DevOps-Security-Agent-Skills: an Agent Skills pack for Kubernetes, Terraform, cloud and compliance work

The gap this fills: agents that know DevOps commands but not your stack

A general coding agent can write a Kubernetes manifest. What it usually cannot do is pick the right resource limits, remember that readOnlyRootFilesystem belongs in the same securityContext block, or know which CLI flags a given cloud provider expects. The repository's answer is a curated set of skill folders rather than a prompt library. The README frames the contrast directly: most awesome lists give you links, while this repo aims to give an agent production-ready knowledge it can act on. The intended reader is a platform, SRE or security engineer who already uses an agent daily and is tired of re-explaining the same conventions. The skill catalog splits into DevOps (40+), Security (35+), Infrastructure (65+), AI Engineering (20+), Compliance (20+) and IT Operations (5+). Those counts come from the README table, and the README also advertises 160+ skills, so the categories overlap or the numbers are approximate. Treat the catalog as the source of truth, not the badge.

How the Agent Skills format loads: discover, match, activate

The mechanism is progressive disclosure, and it is the reason this repo is not just a big markdown dump. Each skill lives in its own directory with a SKILL.md file. That file starts with YAML frontmatter, which the agent reads at startup to decide whether the skill is relevant; the detailed instructions below the frontmatter load only when the skill is activated. The README describes three stages: the agent scans skill folders at startup, matches the user's question against them, and then reads the full SKILL.md and runs scripts as needed. The practical consequence is context cost. A 250 to 400 line SKILL.md costs nothing until someone asks about that topic. The layout inside a skill is consistent: SKILL.md for the knowledge, scripts/ for automation, references/ for deep-dive guides and assets/ for config templates. That uniformity is what lets one CLI install skills from many repositories. It also means quality varies by skill, because each one is an independent document rather than a generated view of a single schema.

Installing it and running a first Kubernetes task

The README's recommended path is the skills CLI, which discovers every SKILL.md in the repository and copies it into your agent's skills directory. The first command installs the whole set. Run it from a shell where npx is available, and expect the CLI to report which skills it found and where it placed them.

npx skills add bagelhole/DevOps-Security-Agent-Skills

If you would rather not install all of them, the README shows a targeted install with the --skill flag repeated per skill, -a to name the agent and -y to skip confirmation. This is the form to use when you only care about Kubernetes and Vault.

npx skills add bagelhole/DevOps-Security-Agent-Skills --skill kubernetes-ops --skill hashicorp-vault -a cursor -y

To see what is available before committing, the README lists a --list invocation. There is also a -g flag for a global install, and a single-skill install by URL. If you prefer to keep the files under your own path, the README offers a plain clone instead of the CLI.

git clone https://github.com/bagelhole/DevOps-Security-Agent-Skills.git ~/.skills/devops-security

After installation, open your agent and ask a question in the domain you installed, for example how to add resource requests and limits to a Deployment. The README's example of what a skill contains is a Deployment manifest with requests and limits set and a securityContext carrying runAsNonRoot and readOnlyRootFilesystem. If the agent answers with that shape of manifest, the skill activated. If it answers generically, the frontmatter did not match and the skill was never read.

What the skill catalog actually covers, and where it is thin

Infrastructure is the largest category by the README's own count, spanning AWS, Azure, GCP, Cloudflare, databases, networking, GPU clusters and local AI. Security covers vulnerability scanning, secrets management, hardening, AI agent security and MCP security. The compliance set names SOC2, HIPAA, GDPR, PCI-DSS, policy-as-code and auditing. Several skills target topics that move fast, including MCP server security, AI coding agent guardrails, eBPF observability with Cilium and Tetragon, platform engineering with Backstage, supply chain attack response, OpenTofu migration, dev containers and Nix, and agent evals as CI/CD gates. That list is the repository's own framing of what is current, dated March 2026 in the README. The thin spots are worth naming. The README does not document how a skill is validated, whether scripts are tested in CI, or how conflicts between two skills that touch the same resource are resolved. There is no versioning scheme described for individual skills, so a skill can change between releases without a signal in its path. Those are documentation gaps, not proof the content is wrong, but they matter when you are deciding how much to trust an instruction an agent will follow.

The limitation: it is a knowledge pack, not a control

Nothing here enforces anything. The scripts and templates are files; whether they run, and what they do when they run, depends on the agent and on you. An agent that reads a hardening skill can still emit a manifest with runAsNonRoot set to false if the surrounding conversation pushes it there, and the repository has no mechanism to block that. Compare it with a policy engine such as Open Policy Agent or Kyverno, which evaluates manifests against rules and rejects them at admission time. That difference is the whole point: a skill teaches, a policy engine decides. If your requirement is an auditable gate in a pipeline, this repository is the wrong tool, and the compliance skills in it are guidance for building that gate rather than the gate itself. A second limitation is format lock-in. The skills are only useful to agents that implement Agent Skills; the README names Claude Code, Cursor, Codex, OpenCode and Cline, and points to the vercel-labs skills repository for a longer list. An agent outside that set gets nothing from the frontmatter and would have to be fed the markdown manually.

Alternatives and how their approach differs

The closest comparison is a plain internal runbook or wiki. A wiki page is read by a human and copied into a prompt by hand; a skill is read by the agent at the moment it is relevant. The trade-off is that a wiki is searchable by anyone and a skill is only as good as its frontmatter description, because a weak description means the agent never opens the file. The second alternative is a policy-as-code toolchain, which the repository itself references in its compliance category. Policy as code produces pass or fail results on real manifests in a pipeline; skills produce suggested manifests in a conversation. Teams that need both will end up with both, and the repository does not pretend otherwise. The third comparison is the awesome-list pattern the README calls out. A link list points a human at a source; this repository inlines the config and the commands so the agent does not have to fetch anything. That inlining is also its maintenance burden, since inlined commands age.

Maintenance, licence and the cost of keeping skills current

The repository is not archived. Its last push was on 2026-05-22, roughly four months before today, and the most recent release is v2.0.0 from 2026-03-24, preceded by v1.0.0 on 2026-01-30. Two releases in the first half of 2026 is a slow cadence for a project whose subject matter includes MCP security and coding agent guardrails, both of which move quickly. The upgrade cost is mostly review time rather than migration work: because skills are markdown with YAML frontmatter and no documented per-skill version, a release can change an instruction your team has come to rely on, and the only way to notice is to diff the skill directories after pulling. The README does not document rollback or a pinning mechanism for individual skills, so plan to keep the clone path under version control yourself if you need that. The licence is MIT, which permits commercial use and modification with the copyright notice retained. The repository ships no licence header guidance for contributed scripts, so if you vendor a script into a proprietary pipeline, confirm the provenance of that specific file rather than assuming the top-level LICENSE covers every asset. This is a description of the licence text, not legal advice.

Editorial conclusion

Adopt it if your team already runs Claude Code, Cursor, Codex or another agent that reads SKILL.md files and you want curated Kubernetes, Terraform, cloud and compliance instructions loaded on demand instead of pasted into prompts. Do not adopt it if you need a scanner, a policy engine or anything that enforces a rule at runtime; this repository supplies text and scripts, and the README does not document rollback for installed skills. Before installing, verify that your agent implements the Agent Skills format, check which of the four top-level directories (compliance/, devops/, infrastructure/, security/) covers your stack, and read one SKILL.md end to end to confirm the depth matches your environment.

Frequently asked questions

Official sources

1. BagelHole/DevOps-Security-Agent-Skills on GitHub
2. Issues
3. License: MIT
4. README
5. Releases