microsoft/autogen · error · ArgumentException

The response from admin is {name}, which is either not in th

Error message

The response from admin is {name}, which is either not in the candidates list or not in the correct format.

What it means

Thrown by RolePlayOrchestrator.GetNextSpeakerAsync when the admin's extracted speaker name (after stripping the 5-character 'From ' prefix) does not case-insensitively match any candidate agent name. The exception message embeds the raw parsed name to show what the admin actually said.

Source

Thrown at dotnet/src/AutoGen.Core/Orchestrator/RolePlayOrchestrator.cs:101

                MaxToken = 128,
                StopSequence = [":"],
                Functions = null,
            },
            cancellationToken: cancellationToken);

        var name = response.GetContent() ?? throw new ArgumentException("No name is returned.");

        // remove From
        name = name!.Substring(5);
        var candidate = candidates.FirstOrDefault(x => x.Name!.ToLower() == name.ToLower());

        if (candidate != null)
        {
            return candidate;
        }

        var errorMessage = $"The response from admin is {name}, which is either not in the candidates list or not in the correct format.";
        throw new ArgumentException(errorMessage);
    }

    private IEnumerable<IMessage> ProcessConversationsForRolePlay(IEnumerable<IMessage> messages)
    {
        return messages.Select((x, i) =>
        {
            var msg = @$"From {x.From}:
{x.GetContent()}
<eof_msg>
round # {i}";

            return new TextMessage(Role.User, content: msg);
        });
    }
}

View on GitHub (pinned to 027ecf0a37)

Solutions

Make agent names simple single tokens (no spaces/punctuation) matching what the admin prompt shows
Strengthen the admin's system prompt to reply exactly 'From <agentname>:' and nothing else
Lower Temperature is already 0; also consider raising MaxToken and few-shot examples in the role-play prompt
Catch ArgumentException around GetNextSpeakerAsync and retry the round, or implement a custom IOrchestrator with fuzzy name matching

Example fix

// before: agents named "Code Reviewer Expert" -> admin replies 'From Code Reviewer Expert:' butSubstring may misalign
// after: use single-token names
var alice = new AssistantAgent(name: "alice", ...);
var bob = new AssistantAgent(name: "bob", ...);

Defensive patterns

Strategy: try-catch

Validate before calling

// Sanity-check candidate names are single tokens the LLM can echo exactly
if (candidates.Any(c => c.Name!.Contains(' ') || c.Name!.Length < 2))
    throw new InvalidOperationException("Rename agents to single-token names before role-play orchestration");

Try / catch

for (int attempt = 0; attempt < 3; attempt++)
{
    try { return await orchestrator.GetNextSpeakerAsync(ctx, ct); }
    catch (ArgumentException e) when (e.Message.Contains("not in the candidates list") && attempt < 2)
    { /* temperature-0 retry; LLM format drift is often transient */ }
}

Prevention

Use simple lowercase single-word agent names
Instruct the admin explicitly: reply with 'From <name>:' only
Consider a custom orchestrator that fuzzy-matches admin output to candidate names

When it happens

Trigger: Admin reply format deviates from 'From <name>:' — e.g. 'From alice smith:', a name with trailing punctuation, a hallucinated agent not in the candidates list, or content shorter than 5 characters (note: Substring(5) itself would throw ArgumentOutOfRangeException first for short strings).

Common situations: Prompt drift where the LLM answers with a sentence instead of a name, agent names containing spaces/casing mismatching the prompt example, stop-sequence ':' behaving differently across model providers, or candidates list changing between rounds.

Related errors

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/51a8f79c88650b86. Report an issue: GitHub.