Parser fails on 'Action: None' - LLM wants direct response but uses wrong format

[npkanaka]

When LLMs output Action: None (or variations like Action: N/A), the parser fails with OutputParserError and leaks internal "Thought:" text to users instead of providing a clean response.

Problem

The ReAct format has a gap in handling scenarios where an LLM recognizes it should use a tool but can't/shouldn't.

In these cases, LLMs commonly output:

Thought: [reasoning about why tool can't be used]
Action: None (direct response required)

This doesn't match the expected formats:

• Action: [tool] + Action Input: {...} → for tool calls
• Final Answer: [answer] → for direct responses

The parser raises OutputParserError, and the error handler returns the raw text including "Thought:" prefix, creating poor UX.

Environment

• CrewAI version: 1.5.0+
• LLM: Any ReAct-style LLM
• Python: 3.10+