More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Autonomous Local Contract Auditor

    1

    Offline first-pass audit of invoices and contracts through your own Ollama model, plus two scripts that check the figures and flag OCR letter/digit mix-ups. Built for CPU-only laptops. Not legal advice.

    $21

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Autonomous Local Contract Auditor

    Autonomous Local Contract Auditor

    Example session with this skill installed

    Run the audit on this scanned invoice and tell me what to double-check before I pay it.

    .\audit.ps1 -File .\tests\02_ocr_noisy_invoice.txt

    ACME DESIGN STUDIO - Page 1 of 2
    lnvoice No. 2O26-0147
    Date: 12 Oct 2O26
    Bill to: Harbor Cafe

    Description Qty Amount
    Logo redesign 1 8OO.OO
    Menu layout 1 4OO.OO

    Sub total 1,2OO.00
    VAT 1O% 12O.OO
    Totai due 1,23O.OO

    Payment terms: Net 3O days from invoice date.
    Late fee: 1.5% per rnonth on overdue amounts.

    ACME DESIGN STUDIO - Page 2 of 2
    Thank you for your business.

    • Read your context and instructions
    • Compiled the autonomous local contract

    Summary

    Audit of a noisy OCR invoice (our test file 02) with llama3.1:8b in QUICK mode. The model copied every figure exactly as written, and two scripts then checked the result.

    1. EXECUTIVE SNAPSHOT

    • Document Type: Invoice
    • Parties Involved: ACME DESIGN STUDIO, Harbor Cafe
    • Effective Date: 12 Oct 2O26
    • Expiration / Term: NOT SPECIFIED
    • Text Quality: CLEAN
    • Overall Risk Rating: NOT RATED (QUICK MODE)

    2. FINANCIAL & TIMELINE OBLIGATIONS

    Obligation / ItemAmount / ValueDue Date / TriggerResponsible Party
    Logo redesign8OO.OO [OCR?]NOT SPECIFIEDNOT SPECIFIED
    Menu layout4OO.OONOT SPECIFIEDNOT SPECIFIED
    Sub total1,2OO.00 [FORMAT?]NOT SPECIFIEDNOT SPECIFIED
    VAT 1O%12O.OO [OCR?]NOT SPECIFIEDNOT SPECIFIED
    Total due1,23O.OO [OCR?]Net 3O daysNOT SPECIFIED
    Late fee1.5% per monthOverdue amountsNOT SPECIFIED

    FLAGS: NONE

    Safety checks (run by the scripts)

    OCR NOISE LIKELY: 9 suspicious figure(s) in the source: 2O26-0147 2O26 8OO.OO 4OO.OO 1,2OO.00 1O% 12O.OO 1,23O.OO 3O
    Figures checked: 10. Not found in the source document: 0.

    Next steps

    • The model wrote CLEAN and FLAGS: NONE, but the scan found 9 figures with the letter O in place of a zero. Treat the model's quality rating as a hint, not a verdict.
    • The invoice lines do not add up to the stated total: 1,200.00 + 120.00 is 1,320.00 if the O's are zeros, but the invoice says 1,230.00. The model did not flag this. Check the original before paying.
    • Time on our CPU-only test laptop: 387.8 s.

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

    About this skill

    Many freelancers and small businesses can't send their invoices and contracts to an online service. This skill runs a local model through Ollama on your own computer and writes a short audit: document type, parties, dates, a table of amounts and deadlines, and a flags line.

    After the model finishes, two scripts check its work:

    • The OCR scan lists numbers in your document that mix the letters O, o, l or I with digits, a common sign of bad scans.
    • The figure check lists any number in the audit that is not written in your document, so an invented or recalculated amount does not slip through.

    Why the scripts are there: the model alone is not good at rating text quality or raising flags. On our noisy test invoice it wrote "CLEAN" while the scan found 9 suspicious figures. On a test lease it wrote an amount that is not in the document, and the figure check caught it.

    What we measured (Windows 10, Intel i3-8130U, 12 GB RAM, no dedicated GPU, llama3.1:8b, QUICK mode, six synthetic test files, one run each):

    • 14 of 21 content checks pass
    • 44 to 388 seconds per document, and it varies a lot between runs

    What is inside the ZIP: SKILL-LITE.md (the prompt for small local models), SKILL.md (full version, not tested on small models), audit.ps1, ocr-scan.ps1, verify-figures.ps1, a Modelfile, six synthetic test documents with our saved outputs, and KNOWN-LIMITS.md.

    Please read the limits: QUICK mode gives a snapshot and a money table, not a risk rating. The model's flags are best-effort. It is a first-pass helper, not legal advice. Compare the result with your original document.

    • Tested only with llama3.1:8b on one Windows 10 CPU-only laptop (Intel i3, 12 GB RAM).
    • 44 to 388 seconds per document; speed varies between runs.
    • The model's flags are best-effort (some missed, one false alarm in our tests). The scripts check the figures.
    • QUICK mode gives no risk rating. FULL mode is not tested across our test files.
    • The OCR scan only finds the letters O, o, l and I inside numbers. Input must be plain text.
    • First-pass helper, not legal advice. Full list: KNOWN-LIMITS.md in the ZIP.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Requires

    Windows with PowerShell
    Ollama
    llama3.1:8b model (one-time download)
    Your document as a .txt file

    Permissions

    Read Files
    Write Files
    Terminal / Shell

    File Scopes

    *.txt
    audits/**

    audit.ps1 only talks to Ollama on this computer (localhost:11434).

    Listed1 day ago

    What's inside

    Frequently Asked Questions

    This skill is specifically optimized for local execution via Ollama. It performs best on models with a context window of at least 8192 tokens to ensure long legal documents are processed without losing critical data.

    Unlike generic prompts, this skill enforces a 'zero inference' policy that prevents the AI from 'fixing' OCR typos or hallucinating currency conversions. This ensures you receive a 100% accurate reflection of the source text rather than a creative interpretation.

    No, this skill is designed for offline use with zero external API calls. Your sensitive legal documents and financial data never leave your local environment, making it ideal for high-privacy audits.

    You receive the specialized auditing logic, the structured extraction schema, and configuration settings for both 'Quick' and 'Full' audit modes. It also includes the formatting rules required to generate clean, copy-pasteable Markdown tables.

    While the skill supports multi-currency tracking, it is optimized for English-language documents. It is currently limited to auditing one document at a time to maintain high precision in risk flagging and data extraction.

    The skill detects and lists 'Missing Protections,' such as the absence of termination rights or liability caps. This helps you identify what is NOT in the contract, which is often more dangerous than what is.

    Only once, to download Ollama and the llama3.1:8b model. After that the audits run on your computer. Our privacy check, with Wi-Fi off, showed Ollama listening on 127.0.0.1 only.

    On our CPU-only test laptop (Intel i3-8130U, 12 GB RAM), 44 to 388 seconds per document, and it varies between runs. A faster computer should be quicker, but we have not tested one.

    Treat them as a hint. On our test files the model missed some flags and raised one false alarm. The two scripts (OCR scan and figure check) are the reliable part. KNOWN-LIMITS.md lists what we found.

    No. It is a first-pass helper. Always compare the result with your original document.

    New in LLM Ops