[BUG] `human_input=True`: the feedback prompt references a "Final Result above" that is never displayed unless `verbose=True`

[coolhandle01]

Description

When a Task has human_input=True but the agent/crew is not verbose, CrewAI shows the 💬 Human Feedback Required panel — "Provide feedback on the Final Result above" — but the result is never printed. The operator is asked to approve output they cannot see.

The two behaviours are gated independently:

• the result render (_show_logs → AgentLogsExecutionEvent) is gated on verbose=agent.verbose or crew.verbose;
• the feedback gate fires whenever human_input=True.

So human_input=True + non-verbose ⇒ a prompt that explicitly references content ("the Final Result above") that was never rendered.

Relationship to #6065

This is separate from #6065 (the AttributeError: 'AgentExecutor' object has no attribute 'ask_for_human_input' crash). On 1.14.5 / 1.14.6 that crash fires first and masks this. The bug here is observable on 1.14.4 (where human_input still works), and will resurface on 1.14.5+ once #6065 / #6069 lands — so it's worth fixing alongside it.

Steps to Reproduce

A minimal reproduction follows.

from crewai import Agent, Crew, Task, Process
from crewai.llms.base_llm import BaseLLM

class StubLLM(BaseLLM):
    def call(self, messages, tools=None, callbacks=None, available_functions=None,
             from_task=None, from_agent=None, response_model=None):
        return "Thought: I know it.\nFinal Answer: The sky is blue."
    def supports_function_calling(self) -> bool:
        return False

agent = Agent(role="Tester", goal="Be a Minimal Reproduction", backstory="bg",
              llm=StubLLM(model="stub"), verbose=False)
task = Task(description="Sky colour?", expected_output="a colour",
            agent=agent, human_input=True)
Crew(agents=[agent], tasks=[task], process=Process.sequential, verbose=False).kickoff()

Press enter at the prompt.

Expected behavior

When human_input is requested, the result under review is displayed (or embedded in the feedback panel) regardless of verbose — the prompt should never reference output that wasn't shown.

Screenshots/Code snippets

N/A (minimal reproduction above)

Operating System

Ubuntu 24.04

Python Version

3.12

crewAI Version

1.14.4

crewAI Tools Version

1.14.4

Virtual Environment

Venv

Evidence

• verbose=False: only the feedback panel appears; no result anywhere.
• verbose=True: the ✅ Agent Final Answer panel renders, and the feedback panel includes a Final Output: line — confirming the result display is purely verbose-gated.

Possible Solution

Root cause

crew_agent_executor.py::_show_logs emits AgentLogsExecutionEvent(verbose=agent.verbose or crew.verbose), so the result render is verbose-gated, while the human-feedback gate (core/providers/human_input.py::_prompt_input) fires on human_input alone and prints a prompt that assumes the result is "above".

Suggested fix

When human_input is active, surface the AgentFinish output regardless of verbose — either emit the result event unconditionally in that path, or include the output directly in the feedback panel so "the Final Result above" is always actually present.

Additional context

Bug diagnosis performed and issue description Generated by Claude Code

[coolhandle01]

⚠️ this issue was generated with the assistance of an LLM, but I am unable to apply that label to this issue

[Gul-A05]

Hi, I'm a newer contributor and I'd love to take a crack at this! Could you assign it to me

[coolhandle01]

Hi stranger!

I think a bunch of us set our Claude's on fixing this, so there are several PRs already. None of us checked the existing issues first it seems. I feel for the maintainers.

[coolhandle01]

@joaomdmoura this seems to be a serious regression, which is why everybody's set their Claude upon it. None of us external would-be contributors have the ability to set the LLM-Assisted label on our PRs or issues.

This would seem to be a very serious regression - what people am I supposed to recommend your library to if human in the loop is fundamentally broken. It's been two weeks. What's going on?

I appreciate that #6075 is probably terribly misinformed, because an AI did it, but it at least includes regression tests to catch this if it happens again.

Please help us to help you. My project is about bug bounty, because I'm trying to teach myself how to make agents safely.. There's no way I can safely develop it without HITL working, though. I'm blocked.

TL;DR: please could a human review #6075 - and/or this issue. Maybe it's stale now and the problem's solved.

<3