fix(examples): steer PR review comments away from GFM table glitches

The model sometimes emitted findings_markdown starting with pipe characters,
which GitHub parses as a markdown table and merges with the heading/summary.

Tighten reviewer instructions: require ### Findings, forbid pipe tables and
backslash-escaped underscores. Add a ### Summary wrapper in the posted comment
body so structure stays clear even if the model drifts.

Made-with: Cursor

Author: leo-aa88

examples/pr-review-github-actions/agents/reviewer.yaml

@@ -25,9 +25,9 @@ spec:
           "evidence": "<what in the PR suggests this>"
         }
       ],
-      "findings_markdown": "<Markdown body for humans: use '### Findings' then a bullet per finding; each bullet starts with **severity** · `file` — **title**, then a short line for evidence. Use '-' bullets only. If findings is empty, use a single line like 'No issues found.'>"
+      "findings_markdown": "<GitHub-flavored Markdown for the PR comment body: start with the line ### Findings, then one '-' bullet per finding. Each bullet: **severity** · `path` — **title**, then a short evidence sentence on the next line (indented is fine). Use '-' bullets only; do not use markdown pipe tables (never begin a line with '|'). Do not backslash-escape underscores in identifiers; use plain text or backticks (e.g. OPENAI_API_KEY or `OPENAI_API_KEY`). If findings is empty, use '### Findings' followed by a line '- No issues found.'>"
     }
     The "findings" array is the structured source of truth; "findings_markdown" must describe the same items in readable form for GitHub comments.
-    If there are no issues, return an empty "findings" array, a positive summary, and findings_markdown like "No issues found."
+    If there are no issues, return an empty "findings" array, a positive summary, and findings_markdown exactly: "### Findings\n- No issues found."
   output:
     schema: ./schemas/review-output.json

examples/pr-review-github-actions/workflows/pr-review-github.yaml

@@ -40,6 +40,8 @@ spec:
         body: |
           ## Automated review
 
+          ### Summary
+
           ${steps.review_diff.output.summary}
 
           ${steps.review_diff.output.findings_markdown}