| name | security-review |
| description | Audit grizzly-platform infrastructure security posture. Use when setting up new machines, roles, or services; when reviewing existing IaC for security gaps; when the user asks about attack surface, hardening, or exposure; or when preparing a machine or service for production use. Covers SSH, firewalls, network segmentation, secrets handling, service exposure, BMC/iDRAC, backup security, and supply chain. |
Security Review
Structured audit of self-hosted infrastructure security. Operate in two modes:
- Targeted review — Audit a specific file, role, playbook, or machine config being worked on.
- Posture assessment — Broad sweep across the repo and (optionally) live machines.
Targeted Review
When reviewing a specific piece of IaC, walk through every applicable category from the checklist in references/checklist.md. For each:
- State whether it's addressed, partially addressed, or missing.
- If missing, state the risk and suggest a concrete fix (IaC, not manual).
- If not applicable, say so briefly.
Output format: a concise table with Category | Status | Finding | Recommendation columns.
Posture Assessment
Read references/checklist.md for the full checklist. Then:
- Read all active Ansible roles, playbooks, inventory files, and scripts in the repo.
- For each machine in the CLAUDE.md architecture table, check what the IaC declares against the checklist.
- Identify gaps — things the IaC doesn't address at all.
- Produce a prioritized gap report: Critical (actively exploitable or data-loss risk), High (missing standard hardening), Medium (defense-in-depth gaps), Low (nice-to-have).
If the user wants live verification (SSH into machines), confirm before doing so. Compare live state against what IaC declares — flag any drift.
Principles
- Never recommend security theater. Every recommendation must address a real threat in a self-hosted infrastructure context (e.g., don't recommend WAF for an internal-only service).
- Recommendations must be implementable as IaC. If something can't be automated, flag it as a manual step requiring documentation.
- Consider the threat model: self-hosted infrastructure on a residential network, internet-exposed via VPN tunnel to VPS, iDRAC/BMC on the local network.
- Don't just find problems — prioritize them. A missing firewall rule on an internet-facing service matters more than a missing log rotation config.