security-review (getsentry/skills): a skill that reviews code for exploitable vulnerabilities

Sentry's skill makes Claude review code for exploitable security flaws and report only confident findings, with a list of items to verify.

Installs: 22K; on skills.sh
Stars: 1.1K; on GitHub
Files: 22; besides SKILL.md
Updated: Oct 2026; last repo commit

The skill sets a strict rule for Claude: research the whole codebase, report only on the file or diff you gave it, and flag only issues where attacker-controlled input is confirmed. Pattern matches alone do not count. Claude detects the type of code, loads the matching reference file, traces where each value comes from, and checks whether the framework or upstream validation already protects it.

It triggers on requests such as "security review", "find vulnerabilities", "audit security" or "OWASP review". Findings come back with severity (Critical to Low), file and line, impact, evidence and a fix. Medium-confidence items go under "Needs verification" and low-confidence ones are not reported. If nothing is found, it says no high-confidence vulnerabilities were identified.

Dedicated guides cover Python (Django, Flask, FastAPI), JavaScript (Node, Express, React, Vue, Next.js) and Docker. For other languages, SKILL.md says to rely on the general reference files.

Good for

Developers who want a security pass on a diff or file before merging
Python and JavaScript teams, the two languages with dedicated guides
Reviewers tired of scanner noise who want findings with evidence and a fix

Install security-review

Claude Code

Copy this prompt.

Install the security-review skill in Claude Code. Run this command:

npx skills add https://github.com/getsentry/skills --skill security-review

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Security code review for vulnerabilities. Use when asked to "security review", "find vulnerabilities", "check for security issues", "audit security", "OWASP review", or review code for injection, XSS, authentication, authorization, cryptography issues. Provides systematic review with confidence-based reporting.

Sections in SKILL.md

Scope: Research vs. Reporting
Confidence Levels
Do Not Flag
General Rules
Server-Controlled Values (NOT Attacker-Controlled)
Framework-Mitigated Patterns
Review Process
1. Detect Context
2. Load Language Guide
3. Load Infrastructure Guide (if applicable)
4. Research Before Flagging
5. Verify Exploitability
6. Report HIGH Confidence Only
Severity Classification
Quick Patterns Reference
Always Flag (Critical)

Key facts

Author: Sentry (getsentry)
License: Skill folder LICENSE (OWASP-derived material, CC BY-SA 4.0); repo LICENSE Apache 2.0
Repository: getsentry/skills
Installs on skills.sh: 22,035 as of 2026-11-02
GitHub stars: 1,107 as of 2026-11-02
Last repo commit: 2026-10-26
Files in the skill folder: 23 files including SKILL.md (17 reference guides, 2 language guides, 1 Docker guide, SPEC.md and LICENSE)
skills.sh audits: Gen Agent Trust Hub pass, Socket fail, Snyk pass
Allowed tools: Read Grep Glob Bash Task
Reference material: Based on the OWASP Cheat Sheet Series

What it includes

Picks reference files by code type
Maps API routes, templates, file handling, crypto, deserialization, external requests, config and CI to the matching reference guide.
1. Detect context
Python or JavaScript guide by file type
python.md for Django, Flask and FastAPI. javascript.md for Node, Express, React, Vue and Next.js.
2. Load a language guide
Docker guide for Dockerfiles
Uses docker.md for Dockerfile and .dockerignore. For GitHub Actions workflows it points to the gha-security-review skill if available, otherwise supply-chain.md.
3. Load an infrastructure guide
Traces input sources across the codebase
Checks where each value comes from, whether it is config or user input, and what validation or framework protection applies.
Read · 4. Research before flagging
Attacker-controlled vs server-controlled input
Compares request data with settings, environment variables and constants, and checks upstream sanitization.
Read · 5. Verify exploitability
Severity-ranked findings plus a Needs verification list
Each finding has location, confidence, issue, impact, evidence and fix. Test files and dead code are not flagged.
6. Report high confidence only
17 topic files from injection to logging
Covers injection, XSS, authorization, authentication, cryptography, deserialization, file security, SSRF, CSRF, data protection, API security, business logic, modern threats, misconfiguration, error handling, supply chain and logging.
Reference guides

From the official docs

Permissions

Low risk

Claude reads your code and may run shell commands while it traces input. Review the commands it asks to run, and keep secrets out of the files you share.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Its allowed tools include Bash and Task
SKILL.md lists Read, Grep, Glob, Bash and Task, so Claude may run shell commands and subagents while tracing code. Approve commands as you normally would.
Your code and findings stay in your Claude session
Anything Claude reads in your repo becomes part of the conversation. Do not point it at files with secrets you do not want in a chat.

Questions

What does the security-review skill do?

It makes Claude review code for exploitable vulnerabilities such as injection, XSS, broken authorization and weak crypto. It reports only high-confidence findings with evidence and a fix, and lists unclear ones as needing verification.

How do I install it in Claude Code, the Claude app or Codex?

Use the install box above. The README documents claude plugin install sentry-skills@sentry-skills for Claude Code, which brings every Sentry skill, and npx skills add getsentry/skills for Claude Code, Cursor, Cline, GitHub Copilot and other agents. It does not document a Claude app upload or Codex.

Does it scan my whole repo?

It researches the whole codebase to confirm a finding, but SKILL.md says to report only on the file, diff or code you gave it.

Why did it find nothing?

By design it skips theoretical issues, framework-protected patterns, test files, dead code and values from settings or environment variables. If nothing clears the high-confidence bar, it answers with 'No high-confidence vulnerabilities identified.'

Is it safe to install?

It bundles no scripts, only Markdown guides. Its allowed tools include Bash and Task, so Claude may run shell commands and subagents while it traces code. Read what it runs, as with any code review you hand to an agent.

Does it replace a security scanner or audit?

No. It is a review checklist for Claude. SKILL.md has it report only what it is confident about, so it will not list every issue a scanner would.

Does it work with other languages?

SKILL.md says only Python and JavaScript have dedicated guides. For other languages Claude uses the general reference files and checks framework protections directly in the code.

Does it need other skills?

No. SKILL.md names two optional ones from the same repo: django-access-review for Django access control and gha-security-review for GitHub Actions workflows, each used only if available. Without them it falls back to its own authorization.md and supply-chain.md guides. Install the whole getsentry/skills repo, or the plugin, if you want them.

Information

updated Nov 2, 2026

Repository: getsentry/skills
Folder: skills/security-review

getsentry/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2

SKILL.md · github.com
getsentry/skills README · github.com