fix(anthropic): sanitize tool_use IDs in convert_to_anthropic_tool_invoke

[Canvinus]

Relevant issues

Encountered invalid_request_error from Anthropic when routing multi-provider conversations — tool call IDs generated by other providers (MiniMax, Composio, etc.) can contain characters like : or . that break Anthropic's ^[a-zA-Z0-9_-]+$ validation on tool_use.id.

Pre-Submission checklist

• I have added testing in tests/llm_translation/test_prompt_factory.py
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🐛 Bug Fix

Changes

_sanitize_anthropic_tool_use_id() already exists and is correctly applied to tool_result blocks in convert_to_anthropic_tool_result (lines 1714, 1728). However, it's missing from tool_use blocks in convert_to_anthropic_tool_invoke — the tool_id is passed raw.

This means when a conversation history contains tool calls from a non-Anthropic provider (e.g. MiniMax generating IDs like sessions_history:183), and the conversation is then sent to an Anthropic model, the request fails with:

messages.75.content.1.tool_use.id: String should match pattern '^[a-zA-Z0-9_-]+$'

The fix applies the same _sanitize_anthropic_tool_use_id() to regular tool_use IDs — 2 lines changed. Both tool_use and tool_result now go through the same sanitizer, so ID matching stays consistent.

Added a test covering colons, dots, and valid IDs passing through unchanged.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 24, 2026 0:38am

[CLAassistant]

All committers have signed the CLA.

[Copilot]

Pull request overview

This PR fixes a bug where tool call IDs from non-Anthropic providers (e.g., MiniMax, Composio) containing invalid characters like colons or dots cause invalid_request_error when routing multi-provider conversations to Anthropic. Anthropic requires tool_use IDs to match the pattern ^[a-zA-Z0-9_-]+$.

Changes:

• Applied existing _sanitize_anthropic_tool_use_id() sanitization to tool_use blocks in convert_to_anthropic_tool_invoke()
• Added comprehensive test coverage for ID sanitization with various invalid characters

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
litellm/litellm_core_utils/prompt_templates/factory.py	Applied ID sanitization to regular tool_use blocks, ensuring consistency with tool_result handling
tests/llm_translation/test_prompt_factory.py	Added test validating sanitization of invalid characters (colons, dots) in tool IDs and that valid IDs pass through unchanged

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[greptile-apps]

Greptile Summary

This PR fixes an invalid_request_error from Anthropic when routing multi-provider conversations by applying the existing _sanitize_anthropic_tool_use_id() sanitizer to tool_use blocks in convert_to_anthropic_tool_invoke. The sanitizer was already correctly applied to tool_result blocks (lines 1714, 1728) but was missing from the tool_use path, causing failures when tool call IDs from non-Anthropic providers (e.g., MiniMax, Composio) contained characters like : or . that violate Anthropic's ^[a-zA-Z0-9_-]+$ pattern.

• Applies _sanitize_anthropic_tool_use_id() to the regular tool_use branch in convert_to_anthropic_tool_invoke, consistent with the existing tool_result handling
• Adds a focused test covering sanitization of colons, dots, and pass-through of valid IDs
• No new functions or abstractions introduced — reuses the existing sanitizer

Confidence Score: 5/5

• This PR is safe to merge — it extends an existing sanitizer to a missing code path with no behavioral changes to valid IDs.
• The change is a 2-line fix that reuses an existing, well-tested sanitizer function. It follows the exact same pattern already applied to tool_result blocks. The regex-based sanitizer is idempotent for valid IDs (characters matching ^[a-zA-Z0-9_-]+$ pass through unchanged). Both tool_use and tool_result now consistently sanitize IDs, maintaining ID matching. The test covers the key edge cases.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/litellm_core_utils/prompt_templates/factory.py	Applies existing _sanitize_anthropic_tool_use_id() to regular tool_use blocks in convert_to_anthropic_tool_invoke, matching the pattern already used for tool_result blocks. Minimal 2-line change, no new logic introduced.
tests/llm_translation/test_prompt_factory.py	Adds test covering sanitization of invalid tool_use IDs (colons, dots) and verifying valid IDs pass through unchanged. Test is well-structured and covers the key scenarios.

Sequence Diagram

sequenceDiagram
    participant External as Non-Anthropic Provider
    participant LiteLLM as LiteLLM Router
    participant Sanitizer as _sanitize_anthropic_tool_use_id
    participant Anthropic as Anthropic API

    External->>LiteLLM: tool_call with id "sessions_history:183"
    LiteLLM->>LiteLLM: convert_to_anthropic_tool_invoke()
    LiteLLM->>Sanitizer: sanitize tool_use id
    Sanitizer-->>LiteLLM: "sessions_history_183"
    LiteLLM->>Anthropic: tool_use block with sanitized id
    Anthropic-->>LiteLLM: Response
    LiteLLM->>LiteLLM: convert_to_anthropic_tool_result()
    LiteLLM->>Sanitizer: sanitize tool_result id (already existed)
    Sanitizer-->>LiteLLM: "sessions_history_183" (consistent)
    LiteLLM->>Anthropic: tool_result with matching id

Loading

_{Last reviewed commit: 02d71bc}

[greptile-apps]

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}