Pipeline runner silently returns empty dict on JSON decode failure

[mcdonalday]

Pipeline runner silently returns empty dict on JSON decode failure

Severity: Medium
File: backend/app/code/pipeline/pipeline_runner.py:129

When parsing LLM-generated JSON responses, the code attempts to extract JSON from text by finding { and }. If parsing fails, it logs a warning and returns an empty dict. Downstream code receives no signal that parsing failed.

except json.JSONDecodeError:
    pass

# Return empty dict if parsing fails
logger.warning(f"Failed to parse JSON from response: {text[:200]}")
return {}

Why it matters

Pipeline steps that depend on structured LLM output proceed with empty data, causing cascading failures or incorrect code generation. The warning log is easily missed in production.