cspm-gcp-cis-benchmark

---
name: cspm-gcp-cis-benchmark
description: >-
license: Apache-2.0
---

# CSPM — GCP CIS Foundations Benchmark v3.0 (subset)

Automated assessment of GCP projects against a curated subset of CIS GCP
Foundations Benchmark v3.0. The full benchmark has 80+ controls; this skill
implements **7 high-impact checks** that cover the most common findings on
real GCP projects. Each check is mapped to NIST CSF 2.0.

> **Honest scope:** the table below lists *only* what `src/checks.py` actually
> implements. See the **Roadmap** at the bottom for controls that are documented
> by CIS but not yet automated here. Contributions welcome — one check per function,
> one finding row per control.

## When to Use

- GCP project security posture assessment
- Pre-audit for SOC 2, ISO 27001, FedRAMP
- Vertex AI deployment security review
- New project baseline validation
- Service account key hygiene audit

## Architecture

Closed loop: scan → finding → fix (PR or console) → re-scan to verify the same `control_id` is now `pass`.

```mermaid
flowchart LR
    subgraph GCP["GCP Project — read-only"]
        IAM["IAM & Service Accounts<br/>3 checks"]
        GCS["Cloud Storage<br/>2 checks"]
        NET["VPC / Firewall<br/>2 checks"]
    end

    CHK["checks.py<br/>7 CIS v3.0 controls<br/>roles/viewer + iam.securityReviewer"]
    OUT["Findings<br/>JSON · Console"]
    FIX["Remediation<br/>IaC PR · console fix<br/>or exception with TTL"]
    VERIFY["Re-scan<br/>control_id == pass"]

    IAM & GCS & NET --> CHK --> OUT --> FIX --> VERIFY
    VERIFY -. drift detected .-> CHK

    style GCP fill:#1e293b,stroke:#475569,color:#e2e8f0
    style CHK fill:#172554,stroke:#3b82f6,color:#e2e8f0
    style OUT fill:#1e3a5f,stroke:#60a5fa,color:#e2e8f0
    style FIX fill:#1a2e35,stroke:#2dd4bf,color:#e2e8f0
    style VERIFY fill:#164e63,stroke:#22d3ee,color:#e2e8f0
```

## Security Guardrails

- **Read-only**: Requires `roles/viewer` + `roles/iam.securityReviewer`. Zero write permissions.
- **No credentials stored**: GCP credentials from ADC (Application Default Credentials) only.
- **No data exfiltration**: Results stay local. No calls beyond GCP SDK.
- **Vertex AI safe**: Checks endpoint auth, VPC-SC, CMEK — does not access model data or training data.
- **Idempotent**: Run as often as needed with no side effects.

## Implemented Controls (7)

Each row maps to one function in `src/checks.py`. If it's not in this table, it's not implemented.

### Section 1 — IAM (3 checks)

| # | CIS Control | Function | Severity | NIST CSF 2.0 |
|---|------------|----------|----------|--------------|
| 1.1 | Corporate credentials only (no personal Gmail) | `check_1_1_no_gmail_accounts` | HIGH | PR.AC-1 |
| 1.3 | No user-managed service account keys | `check_1_3_no_sa_keys` | HIGH | PR.AC-1 |
| 1.4 | Service account key rotation (90 days) | `check_1_4_sa_key_rotation` | MEDIUM | PR.AC-1 |

### Section 2 — Cloud Storage (2 checks)

| # | CIS Control | Function | Severity | NIST CSF 2.0 |
|---|------------|----------|----------|--------------|
| 2.1 | Uniform bucket-level access (no legacy ACL) | `check_2_1_uniform_access` | HIGH | PR.AC-3 |
| 2.3 | No public buckets (allUsers/allAuthenticatedUsers) | `check_2_3_no_public_buckets` | CRITICAL | PR.AC-3 |

### Section 4 — Networking (2 checks)

| # | CIS Control | Function | Severity | NIST CSF 2.0 |
|---|------------|----------|----------|--------------|
| 4.2 | No unrestricted SSH/RDP (0.0.0.0/0 on 22/3389) | `check_4_2_no_unrestricted_ssh_rdp` | HIGH | PR.AC-5 |
| 4.3 | VPC flow logs on all subnets | `check_4_3_vpc_flow_logs` | MEDIUM | DE.CM-1 |

## Roadmap — Documented but Not Yet Automated

These controls are part of the CIS GCP Foundations v3.0 benchmark but are *not* implemented in `checks.py` yet. PRs welcome.

| # | CIS Control | Why it matters |
|---|------------|----------------|
| 1.2 | MFA enforced org-wide | Requires Workspace Admin SDK access, not in this skill's scope |
| 1.5 | No keys for default compute/App Engine SAs | Specialised filter on top of 1.3 |
| 1.6 | No project-wide SSH keys | Compute metadata read |
| 1.7 | SA impersonation scoped | IAM Recommender or policy walker |
| 2.2 | Bucket retention policy on compliance buckets | Requires labeling convention |
| 2.4 | CMEK encryption on sensitive data | KMS + Storage join |
| 3.1–3.4 | Logging coverage and alert policies | `google-cloud-logging` + `google-cloud-monitoring` |
| 4.1 | Default VPC deleted | Compute API VPC enumeration |
| 4.4 | Private Google Access | Subnet attribute |
| 4.5 | SSL policies enforce TLS 1.2+ | Compute API SSL policies |

## Usage

```bash
# Run all checks
python src/checks.py --project my-project-id

# Run specific section
python src/checks.py --project my-project-id --section iam
python src/checks.py --project my-project-id --section vertex-ai

# Output JSON
python src/checks.py --project my-project-id --output json --output-format ocsf > cis-gcp-results.json
```

## Remediation — Critical Findings

```
  FINDING: Public Cloud Storage bucket (2.3)
  ───────────────────────────────────────────
  FIX:     gsutil iam ch -d allUsers gs://BUCKET
           gsutil iam ch -d allAuthenticatedUsers gs://BUCKET
  VERIFY:  gsutil iam get gs://BUCKET | grep -c "allUsers"  # should be 0
```

```
  FINDING: Vertex AI endpoint publicly accessible (V.5)
  ─────────────────────────────────────────────────────
  FIX:     gcloud ai endpoints update ENDPOINT_ID --region=REGION --clear-traffic-split
           # Then configure VPC-SC perimeter for Vertex AI
  VERIFY:  gcloud ai endpoints describe ENDPOINT_ID --format=json | jq '.network'
```

## Posture Metrics

| Metric | Target |
|--------|--------|
| CIS Pass Rate | > 90% |
| Service Accounts with User Keys | 0 |
| Public Buckets | 0 |
| Subnets without Flow Logs | 0 |
| Vertex AI Endpoints without VPC-SC | 0 |
| Audit Logging Coverage | 100% of services |