Copied logs, AI output, hand-edited configuration, and fenced snippets often look like JSON while failing a strict parser. CodePrettify repairs common problems, shows what changed, and lets you continue into a visual transformation recipe.
The repair tool is designed for recognizable JSON-like input. It does not invent missing business data or pretend an ambiguous document is correct.
The payload is wrapped in prose and a Markdown fence, uses JavaScript-style keys and quotes, and contains Python constants.
Captured application payload:
```json
{
service: 'api-gateway',
environment: 'production',
enabled: True,
regions: ['eu-north-1',],
fallback: None,
}
```The result uses quoted keys and strings, JSON booleans and null, and no trailing commas.
{
"service": "api-gateway",
"environment": "production",
"enabled": true,
"regions": [
"eu-north-1"
],
"fallback": null
}The workbench lists the repairs it applied. If the source is incomplete or semantically ambiguous, confirm the values against the originating system.
Use the captured input above for this lesson. The important habit is to treat the repair report as evidence, not as a decorative success message.
Choose More Actions → General tools → JSON Repair & Transform, or find it through the Command Palette. In the Windows app, it is also available from Tools. Select the Repair & Salvage tab.
Paste the complete text, including surrounding prose and code fences, so the report can explain what it removed. If a compatible document is already open, choose Use document. Avoid hand-fixing the source first—you would lose the audit trail for those changes.
Leave Allow partial-record salvage off unless the input is JSON Lines and you deliberately accept dropping reported invalid records. With the option off, one bad record blocks the result. With it on, every excluded source line must be reviewed.
Choose Analyze & repair. For the sample, expect the report to mention the Markdown fence or surrounding text, quoted keys, converted single-quoted strings, Python booleans and null, and removed trailing commas. A change that extracts or semantically replaces content deserves more scrutiny than whitespace cleanup.
Verify that True became JSON true, None became null, and every string value remained unchanged. The tool can normalize recognizable syntax; only you can confirm that production, the region, and the enabled state are correct business values.
If the tool rejects truncated input, missing structural separators, unsafe numbers, excessive nesting, or another unbounded case, return to the producing system. A refusal is safer than a plausible-looking document assembled from guesses.
Copy or download the clean JSON when repair is the only goal. To reshape it, choose Use result in Transform. This creates a separate working result; it does not overwrite the original file or the text you pasted.
Before using a repaired payload in code, a database, or another tool, confirm all six points.
Once the input is valid structured data, add operations in order. Each step receives the result of the previous step, and the live output explains how many rows and steps remain.
role = Admin followed by descending request count returns two rows from six.Keep only records that match a field, comparison operator, and value.
Order matching rows by requests, timestamp, name, or another selected field.
Add another operation, copy the JSON, export it, or open the result as a new document.
Follow the recipe shown in the screenshot: keep administrators, order them by request volume, and retain only the fields needed for a short report. Each operation receives the result of the preceding step.
Choose Use result in Transform after repair, or load valid JSON, JSON Lines, CSV, YAML, TOML, XML, or feed data. Confirm the live result shows the expected starting row count before adding operations.
Choose Filter, select the role field, use the equality operator, and enter Admin. The field picker uses detected fields, types, and sample values to reduce typing mistakes.
Confirm that only administrator rows remain and that the status reports two rows from the original six. If no rows match, inspect capitalization, value type, and field path before adding another step.
Choose Sort, select the request-count field, and use descending order. Because sorting runs after filtering, it orders only the two administrator records instead of doing work on rows that will be removed.
Add Select fields and keep the user name, role, and request count. The output becomes smaller without changing the original source. Rename a field only when the downstream consumer requires a different key.
Move Select fields above Sort. If the selected fields no longer include the sort key, the later step cannot work as intended. Move it back and use this experiment to see why recipe order is part of the transformation.
When the live result and row count are correct, copy, export, or open the transformed JSON as a new document. Keep the source and repair report until the output has been checked by its destination.
Expected result: two administrator objects, sorted from highest to lowest request count, containing only the fields selected for the report.
Select fields, remove fields, or rename properties while keeping a readable step history.
Filter, sort, limit, or remove duplicates from arrays of objects.
Flatten nested objects when downstream tools expect a simpler record shape.
Group rows and calculate aggregates for quick inspection or export.
Move or remove steps and let the workbench recalculate the result from the original input.
The source document remains intact. The transformed output is a separate result you explicitly copy, export, or open.
| What you see | What it means | Best next action |
|---|---|---|
| High-confidence formatting cleanup | The tool made bounded syntax changes such as removing a fence or trailing comma. | Still review the report, then parse or transform the result. |
| Extraction or semantic-literal warning | The tool selected content from a wrapper or changed Python/JavaScript meaning into JSON. | Compare each changed value with the source system. |
| Invalid JSON Lines record | One physical record cannot be parsed safely. | Fix the source, or opt into partial salvage only when losing the reported line is acceptable. |
| Incomplete transformation step | A required field, operator, value, or option is missing. | Complete the step; the previous valid preview remains visible. |
| “No rows matched” | A completed filter removed every row. | Check field path, value type, spelling, and operation order, or remove that step. |
| Unsafe or truncated structure | The tool cannot recover strict JSON without guessing. | Stop and obtain a complete payload from the producer. |
No. It repairs recognizable syntax and wrapper problems. It does not fabricate domain values that are absent from the source.
No. The original remains unchanged until you deliberately copy, export, or open the repaired result.
The available operations depend on the input shape. Row-oriented operations are most useful for arrays of objects; object and nesting operations apply where their required shape exists.
The workbench exposes the ordered steps in the active session. For repeatable automation outside the UI, copy the resulting data or translate the visible recipe into application code.
Paste the JSON-like text, inspect the repair report, and continue into a visible transformation workflow on your device.