Playground eval history: clicking failed eval result crashes with 'e is not iterable' in formatToolUses

[freddypatota]

Describe the bug

When viewing eval results in the ADK playground web UI, clicking on a failed eval case result (to view the session details) crashes with a JavaScript error. Clicking on passed eval results works correctly.

Error from browser console:

Uncaught TypeError: e is not iterable
    at t.formatToolUses (main-QQBY56NS.js:4105:21033)
    at t.addEvalFieldsToBotEvent (main-QQBY56NS.js:4105:21654)
    at t.addEvalCaseResultToEvents (main-QQBY56NS.js:4105:21447)

Root cause

In the eval tab component, addEvalFieldsToBotEvent calls formatToolUses when a metric has evalStatus === 2 (fail) and the metric name is tool_trajectory_avg_score:

// addEvalFieldsToBotEvent calls:
e.actualInvocationToolUses = this.formatToolUses(A.actualInvocation.intermediateData.toolUses)
e.expectedInvocationToolUses = this.formatToolUses(A.expectedInvocation.intermediateData.toolUses)

formatToolUses iterates its argument with for...of but does not guard against null/undefined:

formatToolUses(e) {
    let A = [];
    for (let i of e)  // crashes if e is null/undefined
        A.push({name: i.name, args: i.args});
    return A;
}

When eval cases don't include tool_uses in their intermediate_data (or when intermediateData.toolUses is serialized as null), expectedInvocation.intermediateData.toolUses is null, causing the crash.

This code path is only reached for failed evals because the tool_trajectory_avg_score status check (evalStatus === 2) only triggers for failures. Passed evals never enter this branch.

Steps to reproduce

1. Create a minimal agent that uses AgentTool:

from google.adk.agents import Agent, SequentialAgent
from google.adk.tools.agent_tool import AgentTool

sub_agent = Agent(name="sub", model="gemini-2.5-flash", instruction="Say hello")
workflow = SequentialAgent(name="workflow", sub_agents=[sub_agent])
root_agent = Agent(
    name="my_agent",
    model="gemini-2.5-flash",
    instruction="Use workflow tool to respond.",
    tools=[AgentTool(agent=workflow)],
)

2. Create an evalset file (my_eval.evalset.json) without expected tool_uses:

{
  "eval_set_id": "my_eval",
  "eval_cases": [
    {
      "eval_id": "simple_test",
      "conversation": [
        {
          "user_content": {
            "parts": [{"text": "Hello"}]
          }
        }
      ],
      "session_input": {
        "app_name": "my_agent",
        "user_id": "test_user",
        "state": {}
      }
    }
  ]
}

3. Start the playground: adk web ./agents/ --port 8501
4. Go to the Eval tab, select my_eval, select the test case, run with tool_trajectory_avg_score metric
5. After the eval completes, go to the History section
6. Click on the result button for a FAIL entry -- the UI crashes with the error above

Expected behavior

Clicking on failed eval results should display the session details, same as for passed results.

Suggested fix

Add a null guard in formatToolUses:

formatToolUses(e) {
    if (!e) return [];
    let A = [];
    for (let i of e) A.push({name: i.name, args: i.args});
    return A;
}

Environment

• google-adk version: 1.25.1
• Python: 3.13
• Browser: Chrome (latest)
• OS: macOS

[llalitkumarrr]

Hello @freddypatota ,

Thank you for reporting this. We tried to reproduce the issue using google-adk==1.26.0 and followed your exact steps. We encountered a 404 Not Found error, even though the evaluation results were available in the eval_history dir created after running the test.

This needs to be investigated further.

Hello @xuanyang15 , could you please have a look at this and share your inputs?

[xuanyang15]

@wyf7107 Could you please help take a look?