Interactive code review plugin with multi-agent analysis and conversational walkthrough
npx claudepluginhub benyaminsalimi/review-nowInteractive code review with multi-agent analysis. Two commands: /review-now:pr for pull requests, /review-now:uncommitted for local changes
Interactive code review plugin for Claude Code with multi-agent analysis and conversational walkthrough.
A Claude Code plugin that transforms code review from reading static reports into having interactive conversations. Walk through findings one at a time, apply fixes inline, ask questions, and move fast through your review queue.
Built for two scenarios:
# Add the marketplace
claude plugin marketplace add benyaminsalimi/review-now
# Install the plugin
claude plugin install review-now@review-now-marketplace
git clone https://github.com/benyaminsalimi/review-now.git
cd review-now
claude plugin marketplace add ./
claude plugin install review-now@review-now-marketplace
/review-now:pr <source-branch> <target-branch>Interactive PR review with automatic commits for each fix.
/review-now:pr feature/user-auth main
Features:
Interactive Options:
/review-now:uncommittedInteractive review of all uncommitted changes (no arguments needed).
/review-now:uncommitted
Features:
Interactive Options:
Pre-flight checks
|
Discover custom guidelines (REVIEW_GUIDELINES.md, AGENTS.md)
|
Gather changes (diff, files, commits)
|
Launch 3 agents in parallel
|- Security review agent (Anthropic guidelines)
|- General review agent (Codex patterns)
|- Frontend/Angular review agent (Angular official docs)
|
Deduplicate findings
|
Validate with lightweight agents (filter false positives)
|
Interactive walkthrough
|- Apply fix / Show solution / Custom fix / Ask question / Skip
|- (PR mode: auto-commits after each fix)
|
Final summary with verdict
The plugin automatically picks its strategy:
REVIEW_GUIDELINES.md in repo root? | Strategy |
|---|---|
| Yes | Reviews using your custom guidelines. No agents. |
| No | Dispatches 3 specialized agents in parallel. |
Based on Anthropic's security review guidelines. Focuses on high-confidence vulnerabilities: SQL injection, authentication bypasses, hardcoded secrets, crypto issues, XSS, and data exposure.
Adapted from OpenAI Codex review patterns. Focuses on correctness, maintainability, logic errors, error handling, and code quality.
Built from Angular's official LLM context. Focuses on signal-based architecture, change detection, modern template syntax, component patterns, and performance.
All agents use P0-P3 scale:
| Priority | Meaning |
|---|---|
| P0 | Drop everything. Blocking release/operations. |
| P1 | Urgent. Fix in the next cycle. |
| P2 | Normal. Fix eventually. |
| P3 | Low. Nice to have. |
| Verdict | Condition |
|---|---|
| REQUEST CHANGES | Any P0 or P1 findings |
| NEEDS DISCUSSION | Any P2 findings with confidence > 0.85 |
| APPROVE | Everything else |
Place a REVIEW_GUIDELINES.md file in your repo root to define custom review rules. The plugin will use your guidelines instead of running agents.
Place AGENTS.md files in any directory to define scoped rules. The plugin discovers all AGENTS.md files and checks changed files against rules in their scope (directory and subdirectories).