planguard-k8slisted
Install: claude install-skill vivekbhange-hub/plan-guard-skill
# planguard-k8s: harden the manifest, keep the behaviour
Most Kubernetes risk is not exotic. It is a container that runs as root, a `:latest` image, a
`hostPath` to the Docker socket, a Service open to the internet, a `cluster-admin` binding added
"temporarily". This skill runs `k8s_lint.py` over the manifests and then writes the fix for each
finding so the workload still does what it did.
## Inputs
- Manifest files or a folder (YAML or JSON, many documents per file are fine). For a Helm chart,
the user renders it themselves and saves the output:
```
helm template my-release ./chart -f values.yaml > rendered.yaml
```
Claude does not run `helm`, `kubectl` or any cluster command. Unrendered templates (`{{ }}`)
are reported as "not parsed".
- What the workload does, when it matters: does it need to talk to the Kubernetes API, write to
its root filesystem, bind a port below 1024, run as a fixed UID? Ask before hardening.
- Optional `.planguard-ignore` (`RULE ADDRESS-GLOB reason`) or a `# plan-guard: ignore KG010 reason`
comment inside a manifest file. Commands below say `python3`; on Windows use `python` (or `py -3`).
## Outputs
- A report: counts by severity, a table of findings with file and line, details grouped by rule
with the reason and a patched snippet, and a "Not parsed" list for documents the reader would not guess at.
- Patched manifests (edited copies or a diff) with the same ports, env, volumes and commands.
- A list of what changes at runtime