Delegating to a coworker fails when the model emits a Python-repr list: coworker="['researcher']"

Description

The delegation tools tolerate a coworker that a weak model emitted as a list instead of a string — _get_coworker() unwraps [...] and takes the first entry. But the unwrap keeps the entry's quotes, and role matching only strips double quotes, so the Python-repr spelling (['researcher'], what str(list) and most small models produce) never matches, while the JSON spelling (["researcher"]) does.

The failure is self-contradicting: the tool refuses the role and then prints that same role as an available option.

coworker="['Senior Researcher']"
-> Error executing tool. coworker mentioned not found, it must be one of the following options:
   - senior researcher
   - analyst

Steps to Reproduce

from unittest.mock import patch
from crewai.agent import Agent
from crewai.tools.agent_tools.agent_tools import AgentTools

senior = Agent(role="Senior Researcher", goal="g", backstory="b", allow_delegation=False)
analyst = Agent(role="Analyst", goal="g", backstory="b", allow_delegation=False)
delegate = AgentTools(agents=[senior, analyst]).tools()[0]

with patch.object(Agent, "execute_task", lambda self, task, context=None, tools=None: self.role):
    for coworker in ["Senior Researcher", '["Senior Researcher"]', "['Senior Researcher']"]:
        print(repr(coworker), "->", delegate.run(coworker=coworker, task="t", context="c")[:60])

Actual behavior (measured on main @ 3831e8b)

coworker as the model emits it	value used for role matching	outcome
Senior Researcher	'senior researcher'	delegates
[Senior Researcher]	'senior researcher'	delegates
["Senior Researcher"]	'senior researcher'	delegates
['Senior Researcher']	"'senior researcher'"	"coworker mentioned not found"
['Senior Researcher', 'Analyst']	"'senior researcher'"	"coworker mentioned not found"
'Senior Researcher' (quoted scalar)	"'senior researcher'"	"coworker mentioned not found"

Expected behavior

All six spellings resolve to the Senior Researcher agent. A quoted name is the same defect as a padded name, and a padded name is already tolerated.

Root cause

_get_coworker() (tools/agent_tools/base_agent_tools.py:37) hands the unwrapped entry straight to matching, and sanitize_agent_name() (:20) normalizes whitespace, case, and " only:

normalized = " ".join(name.split())
return normalized.replace('"', "").casefold()

Its own docstring says the value comes back with "quotes removed", and the comment at :65 explains why the tolerance exists at all ("less-powerful LLM's have difficulty producing valid JSON"). Single quotes are the other half of that JSON-shaped problem and are the natural output of str(list) / a Python-repr argument, so the branch misses its main case.

AskQuestionTool shares _get_coworker(), so asking a coworker a question fails the same way.

Possible Solution

Strip the wrapping quotes where the malformed value is already being normalized, in _get_coworker(), so both the list and the scalar form are handled in one place and the agents' own role names are never touched:

if coworker:
    is_list = coworker.startswith("[") and coworker.endswith("]")
    if is_list:
        coworker = coworker[1:-1].split(",")[0]
    coworker = coworker.strip().strip("'\"")

I would not widen sanitize_agent_name() to drop ' instead: it is applied to the agents' real role names as well, so an apostrophe in a role (Data Steward's Office) would stop matching. _get_coworker() only ever sees a model-produced argument.

Additional context

Nothing under lib/crewai/tests/ asserts the current single-quote behavior; the two existing list tests use the unquoted coworker="[researcher]" form, which is why the gap is invisible.

Authored with an AI coding assistant. .github/CONTRIBUTING.md requires the llm-generated label for agent-authored contributions and external contributors cannot apply labels here — maintainers, please add it.

Operating System

Windows 11 (nothing OS specific)

Python Version

3.13

crewAI Version

1.15.22 (main @ 3831e8b)

crewAI Tools Version

Not involved — the defect is in the crewai package.

Virtual Environment

Venv

Start in tools/agent_tools/base_agent_tools.py, especially _get_coworker() and the shared AskQuestionTool path, then review the existing delegation list tests under lib/crewai/tests. Add regression coverage for Python-repr list and quoted scalar coworker values, and confirm all six spellings resolve to the Senior Researcher agent without breaking role names containing apostrophes.