Data File Analyst: Answers With the Query Shown

    1

    Ask questions about your CSV, JSON, Excel or Parquet files and get numbers that come only from executed, shown queries.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Data File Analyst: Answers With the Query Shown

    Example session with this skill installed

    Here's our Q3 orders export and our customer list. Which customer tier brought in the most revenue, how much came from the North District, and what's a typical order size in units?

    • Read your context and instructions
    • Compiled the data file analyst

    Plan: profile both files, resolve every spelling of "North District", check the orders-to-customers join, then one query per figure.

    What the profile found

    • orders: one row per order_id, 120 rows, 2026-07-01 to 2026-09-30. amount is text ($490.00), so sums use num(). qty has a

    possible placeholder -999 in 5 rows.

    • customers: 44 rows, 40 customer ids, no key: tier varies within customer_id.

    The join check flagged double counting

    "relationship": "many-to-many", "inner_join_rows": 137
    a orders row can meet up to 2 customers rows ... (137 joined rows from 120 matched orders rows)
    

    C004, C017 and C031 are listed as Basic and Enterprise; C023 is an exact duplicate. A plain join totals

    47,301.50 against a true

    41,772.50, overcounting by 5,529.00. I kept one row per customer, taking the higher tier as current (tell me if that's wrong):

    TierRevenueOrdersCustomers
    Basic19,380.505318
    Enterprise13,459.004112
    Pro8,933.00268
    WITH c1 AS (SELECT customer_id, substr(MAX(CASE tier WHEN 'Enterprise' THEN '3Enterprise'
      WHEN 'Pro' THEN '2Pro' ELSE '1Basic' END), 2) AS tier FROM customers GROUP BY customer_id)
    SELECT tier, SUM(num(amount)), COUNT(*), COUNT(DISTINCT customer_id)
    FROM orders JOIN c1 USING (customer_id) GROUP BY tier ORDER BY 2 DESC  -- row_count 3
    

    North District

    values found three spellings: North District (37 rows), N District (11), North Dist. (5). It also offered South District as a near spelling; I left it out. Together:

    20,901.50 from 53 orders, 50.0% of revenue. The exact name alone gives 14,577.50.

    SELECT SUM(num(amount)), COUNT(*),
      ROUND(100.0 * SUM(num(amount)) / (SELECT SUM(num(amount)) FROM orders), 1)
    FROM orders WHERE region IN ('North District','N District','North Dist.')  -- row_count 1
    

    Order size

    With -999 as missing: median 6 units, mean 6.43, over 115 of 120 orders. Left in, the mean reads -35.47.

    SELECT median(NULLIF(qty,-999)), AVG(NULLIF(qty,-999)), COUNT(NULLIF(qty,-999))
    FROM orders  -- row_count 1
    

    Next steps

    • Confirm the current tier for C004, C017 and C031.
    • Confirm -999 means "not recorded".
    • Want revenue per customer by tier, or by month? Ask.

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

    About this skill

    Point your agent at your own data files. It profiles every file first (types, nulls, placeholder values such as -999, numbers stored as text, ambiguous dates, duplicates, grain and keys), plans before querying, resolves every spelling of a name before filtering on it, and measures joins for double counting before trusting a total. Every answer comes with the SQL and row count behind each figure, so you can check it. Works with CSV, TSV, JSON and NDJSON out of the box, and Excel .xlsx and Parquet when the needed package is installed.

    Everything loads into memory: about 1 GB of RAM per million rows of 6 columns, measured on a laptop. SQLite dialect (no DATE_TRUNC or PERCENTILE_CONT). No significance tests, forecasting or regression. Excel .xlsx and Parquet need openpyxl or pyarrow, which the skill asks before installing. Compression: .gz only.

    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 9 days ago

    • Free with an Agensi account

    Requires

    Python 3
    openpyxl (optional, for .xlsx)
    pyarrow (optional, for Parquet)
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Runs the bundled script locally on the files the user names. It loads them into an in-memory SQLite database for read-only queries, writes nothing unless asked, and makes no network requests.

    Listed9 days ago

    What's inside

    Frequently Asked Questions

    Unlike standard LLM prompts that 'guess' at data summaries, this skill executes local Python and SQLite code to generate figures. It includes a specific validation layer that flags duplicate rows and inconsistent naming, ensuring your numbers are verifiable rather than hallucinated.

    The skill supports CSV, JSON, NDJSON, and TSV files natively, including compressed .gz versions. With the 'openpyxl' and 'pyarrow' libraries installed, it can also process Excel (.xlsx) and Parquet files.

    Since the skill processes data in an in-memory SQLite database, your files must be small enough to fit into your local system's RAM. It is optimized for efficiency but is not intended for multi-terabyte 'Big Data' environments.

    Python 3 on your machine. The bundled scripts/datafile.py does the profiling, value search, join checks and SQL queries; CSV, TSV, JSON and NDJSON need only the standard library. Your data is loaded into an in-memory SQLite database and stays on your machine.

    The agent performs a 'Join Validation' step before running queries, which checks for many-to-many relationships and potential 'fan-out' issues. This prevents the common error of double-counting values when merging two different datasets.

    SKILL.md (the agent's rules and step-by-step workflow), scripts/datafile.py (profile, values, join-check and query commands, Python standard library only) and reference.md (SQL recipes and pitfalls: whole periods, per-row vs per-entity totals, joins without double counting, rates with their base, top N per group, cohort retention and reconciling parts with the whole).

    New in Data & Databases