lucianfialho Review PR with Specialists
Run targeted code reviews on the current PR using specialist skills.
Usage
/review-pr frontend — React/Next.js, components, a11y, performance
/review-pr backend — API design, security, DB patterns, error handling
/review-pr security — Security-focused checklist (OWASP, secrets, auth)
/review-pr ux — UX heuristics, accessibility, visual hierarchy
/review-pr all — Run all applicable specialists in parallel (same as /batch-review)
If no argument is provided, defaults to all which runs specialists in parallel using subagents (delegates to /batch-review).
Process
1. Get the PR diff
Detect the current PR from the branch:
gh pr view --json number,title,body,files
gh pr diff
If not on a PR branch, ask the user for a PR number.
2. Dispatch to specialist
Based on the argument, load the corresponding specialist skill and provide it with:
- The full PR diff
- The PR description for context
- The review output format (below)
| Argument | Specialist Skill | Focus Area |
|---|---|---|
frontend |
/frontend-dev |
React/Next.js patterns, components, a11y, responsiveness, performance, state management |
backend |
/backend-dev |
API design, database, auth, error handling, input validation, security |
security |
/code-reviewer |
Security-only: injection, XSS, CSRF, secrets, auth bypass, dependencies |
ux |
/ux-designer |
Nielsen's heuristics, accessibility, visual hierarchy, interaction design, empty states |
all |
All of the above | Run each sequentially, combine output |
3. Review output format
Every specialist MUST output findings in this consistent format:
## <Specialist Name> Review
### Summary
<1-2 sentence overview of the review>
### Findings
#### 🔴 Critical
Issues that must be fixed before merge.
- **`file.tsx:42`** — <description>
**Fix:** <concrete suggestion>
#### 🟡 Warning
Issues that should be fixed but aren't blocking.
- **`file.ts:15`** — <description>
**Fix:** <concrete suggestion>
#### 💡 Suggestion
Improvements that would make the code better.
- **`file.ts:88`** — <description>
**Fix:** <concrete suggestion>
### Verdict
<APPROVE | REQUEST_CHANGES | COMMENT>
<1 sentence justification>
4. For review all
Delegate to the /batch-review skill, which runs specialists in parallel using subagents:
- Analyze the PR diff to categorize changed files
- Select applicable specialists based on file types
- Dispatch up to 3 subagents in parallel
- Produce a unified review with combined verdict table
This is equivalent to running /batch-review directly.
Rules
- ALWAYS fetch the actual PR diff — never review from memory
- ALWAYS use the consistent output format above
- NEVER auto-approve — always provide at least the verdict
- If a specialist finds no issues, say so explicitly: "No issues found in "
- For
review all, include a final combined verdict based on the worst finding across all specialists - Be thorough but concise — explain WHY something is an issue
- Provide concrete fix suggestions, not just complaints