Skip to content

skillSecurity

OWASP Top 10 Review

Walks a codebase category by category through the OWASP Top 10 and reports each confirmed weakness with file, line and a concrete fix.

npx workspacetemplates add owasp-top-10-review --tool claude-code

Writes 1 file to .claude/skills/owasp-top-10-review/

.claude/skills/owasp-top-10-review/SKILL.md31 lines · 1.7 KBFile
---
name: owasp-top-10-review
description: Walks a codebase category by category through the OWASP Top 10 and reports each confirmed weakness with file, line and a concrete fix.
---

# OWASP Top 10 review

Review the code against the current OWASP Top 10, one category at a time. Report only what you confirmed by reading the code.

## Steps

1. Map the surface first: entry points (routes, handlers, queue consumers), the auth layer, data stores and every place user input enters. Note the framework and its built-in protections.
2. For each category, search for its typical patterns, then open every hit:
   - Broken access control: handlers that load a record by id without checking the owner or role, missing checks on admin routes.
   - Cryptographic failures: MD5 or SHA-1 for passwords, hard-coded keys, plain HTTP, weak random for tokens.
   - Injection: queries, shell calls, templates and file paths built from input.
   - Insecure design: no rate limit on login or reset, no limits on expensive operations.
   - Misconfiguration: debug mode, default credentials, open CORS, verbose errors.
   - Vulnerable components: outdated packages with known advisories.
   - Authentication failures, integrity failures (unsigned updates, unsafe deserialization), logging gaps, and server-side request forgery.
3. Trace each suspect from input to the dangerous call. If a check on the path stops it, drop the finding.

## Rules

- Never claim a vulnerability you could not trace end to end.
- Do not run exploits or hit live systems. Read and reason only.
- Do not change code unless asked; propose patches.

## Report

Group by category. Per finding: `path:line`, severity, how it is exploited, the fix. List categories with no findings under "Checked, clean".

What it does

A generic "find security bugs" prompt skims the surface and misses whole classes of problems. This skill makes the agent take the ten OWASP categories one at a time, search for the code patterns each one produces, and open every hit before calling it a finding.

You get a short report grouped by category, with the file and line, how an attacker would use the flaw, and the smallest patch. Categories with nothing found are listed as checked, so you know what was covered.

Example

you › Review the orders service against the OWASP Top 10.

agent › A01 Broken access control: src/routes/orders.ts:44 loads an order by id with no owner check, so any logged-in user can read any order. Fix: add userId to the query. A03 Injection: src/search.ts:19 concatenates the sort field into SQL; allowlist the column. Checked, clean: A02, A05, A07, A08, A09, A10. A06 not verifiable offline, run npm audit.

Questions

Does it scan dependencies for known vulnerabilities?
It flags outdated packages it can see in the manifest, but advisory data lives in tools like npm audit or pip-audit. Run one alongside and feed the output in.
Will it find every vulnerability?
No. It is a structured code read, not a penetration test. It is good at access checks, injection and unsafe defaults, and weaker on logic flaws that need running the app.
  • CSRF and CORS Hardening

    SkillSecurity

    Free