refactor: convert workflows to use Claude Code Action

Replace direct Anthropic API calls with Claude Code Action for AI tasks.
This allows the workflows to use the existing CLAUDE_CODE_OAUTH_TOKEN
instead of requiring a separate ANTHROPIC_API_KEY.

Changes:

spec-to-code.yml:
- ✅ Trigger remains label-based ('approved' label on issue)
- ✅ Spec extraction remains unchanged
- ✅ Code generation now uses Claude Code Action
- ✅ Provides detailed prompt with rules and requirements
- ✅ Claude Code reads specs, rules, and generates implementations
- ✅ Self-review loop integrated into Claude Code execution
- ✅ Automatic commit and PR creation

quality-check.yml:
- ✅ Trigger remains workflow_run based
- ✅ Preview image download from GCS unchanged
- ✅ Quality evaluation now uses Claude Code Action
- ✅ Claude Code views images (Vision) and evaluates against specs
- ✅ Uses gh CLI to comment on PR and add labels
- ✅ Scores each implementation (0-100, ≥85 to pass)

Benefits:
- No separate API key needed
- Uses existing Claude Code Max subscription
- Integrated with Claude Code ecosystem
- Cleaner, more maintainable code
- Claude Code handles commits automatically
- Better error handling and retries

Hybrid Approach:
- Label triggers → Automation starts
- Claude Code → AI-powered tasks (generation, evaluation)
- Regular Actions → Infrastructure (tests, GCS upload, PR creation)

This enables full automation using only CLAUDE_CODE_OAUTH_TOKEN
and GCS_CREDENTIALS secrets.

Author: claude

.github/workflows/quality-check.yml

@@ -14,19 +14,13 @@ jobs:
       contents: read
       pull-requests: write
       issues: write
+      id-token: write
 
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
         with:
-          python-version: '3.12'
-
-      - name: Install dependencies
-        run: |
-          pip install anthropic requests
+          fetch-depth: 0
 
       - name: Download preview metadata
         uses: actions/download-artifact@v4
@@ -46,10 +40,12 @@ jobs:
           PR_NUMBER=$(jq -r '.pr_number' preview_metadata.json)
           BUCKET=$(jq -r '.bucket' preview_metadata.json)
           BASE_PATH=$(jq -r '.base_path' preview_metadata.json)
+          CHANGED_FILES=$(jq -r '.changed_files[]' preview_metadata.json | tr '\n' ' ')
 
           echo "pr_number=$PR_NUMBER" >> $GITHUB_OUTPUT
           echo "bucket=$BUCKET" >> $GITHUB_OUTPUT
           echo "base_path=$BASE_PATH" >> $GITHUB_OUTPUT
+          echo "changed_files=$CHANGED_FILES" >> $GITHUB_OUTPUT
 
       - name: Setup Google Cloud authentication
         uses: google-github-actions/auth@v2
@@ -59,255 +55,48 @@ jobs:
       - name: Download preview images
         run: |
           mkdir -p preview_images
-          gsutil -m cp -r "gs://${{ steps.metadata.outputs.bucket }}/${{ steps.metadata.outputs.base_path }}/*" preview_images/
-
-      - name: Run quality evaluation
-        id: quality_eval
-        env:
-          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-        run: |
-          cat > quality_evaluator.py <<'EOF'
-          import os
-          import sys
-          import json
-          import base64
-          from pathlib import Path
-          import anthropic
-
-          def load_spec(spec_id: str) -> str:
-              """Load spec content"""
-              spec_path = Path(f"specs/{spec_id}.md")
-              if not spec_path.exists():
-                  return ""
-              return spec_path.read_text()
-
-          def encode_image(image_path: str) -> str:
-              """Encode image as base64"""
-              with open(image_path, 'rb') as f:
-                  return base64.standard_b64encode(f.read()).decode('utf-8')
-
-          def evaluate_plot(spec_id: str, library: str, variant: str, image_path: str) -> dict:
-              """Evaluate a single plot implementation"""
-              client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))
-
-              # Load spec
-              spec_content = load_spec(spec_id)
-              if not spec_content:
-                  return {"error": f"Spec not found: {spec_id}"}
-
-              # Encode image
-              image_data = encode_image(image_path)
-
-              # Build evaluation prompt
-              prompt = f"""You are an expert data visualization reviewer.
-
-          # Task
-          Evaluate this plot implementation against its specification.
-
-          # Specification
-          {spec_content}
-
-          # Implementation Details
-          - Library: {library}
-          - Variant: {variant}
-
-          # Quality Evaluation Criteria
-          Review the plot image against EACH quality criterion listed in the specification.
-
-          # Response Format
-          Provide your evaluation in the following JSON format:
-
-          {{
-            "overall_score": <0-100>,
-            "verdict": "<PASS|FAIL>",
-            "criteria_results": [
-              {{
-                "criterion": "<criterion text>",
-                "status": "<PASS|FAIL>",
-                "comment": "<brief explanation>"
-              }}
-            ],
-            "strengths": ["<strength 1>", "<strength 2>"],
-            "improvements": ["<improvement 1>", "<improvement 2>"],
-            "summary": "<2-3 sentence overall assessment>"
-          }}
-
-          # Scoring Guidelines
-          - 90-100: Excellent - All criteria met, production ready
-          - 85-89: Good - Minor issues, acceptable
-          - 75-84: Needs improvement - Some criteria failed
-          - <75: Rejected - Major issues, regeneration required
-
-          # Instructions
-          1. Review the image carefully
-          2. Check EACH quality criterion from the spec
-          3. Provide specific, actionable feedback
-          4. Be objective and constructive
-
-          Provide ONLY the JSON response, no additional text."""
-
-              # Call Claude with vision
-              response = client.messages.create(
-                  model="claude-sonnet-4-20250514",
-                  max_tokens=2000,
-                  messages=[{
-                      "role": "user",
-                      "content": [
-                          {
-                              "type": "image",
-                              "source": {
-                                  "type": "base64",
-                                  "media_type": "image/png",
-                                  "data": image_data
-                              }
-                          },
-                          {
-                              "type": "text",
-                              "text": prompt
-                          }
-                      ]
-                  }]
-              )
-
-              # Parse response
-              response_text = response.content[0].text
-
-              # Extract JSON (handle markdown code blocks)
-              if "```json" in response_text:
-                  response_text = response_text.split("```json")[1].split("```")[0].strip()
-              elif "```" in response_text:
-                  response_text = response_text.split("```")[1].split("```")[0].strip()
-
-              try:
-                  result = json.loads(response_text)
-                  return result
-              except json.JSONDecodeError:
-                  return {"error": "Failed to parse response", "raw": response_text}
-
-          # Main execution
-          if __name__ == "__main__":
-              results = []
-
-              # Process all preview images
-              preview_dir = Path("preview_images")
-              for img_file in preview_dir.glob("*.png"):
-                  # Parse filename: {spec_id}_{library}_{variant}.png
-                  parts = img_file.stem.split('_')
-                  if len(parts) >= 3:
-                      spec_id = '_'.join(parts[:-2])
-                      library = parts[-2]
-                      variant = parts[-1]
+          gsutil -m cp -r "gs://${{ steps.metadata.outputs.bucket }}/${{ steps.metadata.outputs.base_path }}/*" preview_images/ || echo "No images to download"
 
-                      print(f"🔍 Evaluating: {spec_id}/{library}/{variant}")
-                      result = evaluate_plot(spec_id, library, variant, str(img_file))
-                      result["spec_id"] = spec_id
-                      result["library"] = library
-                      result["variant"] = variant
-                      results.append(result)
-
-              # Save results
-              with open("quality_results.json", "w") as f:
-                  json.dump(results, f, indent=2)
-
-              # Print summary
-              print("\n" + "="*60)
-              print("Quality Evaluation Results")
-              print("="*60)
-              for r in results:
-                  score = r.get("overall_score", 0)
-                  verdict = r.get("verdict", "UNKNOWN")
-                  print(f"{r['spec_id']}/{r['library']}/{r['variant']}: {score}/100 - {verdict}")
-              print("="*60)
-          EOF
-
-          python quality_evaluator.py
-
-      - name: Parse quality results
-        id: results
-        run: |
-          if [ ! -f quality_results.json ]; then
-            echo "No quality results found"
-            exit 0
-          fi
-
-          # Calculate overall verdict
-          TOTAL=$(jq '. | length' quality_results.json)
-          PASSED=$(jq '[.[] | select(.verdict == "PASS")] | length' quality_results.json)
-          FAILED=$((TOTAL - PASSED))
-
-          MIN_SCORE=$(jq '[.[] | .overall_score] | min' quality_results.json)
-
-          echo "total=$TOTAL" >> $GITHUB_OUTPUT
-          echo "passed=$PASSED" >> $GITHUB_OUTPUT
-          echo "failed=$FAILED" >> $GITHUB_OUTPUT
-          echo "min_score=$MIN_SCORE" >> $GITHUB_OUTPUT
-
-          # Overall verdict: PASS if all pass and min_score >= 85
-          if [ $FAILED -eq 0 ] && [ $MIN_SCORE -ge 85 ]; then
-            echo "overall_verdict=PASS" >> $GITHUB_OUTPUT
-          else
-            echo "overall_verdict=FAIL" >> $GITHUB_OUTPUT
-          fi
-
-      - name: Generate quality report
-        if: steps.results.outputs.total != ''
-        run: |
-          cat > quality_report.md <<'EOF'
-          ## 🤖 Quality Check Results
-
-          **Overall Verdict:** ${{ steps.results.outputs.overall_verdict }}
-
-          **Summary:**
-          - Total implementations: ${{ steps.results.outputs.total }}
-          - Passed: ${{ steps.results.outputs.passed }}
-          - Failed: ${{ steps.results.outputs.failed }}
-          - Minimum score: ${{ steps.results.outputs.min_score }}/100
-
-          ### Detailed Results
-
-          EOF
-
-          # Add detailed results for each implementation
-          jq -r '.[] | "#### \(.spec_id) / \(.library) / \(.variant)\n\n**Score:** \(.overall_score)/100 | **Verdict:** \(.verdict)\n\n**Summary:** \(.summary)\n\n**Strengths:**\n\(.strengths | map("- " + .) | join("\n"))\n\n**Improvements Needed:**\n\(.improvements | map("- " + .) | join("\n"))\n\n---\n"' quality_results.json >> quality_report.md
-
-          echo "" >> quality_report.md
-          echo "*Generated by [quality-check workflow](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})*" >> quality_report.md
-
-      - name: Comment on PR with quality results
-        if: steps.results.outputs.total != ''
-        uses: actions/github-script@v7
-        with:
-          script: |
-            const fs = require('fs');
-            const report = fs.readFileSync('quality_report.md', 'utf8');
-
-            await github.rest.issues.createComment({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              issue_number: ${{ steps.metadata.outputs.pr_number }},
-              body: report
-            });
-
-      - name: Add labels based on verdict
-        if: steps.results.outputs.total != ''
-        uses: actions/github-script@v7
+      - name: Quality evaluation with Claude Code
+        uses: anthropics/claude-code-action@v1
         with:
-          script: |
-            const verdict = '${{ steps.results.outputs.overall_verdict }}';
-            const prNumber = ${{ steps.metadata.outputs.pr_number }};
-
-            if (verdict === 'PASS') {
-              await github.rest.issues.addLabels({
-                owner: context.repo.owner,
-                repo: context.repo.repo,
-                issue_number: prNumber,
-                labels: ['quality-approved']
-              });
-            } else {
-              await github.rest.issues.addLabels({
-                owner: context.repo.owner,
-                repo: context.repo.repo,
-                issue_number: prNumber,
-                labels: ['quality-check-failed']
-              });
-            }
+          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+          prompt: |
+            TASK: Evaluate plot implementations against specifications
+
+            PR: #${{ steps.metadata.outputs.pr_number }}
+            Preview images location: preview_images/
+
+            Instructions:
+            1. List all PNG files in preview_images/ directory
+            2. For each preview image:
+               a. Parse filename to extract: spec_id, library, variant
+               b. Read corresponding spec file: specs/{spec_id}.md
+               c. View the preview image
+               d. Evaluate against quality criteria in spec
+
+            3. For each implementation, check:
+               - Does it meet ALL quality criteria listed in spec?
+               - Are visual elements clear and readable?
+               - Are colors appropriate and accessible?
+               - Is the layout well-structured?
+               - Score: 0-100 (≥85 to pass)
+
+            4. Generate quality report with:
+               - Overall verdict (PASS if all ≥85, FAIL otherwise)
+               - Score for each implementation
+               - Specific feedback for each quality criterion
+               - Strengths and improvements needed
+
+            5. Use gh CLI to:
+               - Post quality report as comment on PR #${{ steps.metadata.outputs.pr_number }}
+               - Add label "quality-approved" if PASS
+               - Add label "quality-check-failed" if FAIL
+
+            Scoring Guidelines:
+            - 90-100: Excellent - All criteria met, production ready
+            - 85-89: Good - Minor issues, acceptable
+            - 75-84: Needs improvement
+            - <75: Rejected - Major issues
+
+            Be objective and constructive in your feedback.