Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    LLM Usage Cost Reconciler

    1

    Reconcile LLM usage exports against dated prices and billing totals, with source-linked costs and explicit unknowns.

    $19.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    LLM Usage Cost Reconciler

    LLM Usage Cost Reconciler

    Example session with this skill installed

    Use the included synthetic fixtures/usage.json, fixtures/prices.json and fixtures/billing.json. Reconcile the usage against price version synthetic-2026-10-v1, generate report.md and report.json, and explain the token/cache/retry breakdown, adjustments, duplicate and unresolved invoice. Do not estimate missing usage.

    • Read your context and instructions
    • Compiled the llm usage cost

    Synthetic fixture reconciliation

    All records, rates and bills in this demonstration are invented. The bundled CLI ran on the included fixtures and produced report.md and report.json. Price table: synthetic-2026-10-v1. Currency: USD.

    ComponentKnown cost (USD)
    Ordinary input5.20
    Output3.00
    Cache reads0.25
    Cache writes in this fixture0.00
    Calculated API usage8.45
    Supplied API usage subtotal8.50
    Unexplained usage difference0.05
    Explicit retry subset, already included0.60

    Invoice totals and adjustments

    The complete synthetic invoice contains USD8.50 usage, USD0.60 tax, USD-0.25 credit and USD20.00 subscription. Its supplied total is USD28.85, so the invoice arithmetic difference is USD0.00. The subscription is separate from token usage. The USD0.05 usage gap remains unexplained; it is not labelled an overcharge or attributed to guessed rounding.

    Evidence and unknowns

    • usage.json#/records/0 and /records/3 are the same record. It is counted once, with both source references retained.
    • usage.json#/records/1 explicitly links to the earlier request ID. Its USD0.60 is an overlapping subset, never an extra charge.
    • /records/2 uses the later dated price row; the earlier rows use the earlier price.
    • The separate incomplete invoice remains

    unknown. Its record has a missing output quantity and no matching model price. The report shows a partial known lower bound, not zero-filled usage or a conclusive discrepancy.

    • billing.json#/invoices/0 and /invoices/1 identify the supplied totals; price-row references and input-file SHA-256 hashes are retained in report.json.

    Deliverables and next steps

    report.md contains the readable comparison. report.json contains component amounts, pricing version, row-level sources, adjustments, quarantine/duplicate evidence and unknowns. Investigate the USD0.05 with the original source evidence. For the incomplete invoice, obtain the missing usage and applicable dated price and confirm export completeness before drawing a conclusion.

    This is the tested synthetic example shipped with version1.0.0, not a real customer billing finding.

    bundled CLI ran on the included fixtures and produced report.md

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

    About this skill

    Explain the part of the bill your evidence supports

    LLM Usage Cost Reconciler 1.0.0 compares supplied usage exports with explicit versioned price tables and billing totals. Run the bundled local Python tool to produce Markdown and JSON reports with file hashes, row references, known costs and visible unknowns.

    What you get

    • Separate input, output, cache-read and cache-write costs, including explicit cache-write TTLs.
    • Exact decimal arithmetic, dated prices, explicit token units, currencies and invoice rounding.
    • Retry costs only from explicit, unambiguous request links, shown as an overlapping subset.
    • Identical duplicate removal and conflicting-identity quarantine.
    • Credits, taxes, subscriptions and other adjustments kept separate from API usage.
    • A documented canonical CSV/JSON schema and narrow offline OpenAI Chat Completions and Anthropic Messages adapters.
    • Synthetic fixtures, tests, schema documentation and runnable examples.

    How it works

    Provide usage, the applicable historical or contract prices, and billing totals. Declare the provider, model, scope, currency, pricing context, period, quantum and export completeness. Run the local CLI, then review its source-linked report and unresolved evidence.

    No current provider rates are embedded. Missing quantities, cache splits and prices remain unknown. Similar prompts do not prove retries, and a monthly chat subscription is not treated as API token usage. A difference is an unexplained amount, not proof of an overcharge.

    Requirements

    Python 3.10+ and local file/terminal access. Runtime uses only the Python standard library. AI-assisted use needs a host that can access the bundled files and run Python; chat-only hosts cannot execute the toolkit. Delivery follows Agensi's agent connection; this listing does not promise a downloadable ZIP.

    The reconciler makes no account connections, API calls, network requests or billing changes. An AI host's handling of supplied files is governed by that host.

    Boundaries

    No FX conversion, live price discovery, PDF/OCR invoice import, missing-token estimation, streaming merge or automatic billing dispute. Adapters cover only their documented standard-text shapes. Results depend on supplied evidence and declared completeness; invoice rounding may differ. All bundled prices and financial examples are synthetic.

    No current provider prices, currency conversion, missing-usage estimation, PDF/OCR invoice import, streaming event merge, live account integration or billing changes. Requires explicit dated rates and declared export completeness. Provider adapters support only documented standard-text final responses. Unknowns remain unknown. Invoice rounding may differ; hashes identify bytes but do not authenticate an export. No savings, invoice-correctness or accounting-compliance guarantee.

    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 10 days ago

    Requires

    Python3.10+
    Local terminal/shell and file access
    No third-party runtime packages, API keys, account connections or network access
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Read only user-supplied usage, price and billing input paths plus bundled references. Write report.md and report.json to the requested fresh output directory. Runtime is offline and never reads credentials or changes billing. Reports contain supplied identifiers; AI-host file handling is governed by that host.

    Listed10 days ago

    What's inside

    Frequently Asked Questions

    No. The runtime reads files you supply and uses your explicit dated or contract prices. Bundled data and rates are synthetic. It needs no API keys, makes no account connections and never changes billing.

    No. Missing quantities, cache splits and unmatched prices remain unknown. A difference is unexplained evidence, not proof of an overcharge. Conclusive comparisons require declared complete exports and matching billing scope.

    Cache buckets are exclusive under the documented schema. A retry is recognized only through an explicit, unambiguous earlier request identifier. Its cost is already included in usage and is shown as an overlapping subset, never an extra charge.

    The documented canonical CSV/JSON schema is the main interface. Narrow offline adapters support documented standard-text final OpenAI Chat Completions and Anthropic Messages responses. Other exports require verified mapping; streaming, PDF and OCR imports are not supported.

    Python 3.10+ and local terminal/file access. Runtime has no third-party dependencies. Use the CLI or an AI host that can access the bundled scripts and execute Python. Chat-only hosts cannot run it. Host subscriptions or normal host usage charges are separate.

    The local runtime is offline and reads only the supplied input paths. It never contacts billing providers or reads credentials. Reports contain supplied identifiers; any AI host handling your files follows that host’s own data practices.