Automated Codex reviews

.github/codex/prompts/review_common.md

@@ -0,0 +1,94 @@
+# Code Review Instructions
+
+You are an expert software engineer conducting a code review. Review the provided pull request changes and respond in JSON format according to the schema.
+
+## Context
+- PR changes: {{ .Files }}
+- Base branch: {{ .BaseBranch }}
+- Head branch: {{ .HeadBranch }}
+
+## Review Criteria
+
+### Critical Issues (Must Fix)
+
+#### 1. Security vulnerabilities
+- Hardcoded secrets/credentials
+- SQL injection risks
+- Unsafe deserialization
+- Missing input validation
+- Insecure direct object references
+
+#### 2. Bugs and logic errors
+- Null pointer exceptions
+- Off-by-one errors
+- Race conditions
+- Infinite loops
+- Incorrect error handling
+
+#### 3. Breaking changes
+- API contract violations
+- Database schema changes without migration
+- Removed public interfaces
+- Changed function signatures
+
+#### 4. Tensor/PyTorch Specific Issues
+- **Hidden type conversions**: Implicit dtype/device casts without explicit `.to()` or `.type()`
+- **Silent fallbacks**: Hidden fallback implementations or broad try/except that hide errors
+- **Batch handling**: Missing support for batch dimension in tensor operations
+- **Shape assumptions**: Hardcoded tensor shapes without validation or documentation
+- **Device mismatches**: Operations between tensors on different devices without explicit movement
+
+### Warning Issues (Should Fix)
+
+#### 1. Code quality
+- Code duplication (DRY violations)
+- Magic numbers/strings
+- Overly complex functions (cyclomatic complexity > 10)
+- Deep nesting (more than 3 levels)
+- Long functions (> 50 lines)
+
+#### 2. Best practices
+- Missing error handling
+- Poor naming conventions
+- Inconsistent code style
+- Lack of comments for complex logic
+- Premature optimization
+
+#### 3. Testing gaps
+- Missing unit tests for new functionality
+- Missing edge cases in tests
+- Low test coverage for changes
+- Tests that don't actually assert anything
+
+#### 4. PyTorch Best Practices
+- **Public APIs**: Data-related public APIs should use `torch.Tensor` as base type
+- **Documentation**: Tensor shapes missing in docstrings (especially for batch-first semantics)
+- **InputData/OutputData**: Missing tests for batch mode + multiple dtypes
+- **Device handling**: Implicit assumptions about device placement
+- **Gradient flow**: Missing `requires_grad` considerations for trainable parameters
+
+### Suggestions (Nice to Have)
+
+#### 1. Performance
+- Inefficient algorithms
+- Unnecessary database queries
+- Missing caching opportunities
+- Memory leaks
+
+#### 2. Maintainability
+- Consider extracting reusable components
+- Add more comprehensive documentation
+- Improve separation of concerns
+- Follow SOLID principles more closely
+
+#### 3. Future-proofing
+- Add deprecation notices for removed features
+- Consider backward compatibility
+- Plan for scalability
+- Add feature flags for risky changes
+
+#### 4. PyTorch Optimizations
+- Use in-place operations where appropriate
+- Consider `torch.jit.script` for performance-critical paths
+- Batch operations instead of loops
+- Use appropriate tensor layouts (contiguous vs strided)

.github/codex/review_schema.json

@@ -0,0 +1,101 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "type": "object",
+  "required": ["approved", "summary", "issues", "strengths", "metrics"],
+
+  "properties": {
+    "approved": {
+      "type": "boolean",
+      "description": "set `true` if the PR is accepted to merge"
+    },
+
+    "summary": {
+      "type": "string",
+      "description": "make short comprehensive summary of the review"
+    },
+
+    "related_issue": {
+      "type": "integer",
+      "description": "print the number of the related issue"
+    },
+
+    "issues": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "required": ["severity", "category", "file", "title", "description", "suggestion"],
+
+        "properties": {
+          "severity": {
+            "type": "string",
+            "enum": ["critical", "warning", "suggestion"]
+          },
+
+          "category": {
+            "type": "string",
+            "enum": [
+              "security", "bug", "tensor", "api", "testing",
+              "quality", "performance", "documentation"
+            ]
+          },
+
+          "subcategory": {
+            "type": "string",
+            "enum": [
+              "device_mismatch", "dtype_conversion", "batch_handling",
+              "hidden_fallback", "missing_tests", "shape_documentation"
+            ]
+          },
+
+          "file": {
+            "type": "string",
+            "pattern": "^[\\w\\-/\\.]+\\.(py|js|ts|java|cpp)$"
+          },
+
+          "line": {
+            "type": "integer",
+            "minimum": 1
+          },
+
+          "title": {
+            "type": "string",
+            "maxLength": 100
+          },
+
+          "description": {
+            "type": "string",
+            "maxLength": 1000
+          },
+
+          "suggestion": {
+            "type": "string",
+            "maxLength": 1000
+          },
+
+          "code_example": {
+            "type": "string"
+          }
+        }
+      }
+    },
+
+    "metrics": {
+      "type": "object",
+      "required": ["files_reviewed", "critical_count", "warning_count", "suggestion_count"],
+
+      "properties": {
+        "files_reviewed": { "type": "integer", "minimum": 0 },
+        "critical_count": { "type": "integer", "minimum": 0 },
+        "warning_count": { "type": "integer", "minimum": 0 },
+        "suggestion_count": { "type": "integer", "minimum": 0 }
+        }
+      }
+    },
+
+    "strengths": {
+      "type": "array",
+      "items": { "type": "string" },
+      "description": "provide advantages of the reviewed PR"
+    }
+  }
+}

.github/workflows/codex-auto-review.yml

@@ -0,0 +1,132 @@
+name: Automated Codex Reviews
+
+on:
+  pull_request:
+    branches: [ master ]
+    types: [ opened, synchronize, ready_for_review ]
+
+jobs:
+
+  codex-review:
+
+    runs-on: ubuntu-latest
+
+    permissions:
+      contents: read
+      pull-requests: write
+
+    steps:
+      - uses: actions/checkout@v5
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+
+      - name: Fetch refs
+        run: |
+          git fetch origin ${{ github.event.pull_request.base.ref }} --depth=1
+          git checkout origin/${{ github.event.pull_request.base.ref }} -- .github/codex/
+          git diff origin/${{ github.event.pull_request.base.ref }}...HEAD > pr.diff
+
+      - name: Set gh CLI
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        uses: cli/gh@v2
+
+      - name: Select issue
+        id: select
+        run: |
+          BODY=$(gh pr view ${{ github.event.pull_request.number }} --json body -q .body)
+
+          ISSUE=$(echo "$BODY" | grep -oE '#[0-9]+' | head -1 | tr -d '#')
+
+          if [ -z "$ISSUE" ]; then
+            ISSUE=$(gh pr view ${{ github.event.pull_request.number }} --json labels -q '.labels[].name' | \
+              grep '^issue/' | head -1 | cut -d'/' -f2)
+          fi
+
+          if [ -n "$ISSUE" ] && [ -f ".github/codex/prompts/review_${ISSUE}.md" ]; then
+            echo "issue_number=$ISSUE" >> $GITHUB_OUTPUT
+            echo "prompt_file=.github/codex/prompts/review_${ISSUE}.md" >> $GITHUB_OUTPUT
+          else
+            echo "issue_number=" >> $GITHUB_OUTPUT
+            echo "prompt_file=.github/codex/prompts/review_common.md" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Run Codex Review
+        id: review
+        uses: openai/codex-action@v1
+        with:
+          openai-api-key: ${{ secrets.OPENAI_API_KEY }}
+          prompt: | 
+            This is PR #${{ github.event.pull_request.number }} for ${{ github.repository }}.
+            ========
+            Read the instructions for the current review below:
+            ----
+            $(cat ${{ steps.select.outputs.prompt_file }})
+            ========
+            Carefully analyze changes introduced by the PR below: 
+            ----
+            $(cat pr.diff)
+            ========
+            Pull request title and body:
+            ----
+            ${{ github.event.pull_request.title }}
+            ${{ github.event.pull_request.body }}
+          output-file: codex-output.json
+          output-schema-file: .github/codex/review_schema.json
+          sandbox: workspace-write
+          safety-strategy: drop-sudo
+
+      - name: Validate JSON
+        id: validate
+        run: |
+          if [ ! -f codex-output.json ]; then
+            echo "approved=false" >> $GITHUB_OUTPUT
+            echo "No output file"
+            exit 0
+          fi
+
+          jq empty codex-output.json || exit 1
+
+          APPROVED=$(jq -r '.approved // false' codex-output.json)
+          echo "approved=$APPROVED" >> $GITHUB_OUTPUT
+
+          if [ "$APPROVED" = "false" ]; then
+            ISSUES=$(jq -r '.issues // [] | length' codex-output.json)
+            echo "FEEDBACK=Review failed: $ISSUES issues found. See codex-output.json" >> $GITHUB_ENV
+          fi
+
+      - name: Validate against schema
+        continue-on-error: false
+        run: |
+          npm install -g ajv-cli
+          ajv validate -s .github/codex/review_schema.json -d codex-output.json
+
+      - name: Post feedback
+        uses: actions/github-script@v7
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          script: |
+            const fs = require('fs');
+            let output = 'No output file generated';
+            const status = '${{ steps.validate.outputs.approved }}' === 'true' ? 'Approved' : 'Failed';
+
+            if (fs.existsSync('codex-output.json')) {
+              output = fs.readFileSync('codex-output.json', 'utf8');
+            }
+
+            const issue = '${{ steps.select.outputs.issue_number }}';
+            const prompt = '${{ steps.select.outputs.prompt_file }}';
+            const approved = '${{ steps.validate.outputs.approved }}';
+
+            await github.rest.issues.createComment({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: context.payload.pull_request.number,
+              body: `## Codex Review \n**Status**: ${status}\n**Issue**: ${issue || 'N/A'}\n**Prompt**: ${prompt || 'N/A'}\n\n**Output**:\n\`\`\`json\n${output}\n\`\`\``
+            });
+
+      - name: Block merge
+        if: steps.validate.outputs.approved != 'true'
+        run: |
+          echo "Codex review failed - blocking merge"
+          exit 1