claim as stated, is the source primary or a laundered secondary write-up, has the claim gone stale. Emits a graded audit table — VERIFIED / MISATTRIBUTED / UNSUPPORTED / STALE / UNCHECKABLE — with a quoted passage as evidence for every grade, saved to out/factcheck/. Advisory: it grades, the human edits. Use when the user wants to check citations, verify sources, or audit a report, post, or brief — e.g. "/factcheck <doc>", "verify the sources in this", "are these citations real?", "is this actually true?". For researching a fresh topic use pulse; for judging code changes use done.

9 Updated 3 days ago
duthaho