Grow AI Skills Saved
✳ PromptAdapted resource

Data Extraction Tool Brief Prompt

Define a bounded document-to-data tool around one input format, an explicit schema and reviewable errors.

By michael.727·Docs & data

Free prompt · No skill to install

What this prompt does

  • Bounded input

    Choose one document format and state unsupported cases.

  • Schema contract

    Define field types, missing values and source traces.

  • Acceptance cases

    Specify representative checks before implementation.

What to provide & what you get

Input
A sample document, target fields, expected volume and runtime or privacy constraints.
Expected output
A focused extraction-tool brief with a data contract, sample output, validation cases and an implementation plan.

How to use

  1. Supply a representative input and the exact fields you need.
  2. Review the proposed schema and uncertain sample values.
  3. Use the acceptance cases to assess an implementation on real documents.

Check the result: Every sample field should be traceable to the input, and every unsupported case should have an explicit outcome.

Copy. Paste. Run.

v1.0.0-gas1

Edit the inputs below, then copy into your AI tool with the requested source material.

Before you start

  • A sample document, target fields, expected volume and runtime or privacy constraints.
Edits stay in this tab

Worked example · Editorial fixture

Specify a text-invoice extractor with review cases

Original fictional inputs and a prepared editorial answer. This reusable prompt was not run against the fixture. Use the files and case-specific checks to review your own result.

Example input

Input format: UTF-8 text, not scanned PDF. Fictional sample: Invoice INV-18; Supplier Example Paper; Total USD 42.00; Tax ID absent. Volume 20/day. Local-only pilot; no external service approved. Target: invoice_id string, supplier string, total_minor integer, currency enum USD, tax_id string|null; all values need source spans.

Prepared output

SAMPLE OUTPUT
{"invoice_id":"INV-18","supplier":"Example Paper","total_minor":4200,"currency":"USD","tax_id":null,"status":"NEEDS_REVIEW"}
Spans: invoice_id → "Invoice INV-18"; total → "Total USD 42.00"; tax_id → absent.
PIPELINE: read UTF-8 → parse this bounded format → validate schema and currency → attach spans → human review → export JSON. No OCR or model service is needed for this initial format.
FIVE CASES
1. USD 42.00 → 4200.
2. Missing tax ID → null + review.
3. Total USD 4x.00 → invalid amount + review, no numeric guess.
4. Total 42.00 without currency → review, no assumed USD.
5. Two conflicting totals → review, preserve both spans.
IMPLEMENT: fixture parser, schema checks, review report, then evaluate on approved real samples.

What to check in this case

  • The sample preserves the missing tax ID as null; it invents neither a tax identifier nor an exchange conversion.
  • Five cases cover value, missing field, malformed value, missing currency and conflict. These are an implementation contract, not measured extraction accuracy.

Keep in mind

  • This produces a tool brief, not a functioning extractor.
  • Accuracy, OCR coverage, service costs and privacy behavior need implementation-specific verification.

Keep exploring

Related skills & prompts

Browse this category ↗