jsonquotefixer 0.3.0

Fixer for broken/unescaped quotes in pseudo-JSON text

jsonfixer

A small utility to repair broken / pseudo-JSON strings that AI models often produce, so they can be parsed by Python's built-in json module.

Motivation

LLM outputs frequently contain JSON that looks correct but is actually invalid:

• Unescaped inner double quotes inside string values
• Raw control characters (literal newlines, tabs, etc.) inside strings
• Smart/curly quotes (“ ” ‘ ’) instead of straight quotes
• The whole JSON wrapped in Markdown triple-backtick code fences
• The whole JSON returned as a quoted string with escaped \n sequences

For example, the following is not valid JSON and json.loads will fail:

{
    "key1": "value1",
    "key2": "value2 is behind the "value1", that is good"
}

jsonfixer detects and repairs these common patterns so you can reliably call json.loads() on the result.

Features

• Escapes stray inner double quotes inside string values (" → \")
• Converts literal control characters inside strings to proper JSON escapes (\n, \t, \r, \b, \f, and \u00XX for other < 0x20 code points)
• Normalizes CRLF (\r\n) inside strings to \n
• Optional extraction of JSON from Markdown code fences (```json ... ```)
• Optional replacement of smart quotes (“ ” ‘ ’) with straight quotes (" ')
• Leaves already-valid JSON unchanged
• Pure Python, no runtime dependencies

Installation

pip install jsonquotefixer

Usage

Basic example

import json
from jsonfixer import fix_quotes

broken = '{"title": "The "Great" Gatsby", "author": "F. Scott Fitzgerald"}'

json.loads(broken)          # raises json.JSONDecodeError

fixed = fix_quotes(broken)
json.loads(fixed)
# {'title': 'The "Great" Gatsby', 'author': 'F. Scott Fitzgerald'}

Extracting from a Markdown code block

Pass parse_code=True when the model wraps the JSON in triple backticks, or returns it as a JSON-encoded string with escaped newlines:

from jsonfixer import fix_quotes

raw = '```json\n{"a": "hello"}\n```'
fix_quotes(raw, parse_code=True)
# '{"a": "hello"}'

Replacing smart quotes

Pass replace_smart=True to convert curly quotes to straight quotes before the fix-up step runs:

from jsonfixer import fix_quotes

raw = '{“a”: “hello”}'
fix_quotes(raw, replace_smart=True)
# '{"a": "hello"}'

All options combined

fix_quotes(raw, parse_code=True, replace_smart=True)

API

fix_quotes(s, parse_code=False, replace_smart=False) -> str

Main entry point. Returns a repaired JSON string.

Argument	Type	Default	Description
s	str	—	The broken / pseudo-JSON string to repair.
parse_code	bool	False	Extract JSON from Markdown code fences, and unwrap outer JSON string quotes.
replace_smart	bool	False	Replace smart quotes (“ ” ‘ ’) with straight quotes (" ') before fixing.

parse_codeblock(s) -> str | None

Extracts the content of the first triple-backtick code fence in s (with or without a language tag). Also handles the case where newlines appear as the literal two-character sequence \n. Returns None if no code fence is found.

replace_smart_quotes(s) -> str

Returns s with smart quotes replaced by their straight-quote equivalents. Does not perform any other repair.

Development

Setting up

git clone https://github.com/abzb1/jsonfixer.git
cd jsonfixer
pip install -e '.[dev]'

Running the tests

Tests are written with pytest:

pytest

Or, to run a specific test and see verbose output:

pytest tests/test_fix_quotes.py -v
pytest tests/test_fix_quotes.py::test_valid_json_unchanged -v

Pre-commit hooks

The repository ships with a .pre-commit-config.yaml. Enable it with:

pre-commit install
pre-commit run --all-files

License

MIT — see LICENSE.