fix: filter thinking parts from agenttool result

[dmora]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Closes: agenttool: Run() leaks thinking parts into tool result #697
• Related: fix: interleaved thought&text aggregation in remoteagent #655, session.Events() and separating thinking & final response #313

2. Or, if no issue exists, describe the change:

Problem:
agenttool.Run() collects all Part.Text from the sub-agent's final response without checking Part.Thought. When thinking models (Gemini 2.5/3 with IncludeThoughts: true) are used as sub-agents, their internal reasoning tokens leak into the tool result map[string]any that the calling agent sees.

This is the same class of bug fixed for remoteagent in #655.

Compare agenttool.Run() (before this fix):

for _, part := range lastContent.Parts {
    if part != nil && part.Text != "" {
        textParts = append(textParts, part.Text)
    }
}

With genai.GenerateContentResponse.Text() which correctly filters thoughts:

if part.Text != "" {
    if part.Thought {
        continue
    }
    texts = append(texts, part.Text)
}

Solution:
Add && !part.Thought to the text extraction loop in agenttool.Run(), matching the genai.Text() behavior.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

New tests added:

• TestAgentTool_Run_FiltersThoughtParts — mixed thought + text parts, with edge case {Text: "", Thought: true}, verifies only non-thought text is returned
• TestAgentTool_Run_AllThoughtPartsReturnsEmpty — all-thought response returns empty map
• TestAgentTool_Run_FiltersThoughtParts_WithOutputSchema — thought parts mixed with JSON output, verifies clean unmarshalling when OutputSchema is set

$ go test ./tool/agenttool/ -v
--- PASS: TestAgentTool_Declaration (0.00s)
--- PASS: TestAgentTool_DeclarationWithoutSchema (0.00s)
--- PASS: TestAgentTool_Run_InputValidation (0.00s)
--- PASS: TestAgentTool_Run_OutputValidation (0.00s)
--- PASS: TestAgentTool_Run_Successful (0.00s)
--- PASS: TestAgentTool_Run_WithoutSchema (0.00s)
--- PASS: TestAgentTool_Run_FiltersThoughtParts (0.00s)
--- PASS: TestAgentTool_Run_AllThoughtPartsReturnsEmpty (0.00s)
--- PASS: TestAgentTool_Run_FiltersThoughtParts_WithOutputSchema (0.00s)
--- PASS: TestAgentTool_Run_EmptyModelResponse (0.00s)
--- PASS: TestAgentTool_Run_SkipSummarization (0.00s)
PASS
ok  google.golang.org/adk/tool/agenttool  0.278s

Manual End-to-End (E2E) Tests:

Verified locally with a real Gemini 2.5 Flash model with ThinkingConfig{IncludeThoughts: true, ThinkingBudget: 2048}. The sub-agent was asked "What is 7 * 8?" and the tool result contained only "56" — no thinking tokens leaked through.

Also verified that reverting the fix causes the mock test to fail with thinking text concatenated into the result:

- "56"
+ "The user is asking about multiplication. 7 times 8 equals 56.\n56\nLet me double-check: 7 * 8 = 56. Yes, that's correct."

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

Aligned with how genai.GenerateContentResponse.Text() handles thought parts in types.go.

Generated with Claude Code