container-host-checks — Claude Code skill for Docker host checks
container-host-checks is a Claude Code skill that investigates suspicious activity on Docker or Podman hosts. It checks new and privileged containers, exposed ports, and signs of host escapes.
Apache-2.0 ★ 71 updated 2026-11-15
npx skills add Problem99-io/kodeblue --skill container-host-checks -a claude-codeWhat container-host-checks does
The skill inspects container settings, Docker logs, and host processes. It checks privileges, mounted paths, network and PID modes, added capabilities, and whether an image matches the expected server role. It uses docker stats to look for sustained CPU use and docker port to inspect public ports.
It also looks for possible escape traces, such as files newly written to /etc/cron.d or /root/.ssh, and host processes whose parent is containerd-shim. In sidecar mode, kodeblue may lack filesystem visibility and access to docker, journalctl, or dmesg; runtime exit and restart details then have to come from the operator. The checks rely on Docker or Podman host access and use host_exec, plus host_list and host_read for file access where available.
What it helps with
- inspect new containers for excessive privileges
- spot public ports that do not match the host profile
- look for container-based miners
- find host-side traces of a container escape
- investigate container exits and restarts from available evidence
Example prompts for the agent
- Investigate a new container on this host and check its privileges and mounts.
- Find containers exposing ports that are missing from the profile.
- Check the host for signs of a container escape or a miner.
How to install container-host-checks
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill container-host-checks -a claude-code2Manually: put the files in place
~/.claude/skills/container-host-checks/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/container-host-checks/
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill container-host-checks -a codex2Manually: put the files in place
.agents/skills/container-host-checks/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill container-host-checks -a cursor2Manually: put the files in place
.cursor/skills/container-host-checks/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Problem99-io/kodeblue blue/skills/container-host-checks2Manually: put the files in place
.github/skills/container-host-checks/SKILL.md1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill container-host-checks -a gemini-cli2Platform CLI
gemini skills install https://github.com/Problem99-io/kodeblue --path blue/skills/container-host-checksManually: put the files in place
.devin/skills/container-host-checks/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about container-host-checks
How do I install container-host-checks in Claude Code?
Run npx skills add Problem99-io/kodeblue --skill container-host-checks -a claude-code. The skill file is at ~/.claude/skills/container-host-checks/SKILL.md.
What does container-host-checks do?
It examines Docker or Podman containers for risky privileges, mounts, exposed ports, mining activity, and possible escapes to the host. It also checks related processes and logs when the environment can access them.
What does container-host-checks need to run?
It needs access to a Docker or Podman host. Checks use host_exec; sidecar file access also uses host_list and host_read when available.
Source text
SKILL.md · Apache-2.0
---
name: container-host-checks
description: Investigate container-related findings on a Docker/Podman host — new containers, privileged containers, exposed ports, escapes.
metadata:
kodeblue-bundled: true
---
# Container host checks
- New container: run `host_exec ["docker","inspect","<name>"]` and check `Image`, `Config.Cmd`, `HostConfig.Privileged`, `HostConfig.Binds` (a mount of `/`, `/etc`, `/root` or `/var/run/docker.sock` is a host-takeover risk), `HostConfig.NetworkMode` (`host`), `HostConfig.PidMode`, `CapAdd`.
- Is the image from the registry the server-role definition expects? Check `["docker","images"]` for creation time.
- Miners often run as containers: `["docker","stats","--no-stream"]` shows sustained CPU.
- A container listening publicly on a port not in the profile should be checked with `["docker","port","<name>"]`.
- Who started it: `journalctl -u docker --since <created> --no-pager`. Check who was logged in then and whether they are in the docker group (docker group membership is effectively root).
- Escape signs on the host: new files in `/etc/cron.d` or `/root/.ssh` written at container start time, or new processes on the host whose parent is `containerd-shim`. For a suspected escape, load `container-breakout-detection`: routes out, host-side traces, response and monitors.
## When kodeblue runs beside the workload (sidecar, pod, no runtime CLI)
kodeblue may share a container's process and network namespaces without seeing
its filesystem or having `docker`, `journalctl` or `dmesg`. `sensor_coverage`
says which sensors are blind there.
- Reach the workload's files as it sees them: `host_list` and `host_read` on
`/proc/<pid>/root/…` (its `/tmp`, `/etc/passwd`, app directory). PID 1 may be
the pod's pause process, not the app, so use the app's PID, never
`/proc/1/root`, and in monitors write `{pid:<exe>}`.
- A parent PID of 0 means the runtime supervises the process from outside the
namespace. That is normal, and so is a new PID after a restart.
- Restarting a container keeps its writable layer: tampered files survive it.
Recovery from a compromise is recreating the container from its image.
- `block_ip` acts in the shared network namespace. When the pod is recreated,
its namespace and the rule are gone (see `containment-verification`).
- What happened to the container (exit codes, OOM kills, restarts) lives with
the runtime. When kodeblue cannot see it, say so and ask the operator (see
`service-restart-forensics`).
© Problem99-io · Apache-2.0 · Problem99-io/kodeblue