Malformed JSON Repair

The problem, from first principles

Even when asked for JSON, models produce things strict parsers reject: the answer wrapped in a markdown code fence, a friendly sentence before it, a trailing comma after the last item, Python-style literals (True, None) instead of true and null. Rather than discarding the response and paying for a retry, an agent can apply a short sequence of conservative repairs and parse again. The repairs must be careful not to change meaning: a comma inside a string value is data, not a trailing comma. This question implements the common fixes in a safe order and gives up cleanly when the text still does not parse.

From theory to code

Implement repair_json.

Constraints

• repair_json(text) returns the parsed Python object, or raises ValueError if no repair makes it valid JSON. Apply these steps in order, parsing with json.loads after each attempt and returning at the first success (try the raw text first).
• 1. Strip a surrounding code fence (a line starting with three backticks, with an optional language name, and the closing fence). 2. Take the substring from the first { or [ to the last matching closing } or ] (the outermost span, ignoring leading and trailing prose). 3. Remove trailing commas: a comma followed only by whitespace and then } or ], outside string literals. 4. Replace the bare words True, False and None (outside string literals, whole words) with true, false and null.
• Steps accumulate: step 3 is applied to the output of step 2, and so on, attempting a parse after each.

Hints