[gaurav0107]

Summary

• openaiResponsesToOpenAIRequest forwarded client tool schemas to the Chat Completions upstream unchanged. Harnesses that deep-truncate their schemas (Codex Desktop) send array keywords replaced by placeholder strings, e.g. "enum": "[MaxDepth]". Ollama Cloud validates tool schemas with Go structs and rejects the request before inference: cannot unmarshal string into Go struct field ToolProperty.enum of type []interface {}.
• A post-pass now runs every emitted function.parameters through the existing stripInvalidSchemaConstructs sanitizer, which the Claude path already uses (as suggested in the issue and confirmed by the maintainer). One pass covers plain function tools, flattened namespace (MCP) sub-tools, tool_search, and tools that arrive already in Chat shape.
• Valid schemas are unchanged. The schemaCoercion.ts doc comment, which said OpenAI targets tolerate these constructs, now notes the strict-validator case.

Related Issues

• Closes fix(providers): Responses->Chat translator forwards string-valued enum in tool schemas; Ollama Cloud rejects with 'cannot unmarshal string into ToolProperty.enum' #18679
• Related to fix(api): cap re-injected tool-output images in Responses-to-Chat to avoid oversized requests #18684 (the image re-injection problem split out of the same report; not addressed here)

Target Branch

• Base is the active release/v3.8.x branch (release/v3.8.53), not main
• New feature? Not applicable: this is a bug fix

Validation

• Change type: routing / translator
• Focused tests: the new test file plus every test file that calls openaiResponsesToOpenAIRequest, and the schema-coercion suites
• eslint + prettier --check on the changed files, and npm run check:open-sse-typecheck
• Reconciled with the current release/v3.8.53 tip
• Production-code change includes a new automated test

Without the fix, the new test fails on the three invalid-schema cases:

✖ #18679 Responses -> Chat drops a string-valued enum from a function tool schema
✖ #18679 Responses -> Chat sanitizes namespace (MCP) sub-tool schemas too
✖ #18679 Responses -> Chat coerces an index-keyed enum object to an array
✔ #18679 Responses -> Chat leaves a valid tool schema unchanged
ℹ tests 4
ℹ pass 1
ℹ fail 3

With the fix:

$ node --import tsx/esm --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit tests/unit/responses-to-chat-tool-schema-sanitize-18679.test.ts
ℹ tests 4
ℹ pass 4
ℹ fail 0

# 56 files: every test that imports openaiResponsesToOpenAIRequest, plus the
# openai-responses / responses-to-chat / schema-coercion suites
ℹ tests 523
ℹ pass 523
ℹ fail 0

$ npm run check:open-sse-typecheck
openSseTypecheckErrors=0

$ npm run check:file-size && npm run check:complexity && npm run check:test-discovery
OK / OK / OK

Tests Added Or Updated

• Changelog fragment: changelog.d/fixes/18806-responses-chat-tool-schema-sanitize.md.

• tests/unit/responses-to-chat-tool-schema-sanitize-18679.test.ts (new) asserts on the translated Chat body:

  • a function tool with "enum": "[MaxDepth]" nested under anyOf[0].properties (the shape from the report) has no string-valued enum left, and its valid sibling enum is kept;
  • the same schema on a namespace (MCP) sub-tool is sanitized the same way;
  • an index-keyed enum object ({"0":"a","1":"b"}) becomes ["a","b"];
  • a valid schema with additionalProperties: false is deep-equal to the input.

Coverage Notes

• The open-sse/translator/request/openai-responses.ts change (the new sanitizeChatToolParameters helper and its call site) is covered by the new test file and by the existing translator suites listed above. schemaCoercion.ts changes only a comment.

Reviewer Notes

• stripInvalidSchemaConstructs also turns numeric-string constraints into numbers (minimum: "5" to 5) and replaces placeholder strings in schema positions with {}. These Chat targets now get the same normalization the Claude path already applies.
• Tools without function.parameters (for example web_search passed through as-is) are returned by reference, unchanged.

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: cheaper-inference/OmniRoute
• Branch: release/v3.8.53
• File: .coderabbit.yaml
• Commit: f000c11
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Advanced
• Run ID: 5de99618-ec39-44f5-bbb0-dc7775417635

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts