Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Performing authorized privilege escalation assessments in AWS environments to identify IAM misconfigurations that allow users or roles to elevate their permissions using Pacu, CloudFox, Principal Mapper, and manual IAM policy analysis techniques.
.claude/skills/performing-aws-privilege-escalation-assessment/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | — | — |
| case-08 | ✗→✓ | ▲ Improved | — | — |
| case-16 | ✗→✓ | ▲ Improved | — | — |
| case-01 | ✗→✓ | ▲ Improved | — | — |
| case-14 | ✗→✗ | = Same ✗ | — | — |
Do not use for unauthorized testing against AWS accounts, for assessing non-IAM attack vectors (SSRF, application vulnerabilities), or as a substitute for comprehensive cloud penetration testing. Always obtain written authorization before testing.
pip install pacu)go install github.com/BishopFox/cloudfox@latest)pip install principalmapper)Establish the baseline permissions of the test principal before attempting escalation.
bash# Get current identity aws sts get-caller-identity # Enumerate inline and attached policies for the current user aws iam list-user-policies --user-name test-user aws iam list-attached-user-policies --user-name test-user # Get group memberships and group policies aws iam list-groups-for-user --user-name test-user for group in $(aws iam list-groups-for-user --user-name test-user --query 'Groups[*].GroupName' --output text); do echo "=== Group: $group ===" aws iam list-group-policies --group-name "$group" aws iam list-attached-group-policies --group-name "$group" done # Simulate specific API calls to map effective permissions aws iam simulate-principal-policy \ --policy-source-arn arn:aws:iam::ACCOUNT:user/test-user \ --action-names iam:CreateUser iam:AttachUserPolicy iam:PassRole \ lambda:CreateFunction ec2:RunInstances sts:AssumeRole \ --query 'EvaluationResults[*].[EvalActionName,EvalDecision]' --output table
Use Pacu's privilege escalation scanner to identify known IAM escalation techniques.
bash# Start Pacu session pacu # Create session and set credentials Pacu (new:session) > set_keys --key-alias privesc-test # Enumerate IAM configuration Pacu > run iam__enum_users_roles_policies_groups Pacu > run iam__enum_permissions # Run privilege escalation scanner Pacu > run iam__privesc_scan # The scanner checks for 21+ known escalation methods including: # - iam:CreatePolicyVersion (create admin policy version) # - iam:SetDefaultPolicyVersion (revert to permissive older version) # - iam:AttachUserPolicy / iam:AttachRolePolicy (attach admin policy) # - iam:PutUserPolicy / iam:PutRolePolicy (create inline admin policy) # - iam:PassRole + lambda:CreateFunction (Lambda with admin role) # - iam:PassRole + ec2:RunInstances (EC2 with admin instance profile) # - iam:CreateLoginProfile / iam:UpdateLoginProfile (set console password) # - iam:CreateAccessKey (create keys for other users) # - sts:AssumeRole (assume more privileged roles) # - glue:CreateDevEndpoint + iam:PassRole (Glue with admin role)
Use Principal Mapper to build a graph of all IAM principals and identify escalation edges.
bash# Collect IAM data for graph construction pmapper graph create --account ACCOUNT_ID # Query for paths to admin pmapper query 'who can do iam:AttachUserPolicy with * on *' pmapper query 'who can do sts:AssumeRole with arn:aws:iam::ACCOUNT:role/AdminRole' # Find all principals that can escalate to admin pmapper analysis # Visualize the privilege escalation graph pmapper visualize --filetype png # Check specific escalation paths pmapper query 'can arn:aws:iam::ACCOUNT:user/test-user do iam:CreatePolicyVersion with *' pmapper query 'can arn:aws:iam::ACCOUNT:user/test-user do sts:AssumeRole with arn:aws:iam::ACCOUNT:role/*'
Evaluate cross-account trust policies for misconfigured role assumptions that allow unauthorized escalation.
bash# List all roles and their trust policies aws iam list-roles --query 'Roles[*].[RoleName,Arn]' --output text | while read name arn; do trust=$(aws iam get-role --role-name "$name" --query 'Role.AssumeRolePolicyDocument' --output json 2>/dev/null) # Check for wildcards or broad trust echo "$trust" | python3 -c " import json, sys doc = json.load(sys.stdin) for stmt in doc.get('Statement', []): principal = stmt.get('Principal', {}) condition = stmt.get('Condition', {}) if isinstance(principal, dict): aws_princ = principal.get('AWS', '') else: aws_princ = principal if '*' in str(aws_princ) or 'root' in str(aws_princ): has_external_id = 'sts:ExternalId' in str(condition) has_mfa = 'aws:MultiFactorAuthPresent' in str(condition) print(f'ROLE: $name') print(f' Principal: {aws_princ}') print(f' ExternalId required: {has_external_id}') print(f' MFA required: {has_mfa}') if not has_external_id and not has_mfa: print(f' WARNING: No ExternalId or MFA condition - confused deputy risk') " 2>/dev/null done # Test role assumption aws sts assume-role \ --role-arn arn:aws:iam::TARGET_ACCOUNT:role/CrossAccountRole \ --role-session-name privesc-test \ --duration-seconds 900
Use CloudFox to identify additional attack surfaces including resource-based policies and service-specific escalation paths.
bash# Run all CloudFox checks cloudfox aws --profile target-account all-checks -o ./cloudfox-output/ # Specific privilege escalation checks cloudfox aws --profile target-account permissions cloudfox aws --profile target-account role-trusts cloudfox aws --profile target-account access-keys cloudfox aws --profile target-account env-vars # Lambda environment variables with secrets cloudfox aws --profile target-account instances # EC2 with instance profiles cloudfox aws --profile target-account endpoints # Exposed services
Compile all discovered escalation paths with proof-of-concept steps and remediation recommendations.
bash# Generate a consolidated report cat > privesc-report.md << 'EOF' # AWS Privilege Escalation Assessment Report ## Tested Escalation Vectors | Vector | Status | Starting Principal | Escalated To | Risk | |--------|--------|--------------------|--------------|------| | iam:CreatePolicyVersion | EXPLOITABLE | test-user | AdministratorAccess | Critical | | iam:PassRole + lambda:CreateFunction | EXPLOITABLE | dev-role | LambdaAdminRole | Critical | | sts:AssumeRole (cross-account) | EXPLOITABLE | test-user | ProdAdminRole | High | | iam:AttachUserPolicy | BLOCKED | test-user | N/A | N/A | | ec2:RunInstances + iam:PassRole | BLOCKED | test-user | N/A | N/A | ## Remediation 1. Apply permission boundaries to all IAM users and roles 2. Remove iam:CreatePolicyVersion from non-admin principals 3. Add sts:ExternalId condition to all cross-account role trust policies 4. Implement SCP guardrails preventing privilege escalation actions EOF
| Term | Definition | |------|------------| | IAM Privilege Escalation | Exploiting overly permissive IAM policies to gain higher-level access than originally granted to a principal | | Permission Boundary | IAM policy that sets the maximum permissions a principal can have, regardless of identity-based policies attached to it | | iam:PassRole | IAM action allowing a principal to pass an IAM role to an AWS service, enabling the service to act with that role's permissions | | Confused Deputy | Attack where an attacker tricks a trusted service into performing actions on their behalf using cross-account role assumption without external ID validation | | Service Control Policy | AWS Organizations policy that sets maximum permissions for member accounts, providing guardrails against privilege escalation | | Principal Mapper | Open-source tool that models IAM principals and their escalation paths as a directed graph for analysis |
Context: During an authorized assessment, a tester discovers that a developer role has the iam:CreatePolicyVersion permission, which allows creating a new version of any customer-managed policy with arbitrary permissions.
Approach:
iam__enum_permissions in Pacuiam:CreatePolicyVersion on its own attached policy"Action": "*", "Resource": "*", "Effect": "Allow"iam:ListUsers, s3:ListBuckets, etc.iam:CreatePolicyVersion and implementing permission boundariesPitfalls: AWS limits managed policies to 5 versions. If all 5 exist, you must delete a version before creating a new one. Always record the original default version to restore it during cleanup. Permission boundaries prevent this escalation if properly configured, so verify boundary policies before declaring a finding.
AWS Privilege Escalation Assessment Report
=============================================
Account: 123456789012 (Production)
Assessment Date: 2026-02-23
Starting Principal: arn:aws:iam::123456789012:user/test-user
Starting Permissions: S3 read-only, Lambda invoke, EC2 describe
Authorization: Signed by CISO, engagement #PT-2026-014
ESCALATION PATHS DISCOVERED: 4
[PRIVESC-001] iam:CreatePolicyVersion -> Admin
Severity: CRITICAL
Starting Permission: iam:CreatePolicyVersion on policy/dev-policy
Escalation: Created policy version 6 with Action:* Resource:*
Time to Exploit: < 2 minutes
Remediation: Remove iam:CreatePolicyVersion, apply permission boundary
[PRIVESC-002] iam:PassRole + lambda:CreateFunction -> LambdaAdminRole
Severity: CRITICAL
Starting Permission: iam:PassRole, lambda:CreateFunction
Escalation: Created Lambda function with AdminRole, invoked to get admin credentials
Time to Exploit: < 5 minutes
Remediation: Restrict iam:PassRole to specific role ARNs with condition key
[PRIVESC-003] sts:AssumeRole -> Cross-Account Admin
Severity: HIGH
Starting Permission: sts:AssumeRole on arn:aws:iam::987654321098:role/SharedRole
Escalation: Role trust policy allows any principal in source account
Remediation: Add sts:ExternalId condition and restrict Principal to specific roles
TOTAL ESCALATION PATHS: 4 (2 Critical, 1 High, 1 Medium)
PERMISSION BOUNDARIES IN PLACE: 0 / 47 IAM principals
SCP GUARDRAILS BLOCKING ESCALATION: 0 / 3 tested vectors| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 comparable cases.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.