Data and AI struggles

Codex fails to return valid structured output

LLM apps4 posts from 2 people2d active+33 posts in the last 7 days, 0 the 7 days before (rising)

Posts per day

Posts per day4 posts, Oct 5 to Oct 21

The posts behind it

4, newest first
GitHub issuesOct 15 6 days ago[BUG] Non-function-calling JSON conversion double-encodes structured responses## Description `Converter.to_json()` and `ato_json()` JSON-encode an LLM response even when that response is already JSON text. For models without function calling, `json.loads()` of the converter result is therefore a string rather than the structured object. The task output...crewAIInc/crewAI · WinorenOpen ↗GitHub issuesOct 15 6 days ago[BUG] Non-object user-data JSON crashes tracing status and updates## Description The shared user-data loader casts any successful `json.loads()` result to a dictionary without checking its shape. If `.crewai_user.json` contains valid JSON with a non-object root (for example `[]` or `null`), tracing helpers raise `AttributeError` on `.get()`...crewAIInc/crewAI · WinorenOpen ↗GitHub issuesOct 15 6 days ago[BUG] Escaped definition names break local schema reference resolution## Description Local JSON Schema references with escaped definition names fail in CrewAI's schema normalization. A valid `$ref` such as `#/$defs/a~1b` refers to the definition named `a/b`, but the resolver looks up the encoded name instead. This affects strict tool schema...crewAIInc/crewAI · WinorenOpen ↗GitHub issuesOct 5 2 weeks ago[Bug]: parse_json_markdown loses literal JSON fence markers inside valid payloads### Bug Description `parse_json_markdown` returns `None` for a complete, valid fenced JSON object or array when a JSON string key or value contains the literal marker json json")[1]`, so a marker inside the JSON payload is treated as another delimiter and truncates the selected...run-llama/llama_index · hulkbigOpen ↗

Companies and products named

About this problem

Evidence

4 posts from 2 people in 2 places, about 2 a week over 11 days. Mostly on crewAIInc/crewAI, run-llama/llama_index. Tools named alongside: Raw, LlamaIndex, CrewAI.

Frustration Frustration 0 of 3· Seen on GitHub issues

How it was grouped

Posts that state a pain and match the "structured-output" rule about Codex. First post Oct 5, 2026, latest Oct 15, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method

History

AddedOct 19, 2026Codex fails to return valid structured output (4 posts)