Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Verifier

    2

    Independent arithmetic audit for finished tables: catch totals that do not add up before a human does.

    Free

    1 useSecurity scanned
    agent-verifier

    agent-verifier

    Example session with this skill installed

    | Item | Amount || --- | --- || Item A | 100 || Item B | 200 || Total | 310 |

    • Read your context and instructions
    • Compiled the agent-verifier

    Verdict: MISMATCH — total row: claimed 310, recomputed 300, difference -10, plus a signed receipt.

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

    About this skill

    Agent Verifier is an independent arithmetic check for AI agents, packaged as an installable skill. Before an agent delivers a table, spreadsheet, CSV, cost breakdown, invoice, quotation or budget to a human, this skill has the agent submit the finished numbers to the Agent Verifier service. The service recomputes every row total, column total and subtotal two independent ways, lists each mismatched cell with the recomputed value and the difference, and returns a signed verification receipt (verification ID, UTC time, SHA-256 content hash).

    The skill covers the full loop: when to check, how to call (MCP tool, bundled script, or plain HTTP), how to read MATCH / MISMATCH / UNABLE_TO_VERIFY, and how to fix and re-verify. It checks arithmetic consistency only — it does not verify source data, and it says so on every receipt.

    Privacy: zero retention. Submissions are processed in memory only by the service and are never stored or logged. The skill makes one disclosed network call: the table text to agent-verifier.agent-verifier.workers.dev.

    Underlying service pricing: 20 checks per day per caller free; beyond that US$0.02 per check via x402, currently on the Base Sepolia test network (test funds only).

    Arithmetic consistency only; source data accuracy is not verified. Input limit 5,000 cells / 64 KB per check. Does not evaluate expressions or percentages. Requires network access to agent-verifier.agent-verifier.workers.dev.

    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

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    • Free with an Agensi account

    Requires

    Workers

    Permissions

    Terminal / Shell
    Network Access

    Allowed Hosts

    agent-verifier.agent-verifier.workers.dev
    Listed15 days ago
    Updated3 days ago

    What's inside

    Frequently Asked Questions

    Standard LLMs often suffer from 'calculator hallucination,' where they confidently confirm incorrect totals. This skill bypasses the model's internal logic, using a dedicated execution environment to perform exact recomputation and provide a cryptographic receipt for the results.

    The skill is compatible with any AI agent or platform that supports Model Context Protocol (MCP) tool calling, as well as any environment capable of making standard HTTP POST requests. This includes popular frameworks like Claude Desktop, LangChain, and custom Python implementations.

    You receive access to a remote MCP server hosted on Cloudflare Workers, a Python-based CLI utility for local auditing, and documentation for integrating the `verify_numbers` tool into your existing workflows.

    Currently, the skill supports tables up to 5,000 cells or 64 KB per request. While it is highly accurate for static numbers, it is designed for finished totals and does not currently evaluate dynamic percentages or complex mixed expressions.

    A 'Verification Receipt' is a signed document containing a unique ID, UTC timestamp, and a SHA-256 content hash of your table. It serves as an immutable audit trail proving that a specific version of your data was mathematically verified at a specific time.

    Simply provide your table data via the `verify_numbers` tool; the skill automatically parses Markdown, CSV, or TSV formats. If any rows or columns don't sum correctly, you'll receive a detailed report showing the exact difference between the claimed and actual values.

    Row totals, column totals, subtotals and line items against stated totals in tables, spreadsheets, CSV/TSV, invoices, quotations and budgets.

    No. Zero retention: content is processed in memory only, never stored or logged.

    20 checks/day free per caller; then US$0.02 per check via x402 (currently test network, test funds only). The skill itself is free.

    New in Productivity