[Infra] UI Build for Release

[yuneng-jiang]

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🚄 Infrastructure

Changes

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 7, 2026 5:08pm

[greptile-apps]

Too many files changed for review. (637 files found, 100 file limit)

In general, the correct fix is to ensure all meta-characters relevant for the target context are escaped in a consistent and complete way, or to use a standard library helper if available. Here, the target context is a Python double-quoted string literal used as the value of the prompt argument in generated code. The existing code only escapes double quotes, which is insufficient; it should also escape backslashes (and optionally newlines) so that any user-supplied prompt is represented safely inside the Python string.

The best minimal fix with no functional changes elsewhere is to mirror the escaping strategy already used in the same function for x (variable O): first escape backslashes, then escape double quotes, and (optionally) escape newlines. Because this snippet is already using .replace(/"/g,'\\"'), we can simply prepend a .replace(/\\/g,"\\\\") so that every backslash in a becomes \\ before quotes are escaped. To keep behavior as close as possible to the current code, we will not introduce newline escaping here unless we see it already being assumed; in this context, Python will accept literal newlines inside a triple-quoted template string, but since the string is in a double-quoted literal in Python code, newlines should also be escaped. However, to avoid surprising changes, we will just fix the backslash escaping per the CodeQL finding.

Concretely, in litellm/proxy/_experimental/out/_next/static/chunks/3c894be4eeb5ff87.js, locate the construction of the transcription code snippet where prompt is added:

file=audio_file${a?`,
\tprompt="${a.replace(/"/g,'\\"')}"`:""}

and change it so that the prompt value uses a.replace(/\\/g,"\\\\").replace(/"/g,'\\"'). No new imports or helper functions are required; we simply extend the existing chain of .replace calls.

In general, when embedding arbitrary text into a string literal in generated code, all relevant meta-characters for that language and string-delimiter style need to be escaped. For a Python double-quoted string ("..."), at minimum you must escape backslashes, double quotes, and newlines (and optionally other control characters) so that the literal compiles and accurately represents the source text.

The best fix here is to mirror the earlier, more robust escaping used for S: escape backslashes first, then double quotes, then newlines. Specifically, instead of only doing n.replace(/"/g,'\\"'), use a chained .replace sequence: n.replace(/\\/g,"\\\\").replace(/"/g,'\\"').replace(/\n/g,"\\n"). This ensures:

• Any \ in n becomes \\ in the Python source.
• Any " in n becomes \", preventing premature string termination.
• Any newline becomes \n, keeping the Python code on one logical line.

Concretely, in litellm/proxy/_experimental/out/_next/static/chunks/a494c4bcbb8f5a7c.js, adjust the interpolation for the transcription prompt around line 392–393 to use this safer escaping chain. No new imports or helper functions are needed; only that one expression must change.