More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    EvidPort Evidence: traceable paper quotations

    1

    Checking an AI's citations by hand means opening every PDF and hunting for every sentence.

    Free

    Security scanned
    EvidPort Evidence: traceable paper quotations

    EvidPort Evidence: traceable paper quotations

    Example session with this skill installed

    I'm drafting the introduction of a review on endometriosis. From my EvidPort library, find what the 2021 transcriptome meta-analysis (PMID 33725527) concludes about eutopic endometrium and give me a sentence I can quote. Check it against the PDF yourself so I don't have to, and tell me if anything in the library needs me.

    • Read your context and instructions
    • Compiled the evidport evidence

    Short answer

    Here is a sentence you can quote. I checked it against your stored PDF myself: right paper, exact wording,

    page 6. The only thing left for you is one look at that page region.

    The quotation

    "The results of our meta-analysis indicate that the eutopic endometrium of women …"
    (shortened here; the full sentence was checked)

    PaperEutopic endometrium from women with endometriosis and chlamydial endometritis share immunological cell types and DNA repair imbalance (2021), doi:10.1016/j.jri.2021.103307
    WherePage 6 of the stored 15-page PDF, located to a region of the page
    Citationevidence:cv_34c53760c40fe7039145fa49b5092efa0edac1b1e3dd1ec788610b9edb117db7#b000083
    Source file SHA-25612046614f7c75de25c4a5f3ad262c1c8069640e77f11b9c7a4ba5e288d59098b
    WordingExact match (quote_matches: true)
    NoteTables and numbers in this PDF are flagged for a closer look; this sentence is an ordinary paragraph with no warnings of its own

    Checked for you

    • The stored PDF is confirmed to be this paper.
    • The sentence is located to a region of page 6.
    • The full sentence matches the stored text exactly.
    • An invented sentence tested against the same passage was rejected (quote_not_found), so a made-up quote would not have got this far.

    Left for you

    • One look at page 6. I can put the sentence next to the page region so you confirm it in one glance.
    • Whether the finding fits your argument.

    Your library

    11 papers, 9 searchable.

    StatusPapersNeeds you?
    Fully extracted4No
    Partly extracted5No, unless you quote from them; I will show the warnings
    File identity unconfirmed2Yes: confirm these two PDFs are the right papers

    Values are from a real EvidPort run on 2026-11-11 against a private library: citation, hash, counts and check results are unchanged. The conversation is written up for readability.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Checking an AI's citations by hand means opening every PDF and hunting for every sentence. This skill has your agent do that work itself through EvidPort, a free open-source local evidence layer, and hand you only what a machine cannot settle.

    Done automatically

    • Finds free-access full text for a DOI, or tells you exactly which PDF to download and what to name it.
    • Confirms the file is the right paper before any of its text is used.
    • Stores the original with its SHA-256 and extracts the text as a pinned version, each passage tied to its page and region.
    • Checks the complete quotation against the stored text. Invented or altered quotes are rejected on the spot.
    • Keeps older versions, so a new PDF or a new extraction never silently replaces what was checked.

    What is left for you

    • One look per quotation you rely on. The agent prepares a packet with the sentence next to its original page region, so you confirm it instead of searching for it.
    • PDFs that are not freely available, and the occasional file whose identity cannot be confirmed automatically.

    In a recorded run

    In 12 extraction runs over 9 papers, all 12 real quotations were located and matched, all 12 fabricated quotations were rejected and a wrong PDF was refused at import, with no person involved. This was an engineering check, not a benchmark.

    What you need

    The skill is instructions only: no scripts, no network calls, no credentials. It makes quotations traceable and checked; whether the science is right is still your call.

    Requires EvidPort installed locally; it is not included. For a quotation to count as reviewed, a person still confirms it against the original page: the skill cuts that to one prepared look but does not remove it. Extraction quality depends on the PDF and parser, and warnings are passed through. No guarantee of exhaustive literature coverage or scientific validity, and the time saved has not been measured. Only Claude Code has been used with the EvidPort MCP server so far.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed today

    • Free with an Agensi account

    Requires

    Git
    Python 3.11+
    EvidPort (free, open source)
    Agent With Local Stdio Support MCP
    Python

    Permissions

    Write Files
    Network Access
    Terminal / Shell

    The skill ships no scripts. Setup runs git clone and pip install in a terminal. The EvidPort server writes its library, queue and receipts to local folders the user configures. It is offline by default; network is used for installation and, only if the user starts the server with --allow-network, for paper metadata and free-PDF providers.

    Listedtoday

    What's inside

    Frequently Asked Questions

    One look per quotation you rely on. The agent finds the file, confirms it is the right paper, locates the sentence and checks the exact wording by itself, then prepares a packet showing the sentence next to its original page region. You confirm it; you do not search for it.

    They are rejected automatically. The complete quotation is compared with the stored text, and anything that is not there comes back as quote_not_found. In a recorded engineering run, all 12 fabricated quotations were rejected.

    Yes. It needs the EvidPort MCP server installed on your own computer. EvidPort is free and open source (Apache-2.0): https://github.com/NingyuSUN/evidport. The skill includes a setup guide.

    Not for checking quotations in papers already in your library: the server is offline by default. To let it fetch paper metadata and free PDFs for you, start the server with --allow-network. Text the agent reads is still sent to your model provider.

    A SKILL.md with the workflow, a setup guide and a tool reference. No scripts. EvidPort itself is not included; you install it separately from GitHub at no cost.

    The exact wording, the paper, the page or region, a citation ID, the SHA-256 of the source file and the check result, plus any extraction warnings.

    New in Research & Analysis