Pre-PR Review Skill

Focused pre-PR review targeting patterns that GitHub Copilot code review catches. NOT a generic code review — use /code-review for that.

Purpose

Catch the 9 recurring pattern categories (contract mismatch, null safety, error handling, scalability, security, logic errors, architecture drift, test quality, naming conflicts) BEFORE creating a PR, reducing review cycles.

When to Use

• Before gh pr create — run /review-pr first
• After significant changes to a feature branch
• When touching API routes AND frontend consumers in same PR

Workflow

Step 1: Detect Scope

# Detect base branch
BASE=$(git merge-base --fork-point main HEAD 2>/dev/null || echo "main")

# Get changed files
CHANGED_FILES=$(git diff --name-only --diff-filter=ACMR "${BASE}...HEAD")

# Count and report scope
FILE_COUNT=$(echo "$CHANGED_FILES" | grep -c . || echo 0)

If no changed files detected, report "No changes found" and exit.

Step 2: Automated Pattern Check (Zero AI Cost)

Run the mechanical pattern checker first:

${CLAUDE_HOME:-.claude}/scripts/code-pattern-check.sh --diff-base "$BASE" --json

Parse the JSON output. Report P1/P2 counts immediately.

Step 3: Read Changed Files

Read the FULL content of each changed file (not just diff). Cross-file analysis requires full context.

Prioritize:

1. API route files (**/api/**/route.ts, **/api/**/route.py)
2. Service/hook files that consume APIs (**/services/*.ts, **/hooks/use*.ts)
3. Type/interface definitions (**/types/*.ts, **/interfaces/*.ts)
4. Test files (**/*.test.ts, **/*.spec.ts, **/test_*.py)

Step 4: AI Cross-File Analysis

Analyze for patterns that code-pattern-check.sh CANNOT catch:

4a. Contract Mismatch (P1)

• Compare API route return types with frontend fetch/hook expectations
• Check Prisma schema field types vs API DTO fields
• Verify request body types match API parameter expectations

4b. Logic Errors (P1)

• Wrong field used in object mapping (e.g., user.name vs user.displayName)
• Regex patterns that match too broadly or too narrowly
• Off-by-one in array/string operations
• Transaction ordering (dependent ops outside $transaction)

4c. Architecture Drift (P2)

• Read active ADRs: ls docs/adr/*.md and check Status: Accepted
• Compare new code patterns against ADR decisions
• Flag contradictions (e.g., localStorage usage when ADR mandates Zustand)

4d. Test Quality (P2)

• Check test files cover edge cases, not just happy path
• Flag brittle assertions (toHaveBeenCalledTimes on implementation details)
• Verify error paths are tested
• Check for shared mutable state between tests

Step 5: Generate Report

Output a structured report with this format:

## Pre-PR Review: {branch_name}

**Scope**: {file_count} files changed | Base: {base_branch}

### Automated Findings (code-pattern-check.sh)

| Severity | Check                  | Count |
| -------- | ---------------------- | ----- |
| P1       | unguarded_json_parse   | 2     |
| P2       | missing_error_boundary | 1     |

### AI Analysis Findings

| #   | Severity | Category           | File                   | Line | Issue                                             |
| --- | -------- | ------------------ | ---------------------- | ---- | ------------------------------------------------- |
| 1   | P1       | contract_mismatch  | src/api/users/route.ts | 45   | Returns `{name}` but hook expects `{displayName}` |
| 2   | P2       | architecture_drift | src/store/auth.ts      | 12   | Uses localStorage, ADR-007 requires Zustand       |

### Recommendations

1. **[P1]** Fix contract mismatch in user API...
2. **[P2]** Migrate auth storage to Zustand per ADR-007...

### Summary

- **P1 (must fix)**: {count}
- **P2 (should fix)**: {count}
- **P3 (consider)**: {count}
- **Verdict**: {READY | NOT READY} for PR

What This Skill Does NOT Do

• Generic code quality review (use /code-review)
• Style/formatting checks (use ESLint/Prettier)
• Performance profiling (use /performance)
• Security audit (use /security-audit)

Reference

• Pattern knowledge base: ${CLAUDE_HOME:-.claude}/reference/copilot-patterns.md
• Automated checker: ${CLAUDE_HOME:-.claude}/scripts/code-pattern-check.sh
• Copilot feedback digest: ${CLAUDE_HOME:-.claude}/scripts/copilot-review-digest.sh
• Full code review: /code-review skill

Integration with Thor

Thor Gate 4b runs code-pattern-check.sh automatically during validation. This skill adds AI analysis on top for pre-PR use.