Skills for getting a change into production without a bad night: separate deploy from release so the worst case is one cohort and one flag, ship a schema change to a live database without a stuck lock, run a cutover that has a point of no return, plan a platform migration that changes one thing at a time, audit a CI/CD workflow's own definition for what an attacker can reach through it, and write release notes someone can act on.
Expert CI/CD pipeline architect, reviewer, optimizer, debugger, and security auditor, running on the session model (always the strongest available Claude) and backed by a self-contained embedded knowledge base covering GitHub Actions, GitLab CI/CD, Jenkins, Buildkite, Tekton, and Kubernetes-native stacks. Self-hosted runner fleet design, supply chain security (SLSA/Sigstore/SBOMs), progressive delivery, DORA metrics, and cross-platform migration.
Manage a CloudPanel v2 server from Claude Code: read, create and maintain sites, databases, certificates and nginx vhosts, with backups, nginx validation and automatic rollback.
Confidently hand off infrastructure work. Helm, Kubernetes and OpenTofu/Terraform skills, path-scoped conventions, and cloud and IaC documentation connectors.
Fits Slurm jobs to what they actually used: trims the CPU and RAM you over-asked for, drives the GPU you under-used, and states the sample size behind every number.
Infrastructure, environment health and template sync — Terraform and Terragrunt work, a health check for this configuration itself, and the update path back to the upstream template.
Resolve Dependabot alerts in parallel across a repo, org, or your own repos: one subagent per package major line per repo in an isolated worktree, ranked by severity and EPSS, with lockfile validation, major-bounded scoped overrides, and pull requests that open ready for review carrying a computed merge-risk rating and a matching merge-risk label. A separate pin audit, run on its own via /gh-security:audit-pins and preflighted against the repo's own open security PRs, finds overrides no longer needed and opens its own removal PR.
LLM-evaluated permission hook for Claude Code — auto-approves safe operations, blocks destructive commands, delegates ambiguous decisions to a lightweight LLM
Connect Claude Code to your WRM PRO over MCP (enter the WRM MCP address and a token from WRM → Settings → AI connections), with skills for safe server work and Linux incident triage.