security-review

Name: security-review
Author: selmakcby

111stars17forks

How to install security-review?

selmakcby/claude-agents-skills/security-review

$npx -y skills add selmakcby/claude-agents-skills --skill security-review

Installs into the current project.

Summary

AI-powered security vulnerability detection. Use PROACTIVELY after writing code that handles user input, authentication, API endpoints, payments, or sensitive data. Flags OWASP Top 10 issues with diff-aware scanning.

Files · 1

View on GitHub

SKILL.md

1	<!--
2	Source: anthropics/claude-code-security-review (official)
3	File: https://github.com/anthropics/claude-code-security-review/blob/main/.claude/commands/security-review.md
4	Used by: reviewer agent
5	-->
6
7	# Security Review
8
9	## When to trigger
10	- After writing code touching auth, payments, API endpoints, file I/O, or user input
11	- Before commit on security-sensitive changes
12	- When user asks `/security-review`
13
14	## Scanning mode
15
16	Diff-aware: only analyze changed files in the current PR/commit.
17	Focus on real vulnerabilities — filter out false positives aggressively.
18
19	## What to check (OWASP Top 10 mapping)
20
21	### 1. Injection
22	- SQL injection → parameterized queries?
23	- NoSQL injection → $where clauses, user-controlled operators?
24	- Command injection → `exec()` with user input?
25	- Prompt injection (AI-specific) → user input flowing into system prompts?
26
27	### 2. Broken authentication
28	- Missing auth checks on protected routes
29	- Weak password hashing (bcrypt/argon2 required)
30	- JWT misuse — missing signature verification, weak secrets, `none` algorithm accepted
31	- Session fixation, missing session rotation on login
32
33	### 3. Sensitive data exposure
34	- Hardcoded secrets in code
35	- API keys in client bundles (import in `"use client"` components)
36	- PII in logs
37	- Missing HTTPS enforcement
38	- Secrets in error messages
39
40	### 4. XML/XXE
41	- XML parsers with external entity processing enabled
42
43	### 5. Broken access control
44	- Missing authorization (authenticated but not authorized)
45	- IDOR (accessing other users' data by changing an ID)
46	- Privilege escalation paths
47
48	### 6. Security misconfiguration
49	- Default credentials
50	- Debug endpoints exposed in prod
51	- Missing security headers (CSP, HSTS, X-Frame-Options)
52
53	### 7. XSS
54	- `dangerouslySetInnerHTML` with user input
55	- Rendering LLM output without sanitization
56	- URL parameters reflected without escaping
57
58	### 8. Insecure deserialization
59	- `JSON.parse` with user-controlled content (usually fine in JS, but watch for prototype pollution)
60
61	### 9. Using components with known vulnerabilities
62	- Run `npm audit` mentally — are any flagged deps in critical paths?
63
64	### 10. Insufficient logging
65	- Security events unlogged (failed logins, permission denials)
66	- Overlogging (logging PII, tokens, passwords)
67
68	## AI-specific checks
69
70	- Webhooks: signature verification present? (Stripe, GitHub, etc.)
71	- Rate limiting: on public endpoints? On AI endpoints especially?
72	- Input length: max length on LLM inputs?
73	- Output validation: Zod schema on LLM responses before rendering?
74	- CORS / CSRF: properly configured for auth endpoints?
75
76	## Severity (same as code-review)
77
78	- CRITICAL — exposed secret, missing auth, unverified webhook, SQL injection, missing rate limit on AI endpoint
79	- HIGH — missing input validation, weak crypto
80	- MEDIUM — logging gaps, overly verbose errors
81	- LOW — defense-in-depth suggestions
82
83	## Output format
84
85	```markdown
86	## CRITICAL
87	- [`<file>:<line>`] <vulnerability> (OWASP: <category>)
88	- Impact: <what an attacker could do>
89	- Fix: <specific mitigation>
90
91	## HIGH / MEDIUM / LOW
92	- ...
93
94	## Verdict
95	PASS \| BLOCK (CRITICAL count: N)
96	```
97
98	## Rules
99
100	- Never hand-wave. "This might be a risk" is not enough — say if it is or isn't, with evidence.
101	- Every finding: severity, OWASP category, file + line, attacker scenario, specific fix.
102	- Rotate suspected exposed secrets immediately and flag at top of report.
103	- False positives are worse than no review — if you're not sure, don't flag it.
104	- AI endpoints get extra scrutiny — token limits, output sanitization, rate limits.

People who liked this also viewed

6 picks

Type

security-review

111stars17forks

View on GitHub

How to install security-review?

selmakcby/claude-agents-skills/security-review

$npx -y skills add selmakcby/claude-agents-skills --skill security-review

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<!--
2	Source: anthropics/claude-code-security-review (official)
3	File: https://github.com/anthropics/claude-code-security-review/blob/main/.claude/commands/security-review.md
4	Used by: reviewer agent
5	-->
6
7	# Security Review
8
9	## When to trigger
10	- After writing code touching auth, payments, API endpoints, file I/O, or user input
11	- Before commit on security-sensitive changes
12	- When user asks `/security-review`
13
14	## Scanning mode
15
16	Diff-aware: only analyze changed files in the current PR/commit.
17	Focus on real vulnerabilities — filter out false positives aggressively.
18
19	## What to check (OWASP Top 10 mapping)
20
21	### 1. Injection
22	- SQL injection → parameterized queries?
23	- NoSQL injection → $where clauses, user-controlled operators?
24	- Command injection → `exec()` with user input?
25	- Prompt injection (AI-specific) → user input flowing into system prompts?
26
27	### 2. Broken authentication
28	- Missing auth checks on protected routes
29	- Weak password hashing (bcrypt/argon2 required)
30	- JWT misuse — missing signature verification, weak secrets, `none` algorithm accepted
31	- Session fixation, missing session rotation on login
32
33	### 3. Sensitive data exposure
34	- Hardcoded secrets in code
35	- API keys in client bundles (import in `"use client"` components)
36	- PII in logs
37	- Missing HTTPS enforcement
38	- Secrets in error messages
39
40	### 4. XML/XXE
41	- XML parsers with external entity processing enabled
42
43	### 5. Broken access control
44	- Missing authorization (authenticated but not authorized)
45	- IDOR (accessing other users' data by changing an ID)
46	- Privilege escalation paths
47
48	### 6. Security misconfiguration
49	- Default credentials
50	- Debug endpoints exposed in prod
51	- Missing security headers (CSP, HSTS, X-Frame-Options)
52
53	### 7. XSS
54	- `dangerouslySetInnerHTML` with user input
55	- Rendering LLM output without sanitization
56	- URL parameters reflected without escaping
57
58	### 8. Insecure deserialization
59	- `JSON.parse` with user-controlled content (usually fine in JS, but watch for prototype pollution)
60
61	### 9. Using components with known vulnerabilities
62	- Run `npm audit` mentally — are any flagged deps in critical paths?
63
64	### 10. Insufficient logging
65	- Security events unlogged (failed logins, permission denials)
66	- Overlogging (logging PII, tokens, passwords)
67
68	## AI-specific checks
69
70	- Webhooks: signature verification present? (Stripe, GitHub, etc.)
71	- Rate limiting: on public endpoints? On AI endpoints especially?
72	- Input length: max length on LLM inputs?
73	- Output validation: Zod schema on LLM responses before rendering?
74	- CORS / CSRF: properly configured for auth endpoints?
75
76	## Severity (same as code-review)
77
78	- CRITICAL — exposed secret, missing auth, unverified webhook, SQL injection, missing rate limit on AI endpoint
79	- HIGH — missing input validation, weak crypto
80	- MEDIUM — logging gaps, overly verbose errors
81	- LOW — defense-in-depth suggestions
82
83	## Output format
84
85	```markdown
86	## CRITICAL
87	- [`<file>:<line>`] <vulnerability> (OWASP: <category>)
88	- Impact: <what an attacker could do>
89	- Fix: <specific mitigation>
90
91	## HIGH / MEDIUM / LOW
92	- ...
93
94	## Verdict
95	PASS \| BLOCK (CRITICAL count: N)
96	```
97
98	## Rules
99
100	- Never hand-wave. "This might be a risk" is not enough — say if it is or isn't, with evidence.
101	- Every finding: severity, OWASP category, file + line, attacker scenario, specific fix.
102	- Rotate suspected exposed secrets immediately and flag at top of report.
103	- False positives are worse than no review — if you're not sure, don't flag it.
104	- AI endpoints get extra scrutiny — token limits, output sanitization, rate limits.

People who liked this also viewed

6 picks

Type