#10232: Fix: Add anthropic mode for tool call ID sanitization

## Summary When switching from other models (e.g., Kimi K2.5) to Claude mid-session, the conversation history may contain tool call IDs with characters like `|` or `:` that Claude rejects. ## Error Message ``` 400 Provider returned error {"message":"messages.3.content.1.tool_use.id: String should match pattern '^[a-zA-Z0-9_-]+$'"} ``` ## Root Cause Claude requires `tool_use.id` to match the pattern `^[a-zA-Z0-9_-]+$`. This is actually less strict than the existing `strict` mode (which only allows alphanumeric characters). Other models like Kimi K2.5 generate tool call IDs with characters like `:` and `|` which are valid for their APIs but not for Anthropic's Claude. ## Solution This PR adds an `anthropic` mode for tool call ID sanitization that: 1. Allows characters `[a-zA-Z0-9_-]` as required by Claude 2. Enables sanitization for all Anthropic models (not just when switching) 3. Is less strict than `strict` mode (preserves underscores and hyphens) ## Changes - `src/agents/tool-call-id.ts`: Added `anthropic` mode to `sanitizeToolCallId` and `isValidCloudCodeAssistToolId` - `src/agents/transcript-policy.ts`: Updated to enable sanitization for Anthropic models with the new `anthropic` mode - Added comprehensive tests for the new mode ## Testing Added tests to verify: - Underscores and hyphens are preserved - Pipes (`|`) and colons (`:`) are stripped - Kimi K2.5 style IDs are properly sanitized - Valid Claude-style IDs remain unchanged - Collisions are handled with underscore suffixes Fixes #10222 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> - Adds an `anthropic` tool-call-id sanitization mode to preserve `[_-]` while stripping characters Claude rejects (e.g. `:`/`|`). - Updates transcript policy so Anthropic models always sanitize tool call IDs using the new mode. - Extends transcript ID rewrite logic/tests to ensure toolCall/toolResult IDs stay paired and collision-free. - Adds targeted unit tests for the new sanitization behavior and collision handling. <h3>Confidence Score: 2/5</h3> - This PR should not merge until the strict-mode collision handling bug is fixed, as it can generate invalid tool IDs in strict mode. - The new anthropic mode itself is straightforward and well-tested, but a boolean logic bug in `makeUniqueToolId` makes strict mode produce underscore-separated IDs on collisions, violating strict’s alphanumeric-only contract and potentially breaking providers that require it. - src/agents/tool-call-id.ts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->