review-pr.sh•11.7 kB
#!/bin/bash
# Automated PR review script using claude -p with allowedTools support
# Usage: ./scripts/review-pr.sh <PR_NUMBER>
set -e
if [ $# -eq 0 ]; then
echo "Usage: $0 <PR_NUMBER>"
echo "Example: $0 123"
exit 1
fi
PR_NUMBER=$1
echo "🤖 Starting automated review for PR #$PR_NUMBER..."
# Verify required tools
for tool in gh jq claude; do
if ! command -v $tool &> /dev/null; then
echo "❌ $tool is required but not installed"
exit 1
fi
done
# Verify PR exists
if ! gh pr view $PR_NUMBER >/dev/null 2>&1; then
echo "❌ PR #$PR_NUMBER not found"
exit 1
fi
echo "📋 Fetching comprehensive PR details..."
# Get comprehensive PR information
echo "🔍 Calling gh pr view..."
PR_INFO=$(gh pr view $PR_NUMBER --json title,body,files,additions,deletions,author,createdAt,baseRefName,headRefName)
echo "✅ PR info retrieved"
echo "🔍 Parsing PR data..."
PR_TITLE=$(echo "$PR_INFO" | jq -r '.title')
PR_BODY=$(echo "$PR_INFO" | jq -r '.body // ""')
PR_AUTHOR=$(echo "$PR_INFO" | jq -r '.author.login')
PR_CREATED=$(echo "$PR_INFO" | jq -r '.createdAt')
BASE_BRANCH=$(echo "$PR_INFO" | jq -r '.baseRefName')
HEAD_BRANCH=$(echo "$PR_INFO" | jq -r '.headRefName')
FILES_COUNT=$(echo "$PR_INFO" | jq -r '.files | length')
ADDITIONS=$(echo "$PR_INFO" | jq -r '.additions')
DELETIONS=$(echo "$PR_INFO" | jq -r '.deletions')
echo "✅ PR data parsed successfully"
# Get all changed files
echo "🔍 Getting changed files..."
FILES_CHANGED=$(echo "$PR_INFO" | jq -r '.files[].path')
echo "✅ Changed files retrieved"
# Get file stats to determine review approach
# Check if this is a very large PR based on file count and line changes
echo "🔍 Calculating PR statistics..."
TOTAL_CHANGES=$((ADDITIONS + DELETIONS))
echo "📊 PR Statistics: $FILES_COUNT files, +$ADDITIONS/-$DELETIONS lines (total: $TOTAL_CHANGES changes)"
# Auto-detect very large PRs to avoid hanging on diff commands
if [ $TOTAL_CHANGES -gt 10000 ] || [ $FILES_COUNT -gt 50 ]; then
echo "⚠️ Very large PR detected based on file count/changes - using file-by-file analysis"
DIFF_SIZE=999999
REVIEW_TYPE="VERY_LARGE_PR"
# Get file summaries for very large PRs
FILE_STATS=$(echo "$PR_INFO" | jq -r '.files[] | "\(.path): +\(.additions)/-\(.deletions)"')
# Try to get sample content from a few key files
SAMPLE_FILES=$(echo "$PR_INFO" | jq -r '.files[0:5][] | .path')
PR_DIFF_SAMPLE="# Very Large PR - Sample from first 5 files:\n"
for file in $SAMPLE_FILES; do
echo "Getting sample from $file..."
# Get individual file diff using the files API
FILE_DIFF=$(gh api repos/{owner}/{repo}/pulls/$PR_NUMBER/files --jq ".[] | select(.filename == \"$file\") | .patch // \"File too large or binary\"" 2>/dev/null || echo "File diff unavailable")
PR_DIFF_SAMPLE="$PR_DIFF_SAMPLE\n## $file\n\`\`\`diff\n$(echo "$FILE_DIFF" | head -20)\n\`\`\`\n"
done
else
# Handle GitHub API diff size limits (20K lines max) for smaller PRs
echo "🔍 Attempting to get PR diff (with 10s timeout)..."
DIFF_OUTPUT=$(timeout --signal=KILL 10s gh pr diff $PR_NUMBER 2>&1)
DIFF_EXIT_CODE=$?
# Check if timeout occurred or API limit hit
if [ $DIFF_EXIT_CODE -eq 124 ] || [ $DIFF_EXIT_CODE -eq 137 ]; then
echo "⚠️ Timeout: PR diff took longer than 10 seconds - treating as very large PR"
DIFF_SIZE=999999
REVIEW_TYPE="VERY_LARGE_PR"
FILE_STATS=$(echo "$PR_INFO" | jq -r '.files[] | "\(.path): +\(.additions)/-\(.deletions)"')
PR_DIFF_SAMPLE="# Timeout occurred while fetching diff - file-level analysis only"
elif [ $DIFF_EXIT_CODE -ne 0 ] && echo "$DIFF_OUTPUT" | grep -q "diff exceeded the maximum number of lines"; then
echo "⚠️ GitHub API limit: diff exceeds 20K line limit - using file-by-file analysis"
DIFF_SIZE=999999
REVIEW_TYPE="VERY_LARGE_PR"
FILE_STATS=$(echo "$PR_INFO" | jq -r '.files[] | "\(.path): +\(.additions)/-\(.deletions)"')
PR_DIFF_SAMPLE="# GitHub API limit exceeded - file-level analysis only"
else
DIFF_SIZE=$(echo "$DIFF_OUTPUT" | wc -c)
echo "📊 Diff size: $DIFF_SIZE characters"
# Determine review approach based on size
if [ $DIFF_SIZE -gt 100000 ]; then
echo "⚠️ Large PR detected - using focused review approach"
FILE_STATS=$(echo "$PR_INFO" | jq -r '.files[] | "\(.path): +\(.additions)/-\(.deletions)"')
PR_DIFF_SAMPLE=$(echo "$DIFF_OUTPUT" | grep -E "^(diff|@@|\+\+\+|---|\+[^+]|\-[^-])" | head -200)
REVIEW_TYPE="LARGE_PR_SUMMARY"
else
echo "📝 Small/Medium PR - using detailed review approach"
PR_DIFF="$DIFF_OUTPUT"
REVIEW_TYPE="FULL_ANALYSIS"
fi
fi
fi
echo "🔍 Running Claude analysis ($REVIEW_TYPE)..."
# Create comprehensive review prompt
cat > /tmp/pr_review_prompt_${PR_NUMBER}.md << EOF
You are an expert code reviewer for the Attio MCP Server project.
Perform a comprehensive review of this Pull Request and provide output in the exact format below:
🎯 **Overview**
Brief summary of what this PR accomplishes and its scope
✅ **Strengths**
- Key positive aspects and good practices followed
- Well-implemented features and patterns
- Good TypeScript usage and type safety
⚠️ **Critical Issues**
- Bugs or problems that must be fixed before merge
- Breaking changes or compatibility issues
- Security vulnerabilities or concerns
💡 **Suggestions**
- Code improvements and optimizations
- MCP schema compliance (avoid oneOf/allOf/anyOf at top level)
- TypeScript best practices and error handling
- Performance considerations
🧪 **Testing Requirements**
- What needs testing before merge
- Specific test scenarios to validate
- Integration test considerations
📋 **Action Items**
- [ ] Required changes for approval
- [ ] Recommended improvements
- [ ] Documentation updates needed
**Recommendation**: [APPROVE / REQUEST CHANGES / NEEDS DISCUSSION]
Focus on Attio MCP Server project conventions from CLAUDE.md, TypeScript standards, MCP protocol compliance, error handling, and actionable feedback.
EOF
# Create data file for claude -p
if [ "$REVIEW_TYPE" = "VERY_LARGE_PR" ]; then
cat > /tmp/pr_data_${PR_NUMBER}.md << EOF
# PR #${PR_NUMBER} Review Data (Very Large PR - File Summary Analysis)
## PR Information
**Title:** ${PR_TITLE}
**Author:** ${PR_AUTHOR}
**Created:** ${PR_CREATED}
**Branch:** ${HEAD_BRANCH} → ${BASE_BRANCH}
**Files Changed:** ${FILES_COUNT}
**Lines:** +${ADDITIONS}/-${DELETIONS}
**Description:**
${PR_BODY}
**File Change Summary:**
${FILE_STATS}
**Sample Code Changes (First 5 files with 20-line previews):**
$(echo -e "$PR_DIFF_SAMPLE")
**Note:** This PR exceeds GitHub's 20K line diff limit. Review focuses on file-level changes, architecture patterns, and high-level impact assessment rather than detailed line-by-line analysis.
**Review Strategy:**
- Focus on architectural changes and patterns
- Identify potential breaking changes or compatibility issues
- Assess security implications of large-scale changes
- Recommend testing strategies for comprehensive changes
- Highlight areas that need careful manual review
EOF
elif [ "$REVIEW_TYPE" = "LARGE_PR_SUMMARY" ]; then
cat > /tmp/pr_data_${PR_NUMBER}.md << EOF
# PR #${PR_NUMBER} Review Data (Large PR - Summary Analysis)
## PR Information
**Title:** ${PR_TITLE}
**Author:** ${PR_AUTHOR}
**Created:** ${PR_CREATED}
**Branch:** ${HEAD_BRANCH} → ${BASE_BRANCH}
**Files Changed:** ${FILES_COUNT}
**Lines:** +${ADDITIONS}/-${DELETIONS}
**Description:**
${PR_BODY}
**File Change Summary:**
${FILE_STATS}
**Key Diff Patterns (Sample - 200 lines):**
\`\`\`diff
${PR_DIFF_SAMPLE}
\`\`\`
**Note:** This is a large PR (${DIFF_SIZE} chars). Review focuses on architecture, patterns, and high-level changes rather than line-by-line analysis.
EOF
else
cat > /tmp/pr_data_${PR_NUMBER}.md << EOF
# PR #${PR_NUMBER} Review Data
## PR Information
**Title:** ${PR_TITLE}
**Author:** ${PR_AUTHOR}
**Created:** ${PR_CREATED}
**Branch:** ${HEAD_BRANCH} → ${BASE_BRANCH}
**Files Changed:** ${FILES_COUNT}
**Lines:** +${ADDITIONS}/-${DELETIONS}
**Description:**
${PR_BODY}
**Files Modified:**
${FILES_CHANGED}
**Complete Diff:**
\`\`\`diff
${PR_DIFF}
\`\`\`
EOF
fi
# Run claude with the prompt and data using the allowed-tools flag
# Note: --allowedTools automatically outputs JSON conversation logs
claude -p /tmp/pr_review_prompt_${PR_NUMBER}.md --allowedTools "mcp__clear-thought-server__sequentialthinking" < /tmp/pr_data_${PR_NUMBER}.md > /tmp/review_output_json_${PR_NUMBER}.md
# Extract the actual review content from the JSON conversation log
# The --allowedTools flag causes Claude to output an array of conversation objects
# We need to find the last object with "type":"result" and extract its "result" field
# First, check if the output is valid JSON
if ! cat /tmp/review_output_json_${PR_NUMBER}.md | jq empty >/dev/null 2>&1; then
echo "⚠️ Output is not valid JSON, treating as plain text"
cp /tmp/review_output_json_${PR_NUMBER}.md /tmp/review_output_${PR_NUMBER}.md
elif cat /tmp/review_output_json_${PR_NUMBER}.md | jq -e '. | type == "array"' >/dev/null 2>&1; then
# Extract from conversation log array - get the last result object
RESULT_CONTENT=$(cat /tmp/review_output_json_${PR_NUMBER}.md | jq -r '.[] | select(.type == "result") | .result' | tail -1)
if [ -n "$RESULT_CONTENT" ] && [ "$RESULT_CONTENT" != "null" ]; then
echo "$RESULT_CONTENT" > /tmp/review_output_${PR_NUMBER}.md
else
echo "⚠️ No result content found in conversation log, using raw output"
cp /tmp/review_output_json_${PR_NUMBER}.md /tmp/review_output_${PR_NUMBER}.md
fi
else
# Fallback: if it's not an array, try direct text extraction
if cat /tmp/review_output_json_${PR_NUMBER}.md | jq -e '. | type == "string"' >/dev/null 2>&1; then
cat /tmp/review_output_json_${PR_NUMBER}.md | jq -r '.' > /tmp/review_output_${PR_NUMBER}.md
else
# Last resort: use raw content (might be plain text that looks like JSON)
cp /tmp/review_output_json_${PR_NUMBER}.md /tmp/review_output_${PR_NUMBER}.md
fi
fi
# Check if review extraction succeeded
if [ ! -s /tmp/review_output_${PR_NUMBER}.md ]; then
echo "⚠️ Failed to extract review content from Claude output."
echo "Raw output saved to /tmp/review_output_json_${PR_NUMBER}.md for debugging"
echo "First 200 characters of raw output:"
head -c 200 /tmp/review_output_json_${PR_NUMBER}.md
echo
exit 1
fi
# Additional check: ensure the extracted content looks like a review (not empty or just JSON)
EXTRACTED_SIZE=$(wc -c < /tmp/review_output_${PR_NUMBER}.md)
if [ $EXTRACTED_SIZE -lt 50 ]; then
echo "⚠️ Extracted review content seems too short ($EXTRACTED_SIZE chars)"
echo "Content preview:"
cat /tmp/review_output_${PR_NUMBER}.md
echo
echo "This may indicate a parsing issue. Raw output saved to /tmp/review_output_json_${PR_NUMBER}.md"
fi
echo "📝 Posting review to PR..."
# Post review as comment
gh pr comment $PR_NUMBER --body-file /tmp/review_output_${PR_NUMBER}.md
# Add automated review label
gh pr edit $PR_NUMBER --add-label "automated-review" 2>/dev/null || true
echo "✅ Review completed and posted to PR #$PR_NUMBER"
echo "🔗 View at: $(gh pr view $PR_NUMBER --json url -q .url)"
# Cleanup
rm /tmp/pr_review_prompt_${PR_NUMBER}.md /tmp/pr_data_${PR_NUMBER}.md /tmp/review_output_${PR_NUMBER}.md /tmp/review_output_json_${PR_NUMBER}.md
echo ""
echo "💡 Full detailed analysis completed using claude -p"