fix(anthropic): sanitize tool_use ids on native /v1/messages path

[Sameerlite]

Summary

• Sanitize tool_use / tool_result ids on the native Anthropic /v1/messages path so replayed cross-provider history (e.g. kimi → Anthropic) no longer 400s on ids like functions.Bash:0
• Normalize ids when converting OpenAI tool_calls to Anthropic tool_use blocks in the pass-through adapter

Fixes LIT-3881

Test plan

• pytest tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py -k "sanitize_tool_use or normalize_anthropic_tool_use"
• pytest tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py::test_translate_openai_content_to_anthropic_sanitizes_colon_dot_tool_call_ids
• pytest tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py::test_anthropic_messages_sanitizes_tool_use_ids_before_dispatch

[codecov]

Codecov Report

❌ Patch coverage is 97.36842% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/llms/anthropic/common_utils.py	97.14%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

This PR adds ID sanitization for tool_use/tool_result blocks on the native Anthropic /v1/messages path, fixing 400 errors caused by replayed cross-provider conversation history containing IDs (e.g. functions.Bash:0) that contain characters invalid under Anthropic's ^[a-zA-Z0-9_-]+$ constraint.

• Introduces normalize_anthropic_tool_use_id in common_utils.py, which strips Gemini thought-signature suffixes and replaces any remaining invalid characters with underscores; the adapter's streaming and non-streaming tool-id normalization is refactored to reuse this shared helper.
• Adds sanitize_tool_use_ids_in_anthropic_messages and calls it in both anthropic_messages (async) and anthropic_messages_handler (sync); a _litellm_messages_presanitized flag prevents redundant double-scanning on the async → sync dispatch path.
• Four targeted tests cover unit normalization, full-message sanitization, adapter transformation, and end-to-end handler dispatch, all mock-only.

Confidence Score: 5/5

Safe to merge — the sanitization is idempotent, copy-on-write, and guarded against double-application.

The change is narrowly scoped to ID normalization on an existing path, all new functions are idempotent (valid IDs pass through unchanged), the double-sanitization shortcut via _litellm_messages_presanitized is correct, and the four added tests are all mock-only with no network calls. The only theoretical concern (two distinct raw IDs colliding after normalization) is highly unlikely with real cross-provider history and would not cause silent data loss — Anthropic would return an error rather than misroute results.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/anthropic/common_utils.py	Adds normalize_anthropic_tool_use_id, _sanitize_tool_use_id_content_block, and sanitize_tool_use_ids_in_anthropic_messages — all idempotent and copy-on-write; clean implementation.
litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py	Refactors both the non-streaming and streaming tool-use ID paths to use the shared normalize_anthropic_tool_use_id helper; thought_sig is still correctly extracted for provider_specific_fields.signature.
litellm/llms/anthropic/experimental_pass_through/messages/handler.py	Adds sanitize_tool_use_ids_in_anthropic_messages call in both anthropic_messages and anthropic_messages_handler; the _litellm_messages_presanitized flag correctly prevents double-sanitization on the async path.
tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py	Adds unit tests for sanitize_tool_use_ids_in_anthropic_messages and normalize_anthropic_tool_use_id; verifies both tool_use.id and tool_result.tool_use_id normalization and non-mutation of original messages.
tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py	Adds test for colon/dot ID sanitization in the OpenAI→Anthropic adapter transformation; all mock-only, no network calls.
tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py	Adds async test verifying that anthropic_messages sanitizes tool IDs before dispatch and does not mutate the caller's original message list.

_{Reviews (3): Last reviewed commit: "Merge remote-tracking branch 'origin/lit..." | Re-trigger Greptile}

[Sameerlite]

@greptileai

[Sameerlite]

@greptileai

[mateo-berri]

LGTM; thanks!