agentsclimarketplace

Iac security

Skill ShieldNet-360/secure-vibe/skills/iac-security

Hardening rules for Terraform, CloudFormation, and Pulumi: state, providers, drift, secretsFrom its SKILL.md

Install
npx -y skills add ShieldNet-360/secure-vibe --skill iac-security

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 15 stars15 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

6.0 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it

Infrastructure-as-Code Security

Rules (for AI agents)

ALWAYS

  • Pin every provider/module to an exact version or a pessimistic constraint (~> 5.42); never >= 0 or unpinned latest.
  • Configure a remote backend with encryption at rest, server-side state locking, and versioning (Terraform: s3 + DynamoDB lock table with kms_key_id; Pulumi: the managed backend or s3://?kmskey=; CloudFormation: managed by AWS).
  • Encrypt every persistent resource by default with a customer-managed KMS key: S3 buckets, EBS volumes, RDS, EFS, DynamoDB, SQS, SNS, CloudWatch log groups.
  • Tag every resource with owner, environment, cost-center, and data-classification via a default tags block.
  • Run terraform plan (or pulumi preview, aws cloudformation deploy --no-execute-changeset) in CI and require a human approval before apply on production stacks.
  • Add a drift-detection job that runs daily and opens an issue when actual cloud state diverges from code (Terraform Cloud drift detection, pulumi refresh, cfn-drift-detect).
  • Use IAM Conditions to scope every role: aws:SourceArn, aws:SourceAccount, aws:PrincipalOrgID, and TLS-only access policies on storage.

NEVER

  • Hardcode provider credentials in the code or .tfvars (access_key, secret_key, client_secret, service_account_key). Use OIDC federation from CI, the provider's instance metadata service, or a secret manager.
  • Commit terraform.tfstate, terraform.tfstate.backup, .pulumi/, or any *.tfvars containing real secrets. They contain plaintext secrets even if the code references variables.
  • Use local_exec / null_resource to fetch secrets at apply time and stash them in state. State is queryable plaintext by anyone with backend read access.
  • Open security groups / firewall rules to 0.0.0.0/0 for ports 22, 3389, 3306, 5432, 1433, 6379, 27017, 9200, 11211 — even for "just dev". Use bastion or VPN.
  • Grant *:* (wildcard action on wildcard resource) IAM policies. Use iam:PassRole with explicit resource ARNs.
  • Disable provider TLS verification (skip_tls_verify, insecure = true).
  • Use count = 0 to "soft-delete" resources you actually want gone — destroy them.

KNOWN FALSE POSITIVES

  • Bastion hosts intentionally exposed on port 22 to the internet with hardened configurations are not the same risk as opening RDS to the world. Document the exception inline.
  • Public CloudFront distributions, ALB listeners on 80/443, API Gateways, and Lambda function URLs that are meant to be internet-facing.
  • Bootstrap resources (the S3 bucket and DynamoDB lock table the backend itself uses) must exist before remote state can; this chicken-and-egg is usually bootstrapped by a one-time local backend that's then migrated.

Context (for humans)

IaC mistakes scale: a single bad module gets terraform apply'd into hundreds of accounts. The classes of issue we cover here — state-secret leaks, unbounded network exposure, wildcard IAM, drift — are exactly what CIS and the cloud providers' own well-architected reviews flag the most. AI assistants are particularly prone to generating "works on my machine" Terraform that pins nothing and uses local state; this skill is the counterbalance.

Verify & lock (triaging a finding)

A scanner/review hit is a candidate, not a confirmed bug. Confirm it, fix it, then lock it so it can't come back.

  1. Confirm it's real (inspect the resolved plan, not the source). Run terraform plan / pulumi preview / cfn change-set and inspect the resolved resource, since a variable, module default, or default-tags block can flip the verdict: a literal 0.0.0.0/0 ingress on 22/3306/5432, a public S3 ACL or bucket policy, an unencrypted volume/bucket/RDS (no kms_key_id), a *:* IAM action/resource, an unpinned provider, or local/unencrypted state. For secrets, grep the plan and the state file — referencing a variable still leaves plaintext in state. Real if the planned resource has the bad property; FP if a constraint already narrows it (e.g. an intentional bastion on 22).
  2. Fix, then lock with a regression test (unit or integration — dev's call): add a policy-as-code check in CI (conftest/OPA, checkov, trivy config, or terraform test) asserting the resource stays constrained — no 0.0.0.0/0 on admin ports, encryption with a CMK, private ACL, no wildcard IAM, providers pinned, remote encrypted state — plus a benign config that passes so the rule is exercised. Commit it so the guard can't be silently dropped.

References

What ships with it: 3 files

9.8 KB alongside SKILL.md

tests/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.