From 0f2079860c7f71ccca6b3c96ca1c56e9e39941ca Mon Sep 17 00:00:00 2001 From: Will Ezell Date: Tue, 23 Jun 2026 22:38:18 -0400 Subject: [PATCH] fix(ai-review): stop models wrapping the whole review in a code block The output-format section showed the expected structure inside a triple-backtick fence as an example. Some models (e.g. qwen3-next-80b) copied that fence literally, wrapping their entire review in a code block so it rendered as raw text instead of markdown. Remove the example fence and instruct the model to reply in raw GitHub-flavored markdown, not wrapped in a code block. Co-Authored-By: Claude Opus 4.8 (1M context) --- .github/prompts/gpt-auto-review.md | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/.github/prompts/gpt-auto-review.md b/.github/prompts/gpt-auto-review.md index 71f93b5954db..d1fcd25c8b26 100644 --- a/.github/prompts/gpt-auto-review.md +++ b/.github/prompts/gpt-auto-review.md @@ -19,9 +19,8 @@ Check for: ## Output format -Always use this exact structure (omit a section entirely if it has no entries): +Respond with raw GitHub-flavored markdown directly in your reply. Do **not** wrap the whole response in a code block or triple backticks — the sections below are the literal output, not an example to fence. Use this exact structure (omit a section entirely if it has no entries): -``` #### New Issues - 🔴 Critical: `path/file:line` — what's wrong and why it matters - 🟠 High: `path/file:line` — what's wrong and why it matters @@ -32,7 +31,6 @@ Always use this exact structure (omit a section entirely if it has no entries): #### Resolved - ✅ `path/file:line` — brief reason it's now fixed or no longer applicable -``` If the PR is clean with no prior findings, write: `No issues found.`