Fix editorial workflow adding code fences to files

[nearestnabors]

Summary

• Added stripCodeFences() helper to remove markdown code fence markers (\``mdx, ```md, ```markdown, or ````) from the beginning and end of LLM responses
• Applied the fix to all three places where LLM output is processed: fixValeIssues(), getEditorialFromAnthropic(), and getEditorialFromOpenAI()

Problem

The editorial review workflow was creating PRs with \``mdxat the beginning and```` at the end of documentation files. This happened because LLMs often wrap their code/markdown output in code fences, even when explicitly instructed to return only the content.

Example: PR #673 shows this issue.

Solution

The fix strips code fence markers from only the start and end of LLM responses, preserving any legitimate code blocks within the actual content.

Test plan

• Run pnpm vale:editorial --pr <number> on a recent merged PR
• Verify the generated editorial PR doesn't have code fences wrapping the file content

🤖 Generated with Claude Code

---

Note

Prevents LLM-generated editorial changes from being wrapped in code fences.

• Adds stripCodeFences() and regexes to remove starting/ending ```mdx/markdown fences
• Applies fence stripping to fixValeIssues(), getEditorialFromAnthropic(), and getEditorialFromOpenAI()
• Minor robustness: handle missing OpenAI content before processing

^{Written by Cursor Bugbot for commit 246d2f1. This will update automatically on new commits. Configure here.}

[vercel]

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

Project	Deployment	Review	Updated (UTC)
docs	Ready	Preview, Comment	Jan 20, 2026 7:30pm

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Missing empty content check after stripping code fences

Medium Severity

The getEditorialFromOpenAI function lost an empty content check during the refactor. The old code checked !revisedContent which caught empty strings after trimming. The new code only checks !rawContent before calling stripCodeFences, but doesn't verify if revisedContent is empty afterward. If an LLM returns whitespace-only content or content that becomes empty after stripping code fences, the function returns a suggestion with empty revisedContent, which would overwrite the documentation file with empty content in createEditorialPR.

 

[cursor]

Missing final trim allows whitespace to break comparison

Low Severity

The stripCodeFences function trims whitespace before stripping code fences but not after. If an LLM wraps its response in code fences and includes leading/trailing whitespace inside (e.g., "```mdx\n NO_CHANGES_NEEDED \n```"), the result will be " NO_CHANGES_NEEDED " with preserved internal whitespace. This causes the "NO_CHANGES_NEEDED" equality check to fail, potentially creating an editorial PR that overwrites documentation with the literal text " NO_CHANGES_NEEDED ".