Cloud Security Assessment

Thin router for security assessment of AWS, GCP, and Azure environments. The body of knowledge lives in per-cloud references/, per-domain methodology/, and per-scope workflows/. Load only what the current task needs.

When to Use

• Multi-cloud or single-cloud security audit (read-only CSPM-style).
• IAM privilege-escalation path discovery.
• Storage (S3 / GCS / Azure Blob) exposure review.
• Network posture and metadata-service checks.
• Secrets-store audit (Secrets Manager / Secret Manager / Key Vault).
• CIS / SOC2 / PCI-DSS benchmark assessment against a live account.
• Cross-cloud identity federation / trust-graph review.

Trigger Phrases

audit AWS security, scan Azure for misconfigurations, check GCP security, test cloud IAM, find S3 bucket issues, cloud penetration test, CIS benchmark audit, cloud privilege escalation, multi-cloud assessment.

When NOT to Use This Skill

This skill assumes you have credentials against a live cloud account. If you only have source artifacts, route to the sibling skills above.

Decision Tree

Target scope?
├─ Single cloud (AWS only)    → workflows/aws_full_assessment.md
├─ Single cloud (GCP only)    → workflows/gcp_full_assessment.md
├─ Single cloud (Azure only)  → workflows/azure_full_assessment.md
└─ 2+ clouds / federated      → workflows/cross_cloud_comparison.md  (fan out per-cloud)

Within a cloud, narrow by domain:
├─ IAM / identity / privesc         → methodology/iam_privilege_escalation.md
├─ Storage exposure (S3/GCS/Blob)   → methodology/storage_misconfig.md
├─ Networking / firewall / IMDS     → methodology/network_security.md
├─ Secrets / KMS / Key Vault        → methodology/secrets_management.md
└─ CIS/compliance full sweep        → the workflow for that cloud (runs scanners)

Need specific commands or tool invocations?
├─ AWS        → references/aws.md
├─ GCP        → references/gcp.md
├─ Azure      → references/azure.md
├─ Tooling    → references/cloud_tools.md
└─ SQL        → references/steampipe_queries.sql

Parallelism Hints

Parallelize freely:

• Per-cloud audits across AWS / GCP / Azure are fully independent — run three concurrent sub-agents.
• Per-service audits within a cloud (IAM, storage, compute, network, secrets, logging) are independent after Phase 0 context gathering.
• Per-region iteration (AWS regions, GCP zones, Azure locations) is independent — fan out.
• Scanner sweep (Prowler + ScoutSuite + Steampipe) runs concurrently — same target, different lenses.

Must be sequential:

• Phase 0 (context / identity / scope) → everything else (enumeration depends on knowing who you are and what you can see).
• IAM enumeration → privesc graph analysis (graph needs the nodes first).
• All per-cloud results → cross-cloud rollup (synchronization point).

Sub-Agent Delegation

Spawn sub-agents when:

• Multi-cloud account: one sub-agent per cloud provider. Each runs its own workflows/<cloud>_full_assessment.md.
• Many subscriptions / projects / accounts in one cloud: one sub-agent per subscription/project.
• Service deep-dives: if an audit surfaces 20+ S3 buckets or 50+ VMs, spawn a dedicated sub-agent for that service.

Coordinator responsibilities: scope enforcement, credential distribution, finding deduplication, cross-cloud correlation.

Reasoning Budget

• Extended thinking for:
  • IAM privilege-escalation path analysis (graph traversal, multi-hop chains, cross-layer — ARM ↔ Entra, AWS IAM → Lambda → DynamoDB).
  • Trust-graph analysis in workflows/cross_cloud_comparison.md Phase F-1.
  • Blast-radius determination for a compromised principal.
  • Effective network reachability through layered NACL + SG + peering + TGW.
• Minimal / no extended thinking for:
  • Misconfiguration enumeration (script-driven, tool output).
  • Scanner orchestration (Prowler / ScoutSuite / Steampipe invocations).
  • Per-bucket / per-resource attribute checks.
  • CIS benchmark cross-referencing.

Multimodal Hooks

• Cloud console screenshots (AWS Console / Azure Portal / GCP Console) are often the clearest evidence for policy findings — capture when proving exploitability.
• ScoutSuite produces an HTML dashboard; screenshots of red "Danger" cards communicate severity well to executives.
• PMapper / ROADrecon visualizations of privesc graphs are worth including as diagrams.

Attach to evidence.screenshot in the finding schema.

Structured Output

All findings conform to schemas/finding.json. Key cloud-specific fields: cloud_provider, account_id, region, service, resource_arn, cis_benchmark_id, compliance (array of CIS / SOC2 / PCI-DSS / HIPAA / GDPR / NIST-800-53 / ISO-27001), privilege_escalation_path, detection.

Workflow Index

Methodology Index

References Index

Tools

Last Validated

• 2026-04
• AWS CLI v2.15+, gcloud 465+, Azure CLI 2.60+
• Prowler v4.0+, ScoutSuite 5.14+, Steampipe 0.22+, Pacu 1.5+
• CIS Benchmarks: AWS v3.0, Azure v2.1, GCP v3.0