fix(streaming): log warning when tool input JSON is malformed

[Ratansairohith]

Fixes #2051

Changes

src/strands/event_loop/streaming.py:

• Added logger.warning() in the except ValueError block (line 282) that logs the tool name and raw input string when JSON parsing fails. The fallback to {} is preserved.

tests/strands/event_loop/test_streaming.py:

• Added a test case for malformed tool input JSON to verify the code path is covered.

Test plan

• All 106 existing streaming tests pass
• New test covers the malformed JSON path

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

Issue: The log message doesn't follow the project's structured logging format defined in AGENTS.md.

The format requires field=<value> pairs separated by commas, followed by a pipe | and a lowercase human-readable message. Compare with the existing warning on line 349 in this same file:

logger.warning(
    "original_stop_reason=<%s>, new_stop_reason=<%s> | "
    "overriding stop reason due to toolUse blocks in response",
    "end_turn",
    "tool_use",
)

Suggestion:

logger.warning(
    "tool_name=<%s>, raw_input=<%s> | failed to parse tool input json, defaulting to empty dict",
    current_tool_use.get("name", "unknown"),
    current_tool_use["input"][:200] if isinstance(current_tool_use.get("input"), str) else "",
)

[github-actions]

Issue: The new parametrized test case verifies the state transformation (malformed JSON → {}) but doesn't assert that logger.warning was actually called. Since the purpose of this PR is to add the warning log, the test should verify it.

Suggestion: Add a dedicated test that uses unittest.mock.patch to assert the warning is emitted with the expected arguments:

@unittest.mock.patch("strands.event_loop.streaming.logger")
def test_handle_content_block_stop_logs_warning_on_malformed_json(mock_logger):
    state = {
        "content": [],
        "current_tool_use": {"toolUseId": "123", "name": "test_tool", "input": "{invalid json}"},
        "text": "",
        "reasoningText": "",
        "citationsContent": [],
        "redactedContent": b"",
    }

    strands.event_loop.streaming.handle_content_block_stop(state)

    mock_logger.warning.assert_called_once()
    call_args = mock_logger.warning.call_args
    assert "test_tool" in str(call_args)
    assert "{invalid json}" in str(call_args)

[github-actions]

Assessment: Comment

Good improvement for observability — logging malformed JSON will help operators debug tool invocation failures, especially with smaller models.

Review Details

• Logging format: The warning message should follow the project's structured logging convention (field=<value> | message) as defined in AGENTS.md and used elsewhere in this file (see line 349).
• Test coverage: The parametrized test verifies the state transformation but doesn't assert that the warning is actually logged. A dedicated test with a mocked logger would strengthen coverage for the core behavior this PR adds.

Clean, small, well-scoped change — just the two items above to address.

[Ratansairohith]

@mkmeral thanks for the approval! Addressed the auto-review feedback in 009fba5:

• Switched the warning to the project's structured logging format (tool_name=<%s>, raw_input=<%s> | ...) per AGENTS.md.
• Added a dedicated test (test_handle_content_block_stop_logs_warning_on_malformed_json) that patches the logger and asserts the warning is emitted with the expected args.

The check-api failure looks like it's from this branch being a few weeks behind main — griffe check --against main is flagging removals/renames on main (e.g. claude-sonnet-4-6 → claude-sonnet-4-20250514, dropped 'checkpoint' stop reason). Happy to rebase onto current main to clear it if you'd like.