Comment by otsaloma

No, it's been working without problems so far. I'm using Anthropic for what it's worth. I ask the LLM to first do some reasoning and then return a JSON array on the final line. Sometimes I've seen some Markdown backticks there, but no irregularities more than that.