AI agent skill · by @awarexone

Cloud Pentest

Post-access cloud exploitation for AWS, GCP, and Azure — what to do AFTER you obtain credentials or reach a metadata endpoint. Covers IAM enumeration and privilege escalation (iam:PassRole, CreatePolicyVersion, AssumeRole chains, GCP service-account impersonation, Azure managed-identity abuse), IMDSv1/v2 metadata credential theft, STS token abuse, S3/GCS/Blob bucket takeover and object exfil, Lambda/Cloud Functions env-var secrets, cross-account and cross-service pivoting, and turning leaked keys into demonstrated impact (read PII, assume admin, exfil data). Assumes authorized scope only. Use when you have AWS/GCP/Azure keys, an SSRF hitting 169.254.169.254 / metadata.google.internal, a leaked service-account JSON, or a token, and need to escalate and prove impact. For finding buckets/origins first, use cloud-recon; for the SSRF entry point, see web2-vuln-classes. 中文触发词：云渗透、AWS提权、IAM枚举、元数据凭证、云安全审计

Archived 2026-10-23 · Source updated 2026-10-09 · License: MIT. Check the original repository for the latest version.

SKILL.md instructions

# CLOUD PENTEST — Post-Access Exploitation

> Recon finds the door; this is what you do inside. A leaked `AKIA...` key or a metadata token is not a finding on its own — the finding is what that identity can DO. Enumerate the identity, find the privesc path, prove real impact. Never touch data you're not authorized to.

---

## AUTHORIZATION FIRST

```
[ ] Target cloud account/subscription is explicitly in scope
[ ] You have written authorization (program scope, engagement doc)
[ ] Read-only enumeration before ANY state change
[ ] No exfil of real customer data — prove access with a benign canary/own-account object
[ ] Discovered accounts/roles you can pivot to are NOT auto-authorized — confirm scope
```

Stop here if any box is unchecked.

---

## 0. IDENTIFY THE IDENTITY (do this first, always)

| You have | First call | Tells you |
|---|---|---|
| AWS keys | `aws sts get-caller-identity` | account id, principal ARN, user vs role |
| AWS metadata | `curl .../iam/security-credentials/<role>` | temp creds + role name |
| GCP SA JSON / token | `gcloud auth ... ` / `curl metadata ...token` | project, service account, scopes |
| Azure token / MI | IMDS `.../identity/oauth2/token` | tenant, object id, resource access |

Never assume privilege. Enumerate what this identity actually has before planning.

---

## 1. AWS — ENUMERATE THEN ESCALATE

**Enumerate (read-only):**
```bash
aws sts get-caller-identity
aws iam get-account-authorization-details 2>/dev/null   # full policy dump if allowed
aws iam list-attached-user-policies --user-name <u>
aws iam list-role-policies / list-attached-role-policies
# No IAM read? Brute the effective perms with enumerate-iam / then map with a policy tool
```

**Classic privesc paths (need the matching permission):**

| Permission held | Escalation |
|---|---|
| `iam:CreatePolicyVersion` | Write a new default `*:*` version onto an attached policy |
| `iam:PassRole` + `ec2:RunInstances` / `lambda:CreateFunction` | Launch compute AS an admin role |
| `iam:AttachUserPolicy` / `PutUserPolicy` | Attach AdministratorAccess to yourself |
| `sts:AssumeRole` (over-broad trust) | Assume a more privileged role |
| `iam:CreateAccessKey` on another user | Mint keys for a privileged user |
| `lambda:UpdateFunctionCode` on privileged fn | Run code with the function's role |
| `iam:UpdateAssumeRolePolicy` | Rewrite trust policy to let you assume it |

**Impact proof (benign):** list a bucket you were told is in scope, `sts assume-role` into the admin role and `get-caller-identity` to show the new ARN, or read a canary object. Don't touch real PII.

---

## 2. METADATA (IMDS) — SSRF LANDS HERE

```
IMDSv1 (no token): GET http://169.254.169.254/latest/meta-data/iam/security-credentials/<role>
IMDSv2 (token):    PUT .../latest/api/token  (X-aws-ec2-metadata-token-ttl-seconds: 21600)
                   then GET with  X-aws-ec2-metadata-token: <token>
GCP:  GET http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/token
      header: Metadata-Flavor: Google
Azure: GET http://169.254.169.254/metadata/identity/oauth2/token?api-version=2018-02-01&resource=https://management.azure.com/
      header: Metadata: true
```

Got temp creds → jump to section 1 (AWS) / 3 (GCP) / 4 (Azure) and enumerate what THAT role can do. SSRF alone is Medium; SSRF → metadata creds → data/admin is Critical.

---

## 3. GCP — IMPERSONATION IS THE PIVOT

```bash
gcloud auth activate-service-account --key-file=sa.json   # or use the token
gcloud projects get-iam-policy <project>
gcloud iam service-accounts list
```

| Permission | Escalation |
|---|---|
| `iam.serviceAccounts.getAccessToken` / `actAs` | Impersonate a higher-priv SA (`--impersonate-service-account`) |
| `iam.serviceAccountKeys.create` | Mint a key for a privileged SA |
| `iam.roles.update` on a custom role you hold | Add permissions to yourself |
| `cloudfunctions.functions.update` / `deploy` | Run code as the function's SA |
| `storage.objects.get` on a sensitive bucket | Read secrets/state |

Owner/Editor at project level = effectively admin. Check for the default Compute SA having Editor — extremely common.

---

## 4. AZURE — MANAGED IDENTITY & RBAC

```bash
az account show
az role assignment list --assignee <objectId> --all
az resource list
```

| Path | Escalation |
|---|---|
| Managed identity with Contributor | Deploy/modify resources, run commands on VMs (`az vm run-command`) |
| `Microsoft.Authorization/*/write` | Assign yourself Owner |
| Key Vault access policy / RBAC | Read secrets, certs, keys |
| Automation Account / Runbook | Execute as the automation identity |
| Storage account key read | Full blob access |

---

## 5. STORAGE — S3 / GCS / BLOB

```
[ ] List: aws s3 ls s3://<bucket> --no-sign-request  (public?)
[ ] Read/Write ACL: get-bucket-acl / put-object test (own canary file only)
[ ] Bucket policy allows *? cross-account?
[ ] Versioning / logging exposes old secrets?
[ ] Website / static hosting → subdomain takeover angle (see web2-vuln-classes)
```

Writable public bucket serving a live site = Critical (content injection / takeover). Readable bucket with secrets/PII = High/Critical.

---

## 6. SECRETS HARVEST (post-access)

```
[ ] Lambda/Cloud Function/App env vars (often hold API keys, DB creds)
[ ] SSM Parameter Store / Secrets Manager (if perms allow)
[ ] EC2 user-data / instance tags
[ ] Terraform state in the bucket you just read
[ ] CI/CD variables reachable from the identity
```

Each secret → re-run "identify the identity" for the NEW credential. Chain until you hit admin or crown-jewel data, then stop and write it up.

---

## 7. REPORT LINE (per finding)

```
Entry: <leaked key | SSRF→metadata | public bucket | SA JSON>
Identity: <ARN / SA / object id> — <what it could do>
Escalation: <exact permission → action that raised privilege>
Impact proven: <assumed admin ARN | read canary | listed in-scope resource>
Blast radius: <accounts/services/data reachable>
Fix: <least-privilege policy | IMDSv2 enforce | block-public-access | rotate>
```

Impact must be demonstrated with a concrete call, using benign/own-account objects. "The key might allow…" is not a finding — show `get-caller-identity` after the escalation.