agentleFS
holon-runholon-run/holon/skills/citation-verifier/SKILL.md

citation-verifier

Skill197 starsChanged yesterdayApache-2.0View on GitHub

Verify that research claims are supported by traceable sources, while preserving scope, freshness, limitations, and contradictions.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 11 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 11

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill after sources have been collected and before a research brief is delivered.

    line 10
  2. Very specific:

    If an input is missing, report the missing field and mark affected claims unverified.

    line 23
  3. Distinctive:

    Check entailment: does the cited passage support the exact proposition, or only a weaker/narrower statement?

    line 30
  4. Distinctive:

    Check scope: compare geography, population, version, methodology, definitions, and date.

    line 33
  5. Distinctive:

    Check freshness: distinguish publication date, last update, and access date.

    line 35
  6. Distinctive:

    Check source quality separately from entailment: identify primary, secondary, official, self-reported, anonymous, or untraceable material …

    line 37
  7. Distinctive:

    Search for counter-evidence when the claim is material, disputed, surprising, or supported by only one weak source.

    line 40
  8. Distinctive:

    Keep the distinction explicit:

    line 62
  9. Distinctive:

    Never invent a URL, DOI, title, date, quotation, page number, or source identity.

    line 73
  10. Distinctive:

    Never cite a source you did not inspect when the task requires claim-level verification.

    line 75
  11. Distinctive:

    Never remove a contradictory source merely because it is inconvenient.

    line 77
  12. Distinctive:

    Do not use citation count, search rank, repository stars, or model agreement as a substitute for entailment and scope checks.

    line 78

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Citation Verifier Skill

Purpose

Use this skill after sources have been collected and before a research brief is delivered. It checks the relationship between a claim and its cited evidence; it does not decide whether a conclusion is desirable and it does not turn search results into evidence.

Required inputs

  • A claim list with stable IDs, wording, and the importance of each claim.
  • A source ledger with stable IDs, title, publisher or author, locator, publication/update date when available, access date, scope, and limitations.
  • The relevant excerpt, table, section, or reproducible locator for each claim/source pair. A homepage or search snippet is not a sufficient locator.

If an input is missing, report the missing field and mark affected claims unverified; do not fill it with a guessed citation.

Verification procedure

  1. Normalize each claim so that it has one testable proposition, a defined population or object, a time window, and any important qualifier.
  2. Check entailment: does the cited passage support the exact proposition, or only a weaker/narrower statement? Record a short paraphrase, not an invented quotation.
  3. Check scope: compare geography, population, version, methodology, definitions, and date. A source can be credible and still be out of scope.
  4. Check freshness: distinguish publication date, last update, and access date. Flag time-sensitive claims whose evidence may have changed.
  5. Check source quality separately from entailment: identify primary, secondary, official, self-reported, anonymous, or untraceable material and record the reason it matters.
  6. Search for counter-evidence when the claim is material, disputed, surprising, or supported by only one weak source. Preserve conflicts rather than collapsing them into an average.
  7. Assign one terminal status: supported, partially-supported, contradicted, unverified, or blocked. Include the reason, source IDs, and the next check if unresolved.

Output contract

Return a claim/evidence matrix with at least these columns:

claim_idclaimsource_idslocatorscope/date checksource qualitycounter-evidencestatusreason

Also return:

  • verified_at: the date/time of the verification pass;
  • coverage: material claims checked versus total material claims;
  • conflicts: a list of unresolved or context-dependent conflicts;
  • unverified: claims that must not appear as established facts;
  • next_checks: the smallest useful follow-up actions.

Keep the distinction explicit:

  • supported means the cited material entails the claim within its scope;
  • partially-supported means only a qualified or narrower claim is entailed;
  • contradicted means credible evidence materially conflicts with the claim;
  • unverified means evidence is missing, inaccessible, or does not entail it;
  • blocked means the check cannot proceed because access or required input is unavailable.

Safety rules

  • Never invent a URL, DOI, title, date, quotation, page number, or source identity.
  • Never cite a source you did not inspect when the task requires claim-level verification. Label operator-provided excerpts as excerpts.
  • Never remove a contradictory source merely because it is inconvenient.
  • Do not use citation count, search rank, repository stars, or model agreement as a substitute for entailment and scope checks.
  • For high-risk topics, report verification limits and require qualified human review before the result is used as advice.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.