test: lock in Bedrock and OpenAI _parse_native_tool_call arg handling (#4972)

[adityasingh2400]

Summary

Issue #4972 reported that AWS Bedrock Converse tool calls had their arguments silently dropped because the dict branch of _parse_native_tool_call used a truthy default in func_info.get("arguments", "{}"), which short-circuited the or before the Bedrock input key was read.

The runtime fix shipped in commit 25fcf39 (fix: preserve Bedrock tool call arguments by removing truthy default), but no test locked in the behaviour, so the issue stayed open. This PR adds the regression coverage.

What is covered

A new TestParseNativeToolCall class in lib/crewai/tests/agents/test_native_tool_calling.py with parametrized cases:

• Bedrock Converse dict shape {"name": ..., "input": {...}, "toolUseId": ...} — asserts toolUseId is used as call id, the tool name is parsed, and input round-trips as the args dict.
• OpenAI dict shape {"id": ..., "function": {"name": ..., "arguments": "...json..."}} — asserts the id is preserved and the JSON arguments string is returned unchanged.
• A third case asserts the parsed args dict equals the original Bedrock input (and is not the empty dict that triggered the bug).

I verified the tests catch the regression by temporarily restoring the buggy default (func_info.get("arguments", "{}")): the two Bedrock cases fail with '{}' == {...} and the OpenAI case still passes, exactly matching the bug surface in the issue.

This is a test-only PR, no source changes.

Test plan

• pytest tests/agents/test_native_tool_calling.py -v -k parse_native_tool_call passes (3/3)
• Reverted the fix locally and confirmed the two Bedrock cases fail with the exact symptom from [BUG] _parse_native_tool_call drops Bedrock Converse API tool arguments — always passes empty dict #4972
• Restored the fix and confirmed all three cases pass

Fixes #4972

Summary by CodeRabbit

• Tests
  • Added a regression test suite validating parsing of native tool-call formats from multiple LLM providers, ensuring call IDs, tool/function names, and arguments are correctly preserved and extracted across provider-specific shapes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 22f66fd7-d668-4e1c-af09-97fcdbfb8627

📥 Commits

Reviewing files that changed from the base of the PR and between dc8f655 and 3a0709b.

📒 Files selected for processing (1)

• lib/crewai/tests/agents/test_native_tool_calling.py

---

📝 Walkthrough

Walkthrough

This PR adds regression test coverage for the _parse_native_tool_call method, verifying parsing of Bedrock Converse ({"name","input","toolUseId"}) and OpenAI-style ({"function":{"name","arguments"}}) tool-call dicts and ensuring Bedrock tool arguments are preserved.

Changes

Native tool-call parsing regression tests

Layer / File(s)	Summary
Regression test suite for _parse_native_tool_call lib/crewai/tests/agents/test_native_tool_calling.py	Parametrized test class TestParseNativeToolCall covers both Bedrock Converse dict shape ({"name", "input", "toolUseId"}) and OpenAI-style dict shape ({"function": {"name", "arguments"}}). Dedicated assertion ensures parsed Bedrock arguments match the original input dict and are not the empty dict that caused the regression.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 A tiny test hopped into the suite,
To catch where Bedrock's inputs lost their feet.
It checks both styles, precise and small—
So args stand proud and never fall.
🎉

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately and concisely describes the main change—adding regression tests to lock in Bedrock and OpenAI tool call argument handling in _parse_native_tool_call.
Linked Issues check	✅ Passed	The PR adds comprehensive regression tests covering both Bedrock Converse dict shape and OpenAI dict shape, directly addressing issue #4972's requirements to validate argument preservation in _parse_native_tool_call.
Out of Scope Changes check	✅ Passed	All changes are in-scope test additions directly related to the linked issue #4972, with no runtime code modifications or unrelated alterations.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[adityasingh2400]

Hi maintainers, this PR is from a fork and the required CI workflows (Lint, Run Type Checks, Run Tests on 3.10 to 3.13) are sitting in the action_required state, which is the first-time-contributor gate. Could one of you approve the workflow runs so the required checks can complete and the PR can be merged? The change is test-only (one new file under lib/crewai/tests/agents/) and locks in the Bedrock arg-handling fix from #4972. Happy to rebase if you need anything else. Thanks.