fix(parser): wrap non-dict message and rate_limit_info in MessageParseError

[Amey-Thakur]

Summary

parse_message wraps malformed frames in MessageParseError so callers can catch one documented exception and recover the offending frame from error.data. Two frame shapes slipped past that contract and raised a raw TypeError instead, which escapes parse_message and tears down the receive_messages() / query() async generator.

This continues the hardening from #1064 (non-dict assistant content and non-dict content blocks) and the rate_limit_event field guards (#599 / #601 / #689, which covered missing keys). The two sibling shapes below were still unguarded.

The gap

Both the user and assistant branches index data["message"]["content"] before checking that message is a dict, and the rate_limit_event branch reads info["status"] before checking that rate_limit_info is a dict.

from claude_agent_sdk._internal.message_parser import parse_message

parse_message({"type": "assistant", "message": "oops"})
# TypeError: string indices must be integers, not 'str'

parse_message({"type": "user", "message": 5})
# TypeError: 'int' object is not subscriptable

parse_message({"type": "rate_limit_event", "uuid": "u", "session_id": "s",
               "rate_limit_info": 5})
# TypeError: 'int' object is not subscriptable

Because the TypeError is not wrapped in MessageParseError, a consumer catching MessageParseError does not catch it, the raw frame is not attached for diagnosis, and the message stream is torn down. This is the same failure mode #1064 describes, and as that issue notes, Anthropic-compatible backends that emit slightly different frame shapes are the common trigger.

Fix

Add the isinstance(..., dict) guard, with the same MessageParseError wording already used for non-dict content blocks, at the two sites that were missing it:

• user / assistant: validate data["message"] is a dict before indexing into it.
• rate_limit_event: validate data["rate_limit_info"] is a dict before reading its keys.

Every frame that already parsed correctly is unchanged; only the three previously-crashing shapes now raise the documented MessageParseError instead of a raw TypeError.

Tests

tests/test_message_parser.py:

• test_non_dict_message_raises_documented_error — parametrized over assistant / user and over a string, int, list, and None message.
• test_non_dict_rate_limit_info_raises_documented_error — parametrized over the same non-dict values.

Both assert MessageParseError and the expected dict message. They fail with a raw TypeError without this change and pass with it. The full tests/test_message_parser.py suite is green (92 passed). ruff check, ruff format --check, and mypy on the changed files are clean.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.