Claude Code Plugins

Community-maintained marketplace

Feedback

Security-first PR review checklist for this repo. Use when reviewing diffs/PRs, especially changes involving auth, networking, sensitive data, or dependency/lockfile updates. Focus on secret/PII leakage risk, supply-chain risk (npm + node_modules inspection), cross-platform architecture (extension/mobile/desktop/web), and React performance (hooks + re-render hotspots). Avoid UI style nitpicks. PR Review.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name pr-review
description Security-first PR review checklist for this repo. Use when reviewing diffs/PRs, especially changes involving auth, networking, sensitive data, or dependency/lockfile updates. Focus on secret/PII leakage risk, supply-chain risk (npm + node_modules inspection), cross-platform architecture (extension/mobile/desktop/web), and React performance (hooks + re-render hotspots). Avoid UI style nitpicks. PR Review.
allowed-tools Read, Grep, Glob, Bash

Secure PR Review

Follow this workflow when reviewing code changes. Prioritize security > correctness > architecture > performance.

Review scope (base branch)

  • Review scope: treat x as the base (main) branch. Always review the PR as the diff between the current branch (HEAD) and x (i.e., changes introduced by this branch vs x).
  • Use PR semantics when generating the diff: git fetch origin && git diff origin/x...HEAD (triple-dot) to review only the changes introduced on this branch relative to x.

0) Scope the change

  • Identify what changed (files, modules, entrypoints, routes/screens).
  • Identify risk areas: auth flows, signing/keys, networking, analytics, storage, dependency updates.

1) Secrets / PII / privacy (MUST)

  • Do not allow logs/telemetry/error reports to include: mnemonics/seed phrases, private keys, signing payloads, API keys, tokens, cookies, session IDs, addresses tied to identity, or any PII.
  • Inspect all “exfil paths”: console.*, logging utilities, analytics SDKs, error reporting, network requests, and persistence:
    • Web: localStorage / IndexedDB
    • RN: AsyncStorage / secure storage
    • Desktop: filesystem / keychain / sqlite
  • If any potential leak exists, explicitly document:
    • source (what sensitive data),
    • sink (where it goes),
    • trigger (when it happens),
    • impact (who/what is exposed),
    • fix (concrete remediation).

2) AuthN / AuthZ (MUST)

  • Verify authentication middleware/guards wrap every protected route and cannot be bypassed.
  • Verify authorization checks (roles/permissions) are correct and consistent.
  • Verify server/client trust boundaries: never trust client input for authorization decisions.

3) Dependency & supply-chain security (HIGHEST PRIORITY)

If package.json / lockfiles changed, you MUST do all of the following:

3.1 Enumerate changes

  • List every added/updated/removed dependency with name + from→to version and the reason (if stated in PR).

3.2 Quick ecosystem risk check (before approve)

  • For each changed package:
    • check for recent maintainer/ownership changes, suspicious release cadence, known advisories/CVEs, typosquatting risk.
    • if your environment supports it, run commands like: npm view <pkg> time maintainers repository dist.tarball.

3.3 Source inspection (node_modules) — REQUIRED when risk is non-trivial

  • Inspect the dependency’s node_modules/<pkg>/package.json and entrypoints (main / module / exports).
  • Grep for high-risk behavior (examples; expand as needed):
    • outbound/network: fetch(, axios, XMLHttpRequest, http, https, ws, request, net, dns
    • dynamic execution: eval, new Function, dynamic require, remote script loading
    • install hooks: postinstall, preinstall, install, binary downloads
    • privilege access: filesystem, clipboard, keychain/keystore, environment variables
  • Treat as HIGH RISK and block approval unless justified + isolated:
    • any telemetry / remote config fetch / unexpected outbound requests
    • any dynamic execution or install-time script behavior
    • any access to sensitive storage or wallet-related data

3.4 React Native native-layer inspection (REQUIRED for RN libraries)

  • For React Native dependencies (or any package with native bindings: .podspec, ios/, android/, react-native.config.js, TurboModules/Fabric):
    • Inspect iOS/Android native sources for security + performance.
    • Confirm there are no unexpected outbound requests, no telemetry/upload without explicit product intent, and no access to wallet secrets/private keys/seed data.
    • If necessary, drill into third-party native dependencies:
      • iOS: CocoaPods / Pods/ sources, vendored frameworks, build scripts
      • Android: Gradle/Maven artifacts, JNI/native libs, build-time tasks
    • Treat any hidden network behavior, dynamic loading, install/build scripts, or obfuscated native code as HIGH RISK unless explicitly justified and isolated.

4) Mandatory callout when node_modules performs outbound requests

If node_modules code performs any outbound network/API request (directly or indirectly), call it out clearly in the review:

  • exact call site (file path + function)
  • destination (full URL/host)
  • payload fields (what data is sent)
  • headers/auth (tokens/cookies/identifiers)
  • trigger conditions (when/how it runs)
  • cross-platform impact (extension/mobile/desktop/web)

4.1 Extension manifest permissions changes (HIGHEST PRIORITY)

  • If manifest.json (permissions, host_permissions, optional_permissions) changes:
    • Call it out prominently as the top review item.
    • Enumerate added/removed permissions and explain what new capabilities they enable.
    • Assess least-privilege: confirm the permission is strictly necessary, scoped to minimal hosts, and does not broaden data access/exfil paths.
    • Re-check data exposure surfaces introduced by the permission change (network, storage, messaging, content scripts, background/service worker).

5) Cross-platform architecture review (extension/mobile/desktop/web)

Review the implementation as a senior multi-platform architect:

  • Is the approach the simplest correct solution with good maintainability/testability?
  • Identify platform pitfalls:
    • Extension constraints (MV3/service worker lifetimes, permissions, CSP)
    • RN constraints (WebView, native modules, backgrounding)
    • Desktop (Electron security boundaries, IPC, nodeIntegration)
    • Web (CORS, storage, XSS, bundle size, runtime differences)
  • If not optimal, propose a better alternative with tradeoffs.

6) React performance (hooks + re-render hotspots)

For new/modified components:

  • Check for unnecessary re-renders from unstable references:
    • inline objects/functions passed to children
    • incorrect hook dependency arrays
    • state placed too high causing wide re-render fanout
  • Validate memoization strategy (memo, useMemo, useCallback) is correct (no stale closures / broken deps).
  • Watch for expensive work in render, list rendering issues, and missing cleanup for subscriptions/listeners.
  • Apply stricter scrutiny to new parent/child boundaries and call out any likely re-render hotspots.

7) Review output format (keep it actionable)

  • Focus on security/correctness/architecture/performance.
  • Avoid UI style / comment nitpicks unless they cause real bugs, security risk, or measurable perf regression.
  • Provide findings as:
    • Blockers (must fix)
    • High risk (strongly recommended)
    • Suggestions (nice-to-have)
    • Questions (needs clarification)

Additional resources