Merge pull request #369 from netwrix/doc-review-diff-only

Doc review diff only

Author: jth-nw

.github/workflows/claude-documentation-fixer.yml

@@ -97,13 +97,44 @@ jobs:
             --allowedTools "Read,Write,Edit,Bash(vale:*),Bash(gh pr view:*),Bash(gh pr diff:*),Bash(git config:*),Bash(git add:*),Bash(git commit:*),Bash(git push:*),Bash(git status:*),Bash(git diff:*)"
             --append-system-prompt "${{ steps.read-prompt.outputs.prompt }}"
 
+      - name: Record last comment ID
+        id: pre-claude
+        if: steps.pr-info.outputs.is_fork == 'false' && steps.cmd-type.outputs.is_preexisting == 'true'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          LAST_ID=$(gh api repos/${{ github.repository }}/issues/${{ steps.pr-info.outputs.number }}/comments \
+            --jq 'if length > 0 then .[-1].id else 0 end' 2>/dev/null || echo "0")
+          echo "last_comment_id=$LAST_ID" >> "$GITHUB_OUTPUT"
+
       - name: Detect preexisting issues
         if: steps.pr-info.outputs.is_fork == 'false' && steps.cmd-type.outputs.is_preexisting == 'true'
         uses: anthropics/claude-code-action@v1
         with:
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
           github_token: ${{ secrets.GITHUB_TOKEN }}
           show_full_output: false
+          prompt: |
+            Detect preexisting issues in PR ${{ steps.pr-info.outputs.number }}.
+
+            Follow these steps in order:
+
+            1. Use `gh pr diff ${{ steps.pr-info.outputs.number }}` to identify which lines are part of the PR diff.
+
+            2. Use `gh pr view ${{ steps.pr-info.outputs.number }}` to get the list of changed files, then read the full content of each changed markdown file.
+
+            3. Run all three review passes on each file per your instructions, but report ONLY issues on lines that are NOT part of the PR diff.
+
+            4. You MUST write your results ONLY to `/tmp/preexisting-issues.md` — always, even if there are no issues. Do not post a comment. Do not write anything else. Use this exact format:
+
+            ## Preexisting issues
+            ### path/to/file.md
+            - Line N: issue. Suggested change: '...'
+
+            Or if there are no issues:
+
+            ## Preexisting issues
+            None.
           claude_args: |
             --model claude-sonnet-4-5-20250929
             --allowedTools "Read,Write,Edit,Bash(gh pr view:*),Bash(gh pr diff:*)"
@@ -115,6 +146,7 @@ jobs:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           PR_NUMBER: ${{ steps.pr-info.outputs.number }}
           REPO: ${{ github.repository }}
+          LAST_COMMENT_ID: ${{ steps.pre-claude.outputs.last_comment_id }}
         run: |
           python3 << 'PYTHON_EOF'
           import os
@@ -123,6 +155,7 @@ jobs:
 
           pr_number = os.environ['PR_NUMBER']
           repo = os.environ['REPO']
+          last_comment_id = int(os.environ.get('LAST_COMMENT_ID', '0'))
 
           FOOTER = (
               "\n\n---\n\n"
@@ -157,8 +190,28 @@ jobs:
           else:
               clean_body = '## Preexisting issues\nNone.' + FOOTER
 
-          subprocess.run(
-              ['gh', 'pr', 'comment', pr_number, '--repo', repo, '--body', clean_body],
-              check=True,
+          # Find the action's auto-posted comment (ID > last recorded, posted by a bot)
+          result = subprocess.run(
+              ['gh', 'api', f'repos/{repo}/issues/{pr_number}/comments'],
+              capture_output=True, text=True, check=True,
           )
+          comments = json.loads(result.stdout)
+          new_bot_comments = [c for c in comments
+                              if c['id'] > last_comment_id
+                              and c['user']['login'].endswith('[bot]')]
+
+          if new_bot_comments:
+              # Replace the action's auto-comment in-place with our formatted output
+              target_id = new_bot_comments[-1]['id']
+              subprocess.run(
+                  ['gh', 'api', f'repos/{repo}/issues/comments/{target_id}',
+                   '-X', 'PATCH', '--input', '-'],
+                  input=json.dumps({'body': clean_body}),
+                  capture_output=True, text=True, check=True,
+              )
+          else:
+              subprocess.run(
+                  ['gh', 'pr', 'comment', pr_number, '--repo', repo, '--body', clean_body],
+                  check=True,
+              )
           PYTHON_EOF

.github/workflows/claude-documentation-reviewer.yml

@@ -251,16 +251,39 @@ jobs:
               return valid
 
           def normalize_review_body(body):
-              """Convert any heading-formatted issue lines to bullet points."""
-              lines = []
-              for line in body.split('\n'):
-                  # Convert "### Line N: ..." or "#### Line N: ..." etc. to "- Line N: ..."
+              """Normalize issue formatting to single-line bullet points."""
+              src = body.split('\n')
+              result = []
+              i = 0
+              while i < len(src):
+                  line = src[i]
+
+                  # Convert heading format: ### Line N: ... → - Line N: ...
                   m = re.match(r'^#{1,6}\s+(Line \d+:.+)$', line)
                   if m:
-                      lines.append(f'- {m.group(1)}')
-                  else:
-                      lines.append(line)
-              return '\n'.join(lines)
+                      result.append(f'- {m.group(1)}')
+                      i += 1
+                      continue
+
+                  # Convert bold format: **Line N: title** + sub-bullets → - Line N: single line
+                  m = re.match(r'^\*\*(Line \d+:.*?)\*\*\s*$', line)
+                  if m:
+                      title = m.group(1).rstrip('.')
+                      i += 1
+                      parts = []
+                      while i < len(src) and re.match(r'^\s*[-*]\s+', src[i]):
+                          sub = re.sub(r'^\s*[-*]\s+', '', src[i])
+                          sub = re.sub(r'^(Issue|Fix|Description|Suggested change):\s*', '', sub, flags=re.IGNORECASE)
+                          if sub.strip():
+                              parts.append(sub.strip().rstrip('.'))
+                          i += 1
+                      combined = f'- {title}. {". ".join(parts)}.' if parts else f'- {title}.'
+                      result.append(combined)
+                      continue
+
+                  result.append(line)
+                  i += 1
+              return '\n'.join(result)
 
           # Read the review summary Claude wrote
           summary_path = '/tmp/review-summary.md'