CSV to JSON for automation: preserve data before connecting tools

Keep the meaning of each field

An account code like 00123 should not become 123 just because a conversion tool guessed its type. An empty cell should not quietly become zero either. Preserve the export first, then decide which fields are numbers, dates or booleans. That is the contract your next integration needs.

RFC 4180 documents common CSV conventions including quoted fields, embedded commas and doubled quotes. It is an informational description of a common format; real exports may use different delimiters and newline styles. Format reference.

Why splitting on commas fails

The row "Shah, Fazal",draft contains two fields, not three. A quoted field can also contain a newline. An escaped double quote appears as two consecutive quotes inside a quoted field. A line-by-line split followed by a comma split loses these boundaries.

Use a parser that handles those cases and reports malformed input. A duplicate header should produce an explicit error rather than silently replacing a value. Rows with too few or too many fields should stop conversion so you can fix the export.

Use the converter on an evaluation dataset

Prepare columns such as task_id, expected_action and requires_approval. Convert the records, inspect the JSON and validate every field against your workflow schema. Keep task_id stable so you can compare results across versions.

The CSV to JSON tool runs in your browser, supports comma, semicolon and tab delimiters, and preserves all values as strings. It accepts up to 250,000 characters. It skips fully empty records and rejects duplicate headers, inconsistent row widths and malformed quotes. Those are implementation choices, not a claim to cover every spreadsheet dialect.

Test before connecting a write operation

Try a quoted comma, an embedded newline, an escaped quote and a leading-zero identifier. Add a malformed record and confirm that conversion fails clearly. Test your destination’s schema separately: valid JSON does not imply valid business data.

Do not let imported fields become executable instructions. Treat every cell as data, keep HTML escaped when displaying it and use the destination API’s structured arguments. An evaluation input that contains an instruction to change permissions should not change the executor’s policy.

Next action

Convert a sample using the browser tool, then build an agent evaluation set. When replacing a production data path, run a dry import and compare record counts and identifiers before enabling writes.

Keep a conversion acceptance fixture

Before adopting a parser in a workflow, save a tiny fixture with the cases your exports actually contain. Include a name with a comma, text with a newline, a quoted phrase, an empty optional field and an identifier with leading zeros. Write the expected object for each row and compare it with the produced JSON.

Add a duplicate header and a row with a missing field as deliberate failures. Confirm the tool rejects them visibly instead of producing a plausible but incomplete object. A conversion that silently overwrites one column can be more damaging than one that stops with a clear error.

Keep this fixture under version control with the integration. If an exporting system changes its delimiter, header names or date format, update the contract deliberately and rerun the fixture. Record the original export version and conversion settings in an import log. This creates a repeatable boundary between a spreadsheet and an application rather than relying on a one-time visual inspection.

Sources and review notes

RFC 4180 checked October 14, 2026. Data-contract advice and examples are original guidance. Tool behavior is verified against this repository’s parser tests; no customer dataset is used.