sentinel-iac 0.5.1

Sentinel

The local-first, cloud-agnostic IaC auditor that fuses cost + security in a single pass — across AWS, Azure, GCP, Kubernetes and on-prem — and writes the fix, not just the flag.

Install · Quick start · Rules · Architecture · Contribute

Sentinel scans your Terraform on your machine, applies security, cost, reliability and governance rules in one pass, reconciles the trade-offs between them, and proposes concrete, ready-to-apply fixes.

$ sentinel scan ./infra

Sentinel found 25 issue(s) in ./infra (6 cost, 6 governance, 3 reliability, 10 security):

1. [CRITICAL/SECURITY] aws_security_group.web  (SEC001)
   at:      infra/main.tf:3
   problem: Ingress on port 22 is open to the entire internet (0.0.0.0/0) and covers SSH/RDP.
   fix:     Restrict cidr_blocks to known ranges, or front the service with a
            load balancer or SSM Session Manager instead of exposing it directly.
   patch:
     cidr_blocks = ["10.0.0.0/8"]  # replace with your trusted CIDR

9. [MEDIUM/COST] aws_instance.batch  (COST001)
   problem: Instance type 'm5.4xlarge' is very large (~$561/month on-demand) and likely over-provisioned.
   saves:   ~$420.48/month (estimate)
   patch:
     instance_type = "m5.xlarge"

Trade-offs (cost x risk on the same resource):
  ⚖ aws_db_instance.main [COST + SECURITY]
    Sequence the security fix first, then right-size — resizing an exposed
    resource first just makes the breach cheaper to run.

Estimated recoverable spend: ~$1,717.53/month
9 resource(s), 1 file(s), 0.12s. Local-only scan; nothing left this machine.

Why Sentinel

Cost lives in one tool (Infracost, Kubecost), security in another (Checkov, Trivy), and nobody reconciles them at the moment of the decision. A change that saves money can widen your attack surface — and no scanner tells you that. Sentinel is different by design:

• Fuses cost + security. One pass, one ranked report, explicit trade-offs when both hit the same resource, with estimated dollars next to every cost finding.
• Fixes, not just flags. Every finding carries plain-English remediation and, where safe, a ready-to-apply HCL patch.
• Local-first, by architecture. No SaaS, no upload, no telemetry. Your code, secrets and infrastructure data never leave your machine. LLM explanations are strictly opt-in and bring-your-own-key (or fully local via Ollama).
• Policy-as-code around the scanner. Encode your org's standards — required tags, severity gates, grandfathered ignores — in a reviewable .sentinel.yaml.
• Built to operate, eventually. Every action is governed by the Trust Ladder: today read-only with an audit trail; auto-fix PRs and policy-gated remediation come only after trust is earned.

Installation

Requires Python 3.10+.

pip install sentinel-iac

Or run from source:

git clone https://github.com/getsentinel-oss/sentinel && cd sentinel
pip install -e .

Or via Docker:

docker build -t sentinel . && docker run --rm -v "$PWD:/scan" sentinel scan /scan

Quick start

# Scan a directory (recursive; .terraform/ and vendored dirs are skipped)
sentinel scan ./infra

# Machine formats for CI and code scanning
sentinel scan ./infra --format json   --output report.json
sentinel scan ./infra --format sarif  --output report.sarif   # GitHub Code Scanning
sentinel scan ./infra --format markdown --output report.md    # paste into a PR

# Gate CI: exit 1 only on HIGH or CRITICAL (default), or choose your line
sentinel scan ./infra --fail-on CRITICAL
sentinel scan ./infra --fail-on NEVER        # report, never fail

# Draft the fixes as a review-ready proposal (changes nothing)
sentinel propose ./infra --output proposal.md

# What does this branch do to the bill? (exit 1 if it breaches your budget)
sentinel diff ./infra-main ./infra-branch

# See every rule
sentinel rules

Policy-as-code

Drop a .sentinel.yaml next to your Terraform:

min_severity: LOW
fail_on: HIGH
required_tags: [Owner, CostCenter]
disabled_rules: [GOV001]
severity_overrides:
  COST002: MEDIUM
ignores:
  - rule: SEC001
    resource: aws_security_group.legacy_*
    reason: grandfathered until the Q3 migration

Need a documented exception? Suppress one rule for one resource, inline, with a mandatory reason:

# sentinel:ignore SEC001 reason=reachable only through the corporate VPN

See docs/configuration.md for the full reference.

Opt-in AI explanations (BYO-key or fully local)

Sentinel never sends anything anywhere by default. If you want tailored explanations, configure a provider and pass --explain:

llm:
  provider: ollama      # fully local — data still never leaves your machine
  model: llama3.1

sentinel scan ./infra --explain

openai and anthropic providers use your own API key from the environment (OPENAI_API_KEY / ANTHROPIC_API_KEY). Only the finding itself (rule, resource address, message) is sent — never your files. Details in SECURITY.md.

GitHub Actions

- uses: getsentinel-oss/sentinel@v0
  with:
    path: ./infra
    fail-on: HIGH

Or upload SARIF to Code Scanning — see docs/configuration.md#ci.

What it checks (42 rules across 5 providers)

Full catalogue with rationale and references: docs/rules.md.

Run sentinel providers for live coverage. Terraform for any provider is parsed; resources from providers without a rule pack are reported as not scanned (never a misleading "clean"). Coverage detail: docs/coverage.md.

Exit codes

Project status & roadmap

Sentinel is at v0.5 — Terraform + Kubernetes scanning, propose (Trust Ladder rung 1), policy packs, pre-merge cost diffing, and the open-core boundary drawn in code. Shipped and designed releases are in ROADMAP.md. Architecture and decision records live in docs/.

Contributing & governance

Rules are single, small Python functions — adding one is a great first PR. Start with CONTRIBUTING.md; propose a rule with the rule proposal template.

Everyone is welcome to contribute; maintainers make the final merge after reviewing each pull request, and only maintainers cut releases. The model — roles, the merge gate, and how production is protected — is in GOVERNANCE.md and docs/merge-and-security-policy.md.

Free vs. paid

Everything in this repository is free forever — enumerated in code and tested (sentinel license prints the list). Team/Enterprise capabilities (org rollups, SSO, compliance evidence) live in a separate commercial package and never restrict the core.

License

MIT. The core is free forever — see the project principles in docs/architecture.md.