Hunt cloud / infrastructure misconfigurations. AWS: public S3 buckets (s3:GetObject anonymous), permissive bucket policies (PutObjectAcl public-write), exposed CloudFront origin, public Lambda function URL, public RDS snapshot, IAM credentials in JS bundles, AWS metadata accessible via SSRF. GCP: public GCS buckets, exposed Cloud Run services, leaked service account JSON. Azure: public blob containers, exposed Function App. (Kubernetes/Docker exposure is owned by hunt-k8s; CI/CD pipeline attacks by hunt-cicd; post-credential IAM escalation by cloud-iam-deep.) Detection: targeted dorking, certificate transparency, JS bundle secret extraction, port scan for known service ports. Validate: actual data read / write / RCE. Use when hunting cloud-native storage and compute misconfig (S3/GCS/Blob, IMDS-via-SSRF, serverless, public managed services).
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Hunt cloud / infrastructure misconfigurations. AWS: public S3 buckets (s3:GetObject anonymous), permissive bucket policies (PutObjectAcl public-write), exposed CloudFront origin, public Lambda function URL, public RDS snapshot, IAM credentials in JS bundles, AWS metadata accessible via SSRF. GCP: public GCS buckets, exposed Cloud Run services, leaked service account JSON. Azure: public blob containers, exposed Function App. (Kubernetes/Docker exposure is owned by hunt-k8s; CI/CD pipeline attacks by hunt-cicd; post-credential IAM escalation by cloud-iam-deep.) Detection: targeted dorking, certificate transparency, JS bundle secret extraction, port scan for known service ports. Validate: actual data read / write / RCE. Use when hunting cloud-native storage and compute misconfig (S3/GCS/Blob, IMDS-via-SSRF, serverless, public managed services).
# S3 listing
curl --max-time 30 --connect-timeout 10 -s "https://TARGET-NAME.s3.amazonaws.com/?max-keys=10"
aws s3 ls s3://target-bucket-name --no-sign-request
# Try common bucket namesfor name in target target-backup target-assets target-prod target-staging; do
curl --max-time 30 --connect-timeout 10 -s -o /dev/null -w "$name: %{http_code}\n""https://$name.s3.amazonaws.com/"done# Firebase open rules
curl --max-time 30 --connect-timeout 10 -s "https://TARGET-APP.firebaseio.com/.json"# read
curl --max-time 30 --connect-timeout 10 -s -X PUT "https://TARGET-APP.firebaseio.com/test.json" -d '"pwned"'# write
EC2 Metadata (via SSRF)
http://[REDACTED_IP]/latest/meta-data/iam/security-credentials/ # role name
http://[REDACTED_IP]/latest/meta-data/iam/security-credentials/ROLE-NAME # keys
For testing cloud-misconfig findings against a local AWS sim before/instead of hitting real cloud:
# LocalStack 3.0 community (pin the version — 4.x requires a Pro license)
docker run -d --name lab-localstack -p 14566:4566 localstack/localstack:3.0
# awscli ≥ 2.30 + LocalStack 3.0 incompatibility workaround (x-amz-trailer header):export AWS_REQUEST_CHECKSUM_CALCULATION=when_required
export AWS_RESPONSE_CHECKSUM_VALIDATION=when_required
export AWS_ENDPOINT_URL=http://localhost:14566
export AWS_ACCESS_KEY_ID= AWS_SECRET_ACCESS_KEY= AWS_DEFAULT_REGION=us-east-1
test
test
Without those env vars, aws s3 cp/sync fails with InvalidRequest. Document this for the team. See docs/verification/phase2j-cloud-localstack.md for the full reproducible flow.
CloudWatch RUM Weaponization (2024-2026 surface)
AWS CloudWatch RUM (Real-User Monitoring) is a client-side telemetry service launched late 2021. Customers embed a JS snippet on their pages that sends performance/error events to dataplane.rum.<region>.amazonaws.com. The snippet's AppMonitor config contains an identityPoolId (Cognito) and guestRoleArn (IAM role) — both public by design. The IAM role policy is the security boundary, and when developers leave it broader than the documented minimum (rum:PutRumEvents on the AppMonitor ARN), the entire pool becomes the unauthenticated AWS-credential vending machine described in cloud-iam-deep → Cognito Identity Pool chain.
Full chain documented in cloud-iam-deep → Cognito Identity Pool unauthenticated chain. RUM is one common embedding context.
Chain B — Telemetry endpoint covert exfil.dataplane.rum.<region>.amazonaws.com is an AWS-owned domain on every enterprise allowlist. The PutRumEvents payload accepts arbitrary userDetails and customEvents string fields:
Defenders watching egress see traffic to a known-good AWS hostname; DLP doesn't parse the JSON body; SIEM rules typically don't ingest customer RUM telemetry.
Chain C — DOM injection via snippet source poisoning. Many customers either self-host cwr.js on their own CDN (assets.target.com/cwr.js) or bundle aws-rum-web and serve from static.target.com/main.<hash>.js. Subdomain takeover on the JS host or supply-chain compromise (npm typosquat against aws-rum-webb) gives persistent JS execution on every page-load with the trust of the aws-rum-web SDK — including its already-granted Cognito permissions.
Chain D — Telemetry injection / dashboard poisoning. With the public identityPoolId + applicationId, an external attacker can flood PutRumEvents with fake error spikes (drown real alerts), inject XSS payloads into page-URL telemetry that fire when an SOC analyst views the CloudWatch dashboard, and inflate billable RUM event counts (financial DoS).
Severity rubric
Finding
Severity
Justification
guestRoleArn with *:* or wildcards on multiple services
Critical (9.1+)
Anonymous full AWS access
guestRoleArn with s3:*, dynamodb:*, secretsmanager:*, lambda:Invoke* on production resources
High (7.5-8.8)
Data exfil / RCE depending on resource
guestRoleArn with cognito-identity:* or iam:PassRole
High (8.0)
Privilege escalation primitive
guestRoleArn with only rum:PutRumEvents + endpoint-scoped resource
Informational
Documented, intended config
RUM userDetails logging PII into events viewable in CloudWatch console
Medium (5.3-6.5)
Sensitive data exposure via dashboard sharing
RUM AppMonitor accepts PutRumEvents from arbitrary internet sources (telemetry injection)
Low-Medium (4.3)
Dashboard poisoning, alert evasion, billing DoS
Self-hosted cwr.js on takeoverable subdomain
Critical (9.8) when chained
Persistent stored XSS across every customer page
Disclosed cases / authoritative writeups
No CVE assigned specifically to AWS RUM as of 2026-05. The attack class is documented in research but specific named bug-bounty payouts on RUM are rare in public hacktivity. The pattern is "Cognito identity pool over-permission via embedded SDK" — RUM is one common embedding.
Andres Riancho — "Misconfigured Cognito Identity Pools" (2020/2023) — establishes the attack class. andresriancho.com
Rhino Security Labs — Pacu cognito__enum_identity_pools — production tooling that automates Chain A. github.com/RhinoSecurityLabs/pacu
NotSoSecure / Claranet — "Exploiting weak configurations in Amazon Cognito" (Nov 2023) — explicitly calls out RUM as one of three SDKs commonly leaking the pool ID. notsosecure.com
Run aws sts get-caller-identity and screenshot the role ARN.
Run enumerate-iam / Pacu iam__enum_permissions — capture at least one allowed action beyond rum:PutRumEvents. Without this, the finding is Informational.
Demonstrate at least one read/list against a real resource (S3 bucket list, DynamoDB scan, Lambda invoke).
Do not modify/delete data even if permitted — read-only PoC only.
S3 bucket listing without read access — finding a listable bucket is recon, not a vulnerability. Need to demonstrate readable objects with sensitive content.
Firebase public read without sensitive data — an open Firebase RTDB with only public app config is informational. Need PII, credentials, or internal data.
GCP/Azure storage without credential impact — public blob storage containing static assets is not a finding. Need secrets, source code, or customer data.
Cloud metadata endpoint accessible from outside — SSRF to 192.0.2.1 is only exploitable from inside the cloud. External access to metadata is a different (and more severe) finding.
Bucket ACL misconfig vs policy misconfig — bucket policies and ACLs are different mechanisms. Test both before concluding the bucket is public.
CloudFront/Cloudflare CDN origin leakage — bypassing CDN to reach the origin is recon until you demonstrate what's different (debug endpoints, auth-less access).
Unrestricted cloud function triggers — publicly-invokable functions may be intentional. Need to demonstrate the function does something sensitive (data access, resource modification).
Related Skills & Chains
hunt-subdomain — Stale CNAMEs pointing to deleted buckets are a takeover gold mine. Chain primitive: Cloud misconfig (S3 public/deleted) + hunt-subdomain → unclaimed CNAME points to bucket → assets.target.com takeover.
cloud-iam-deep — A leaked SA JSON / AWS key in a public bucket is only half the bug. Chain primitive: Public S3 + leaked AWS key in .env → cloud-iam-deep enumeration → cross-service iam:PassRole escalation.
hunt-ssrf — Metadata service is reachable only from inside the VPC; SSRF is the bridge. Chain primitive: SSRF + cloud misconfig (IMDSv1 still enabled) → instance role keys → S3/RDS data read.
supply-chain-attack-recon — Exposed CI/CD endpoints and SBOMs reveal internal package names. Chain primitive: Exposed Jenkins/GitLab + internal package name leak → npm/PyPI dependency-confusion publish → CI build pwn.
security-arsenal — Load the Cloud Bucket Wordlist (target-prod / target-backup / target-staging permutations) and the Admin-Panel Path List for fast enumeration.
triage-validation — Apply the Unique-Marker gate: any "writable bucket" claim requires a write of a unique marker file and a read-back from a clean session before report submission.
