citations
Skill0 starsChanged 37 days agoMITView on GitHub
Use for academic citation requests, including citing or formatting a paper or article in APA, MLA, Chicago, or Harvard style (even a single one), fixing or checking BibTeX, verifying references or a bibliography, auditing a .bib file, or exporting citations as BibTeX, Markdown, or JSON. Verifies, corrects, formats, and exports references with the Corbis research connector.
8 other projects use a version of this.
Adopted by
Most recently taken up by
- liangdabiao/claudesdk-deep-research-chattheir own version, first seen Nov 8, 2026
- matthull/my-skillstheir own version, first seen Oct 20, 2026
- ZimoLiao/scholaraiotheir own version, first seen Oct 8, 2026
- polycode-public/claude-code-idea-refinerytheir own version, first seen Aug 23, 2026
- blonderoofrat/agent-skillstheir own version, first seen Jul 31, 2026
- matteing/opaltheir own version, first seen Feb 16, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- ZimoLiao/scholaraio★ 585same name
- matteing/opal★ 59same name
- liangdabiao/social_research_agent★ 45same name
- liangdabiao/simple_claude_deep_research_agent★ 37same name
- liangdabiao/claudesdk-deep-research-chat★ 7same name
- blonderoofrat/agent-skills★ 2same name
- polycode-public/claude-code-idea-refinery★ 1same name
- matthull/my-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- investment-allocation-briefSkill
- literature-reviewSkill
- onboardingSkill
- paper-reviewSkill
- real-estate-investment-briefSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 7 edited versions we read keep these.
3
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
- Distinctive: line 8
Check references against the Corbis paper index and format them without adding anything the tools did not return.
- Distinctive: line 26
Unverifiable: no confident match. Keep the user's entry as written and say what is missing or ambiguous.
- Very specific: line 33
Run Quick steps 1 to 4, except that full mode has no 5-call cap: keep calling verify_bibtex in batches at the entry cap until every entry …
- Distinctive: line 36
Keep unverifiable entries out of verified output and list them separately for the user to fix.
“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.
Files
Citations
Check references against the Corbis paper index and format them without adding anything the tools did not return.
Ground rules
- Cite only papers, data, and quotes that a Corbis tool returned in this conversation or that the user supplied. Never invent a DOI, author, year, title, or quote.
- Say when evidence is thin, old, mixed, or limited to one market or domain. Never describe the output as free of errors or hallucinations.
- Some optional steps use premium tools that need an enterprise plan. If a premium tool is not in your tool list, or returns an access or plan error, tell the user once that the step needs an enterprise plan (current plans: https://www.corbis.ai/pricing), then continue with the standard tools. Never retry a denied tool.
- Research tool calls may use Corbis credits. Use a batch tool when one exists, and never re-fetch details for a paper already retrieved in this conversation.
- Quick mode is the default for a narrow question: at most 5 Corbis tool calls, a short cited answer, then an offer to run the full workflow. Use full mode when the user asks for a review, report, or audit, or accepts that offer.
- If no Corbis tools are available, or a Corbis call returns a sign-in or connection error, first retry the call once through any other Corbis connection in your tool list (the same tool name from another Corbis server). That retry counts toward the quick-mode limit, so skip it when no calls remain. If none works, ask the user to connect or reconnect Corbis (the plugin's Connectors tab, or
/mcpin Claude Code) instead of answering from memory. Do not switch to another source, such as a web search or another citation database, unless the user asks, and label anything from another source as not verified by Corbis.
Quick: verify BibTeX
- Take the BibTeX the user pasted or uploaded. If they gave formatted references instead, ask whether to convert them to BibTeX entries first, and convert only the fields they wrote.
- Call
verify_bibtex, settingmaxEntriesto the number of entries you have, up to the tool's maximum. A bibliography that needs more than one call to cover, staying within 5 total Corbis tool calls, can split across calls in quick mode; a bibliography that needs more than 5 calls does not fit quick mode, so say so and offer full mode instead. - Compare the tool's total entry count against how many it actually checked, then report every entry in one of four groups:
- Matched: the entry agrees with the index.
- Corrected: list each changed field as old value and new value. An
authorcorrection is only a first-author surname hint, not a full author list: keep the user'sauthorfield and flag the mismatch instead of replacing it. - Unverifiable: no confident match. Keep the user's entry as written and say what is missing or ambiguous. Never fill in guessed fields.
- Not checked: any entry beyond what the tool actually verified. Never silently drop these.
- Give one corrected BibTeX block that uses only the tool's corrections (except
authorhints, per step 3) and the user's original fields. - Offer to format or export the verified list.
Full: verify, then format or export
- Run Quick steps 1 to 4, except that full mode has no 5-call cap: keep calling
verify_bibtexin batches at the entry cap until every entry is checked, then report all of them. - For matched and corrected entries, call
render_citationswith the matched work IDs or DOIs thatverify_bibtexreturned, in the requested format (BibTeX, Markdown, or JSON). Itsexportsalready hold the file content and a suggested filename, built from full corpus metadata including complete author lists: give those to the user as the file, exactly as returned, without editing them. Any BibTeX you compose (such as the corrected block from Quick step 4) uses only fields the user supplied or a tool returned in this conversation: never add a volume, issue, pages, publisher, or any other field from memory. Useexport_citationsonly for metadata the user supplied, never to rebuild records from rendered text. - For a named style (APA, MLA, Chicago, or Harvard), first call
get_paper_details_batchon the matched IDs or DOIs, at most 25 per call, becauseverify_bibtexreturns match IDs and correction notes, not full records. Then callformat_citationwith that style and the returned metadata, at most 50 papers per call, and combine the results in order. For an entry with no match, use only complete metadata the user supplied, and say that formatting supplied fields does not verify them. - Keep unverifiable entries out of verified output and list them separately for the user to fix.
Single citation
For one paper in a named style ("cite this in APA"): call format_citation with that style, using the metadata a Corbis tool already returned if the paper came from Corbis in this conversation, or the metadata the user gave otherwise. For a Markdown or BibTeX reference on a paper that came from Corbis, use render_citations instead. Either way that is one tool call.
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
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.

