Competition Kernel Container Escape is a downstream skill of the CTF Sandbox Orchestrator. It applies when the decisive step in a challenge is proving a boundary crossing between a containerized context and the host or a higher-privilege kernel context.
The agent maps runtime isolation first: namespaces, cgroups, seccomp, capabilities, LSM state and mounts. It then separates the exploit prerequisite, the primitive and the crossing proof, and keeps instrumented observations apart from the pristine challenge path.
What You Get
- A quick start that maps isolation before any exploit discussion
- A checklist of namespace, cgroup, capability, seccomp and LSM state
- A rule to separate crash-only behavior from stable capability gain
- A four-part chain from prerequisite state to host-level capability
- A hand-off to the Linux credential pivot skill when the path relies on credential replay
Common Use Cases
- Judging whether a container misconfiguration is a real escape in a lab
- Recording the isolation baseline of a challenge container
- Stating whether the root cause is a kernel bug, runtime setting, capability grant or namespace leak
- Documenting a primitive-to-crossing chain for a write-up
What Makes It Unique
It demands evidence of crossing and treats container root as insufficient on its own. Crash artifacts do not count as an escape until a stable capability appears, and the chain must reproduce from a reset baseline, which keeps claims checkable.
Under The Hood
The skill is instruction text only: the SKILL.md body, one reference checklist and a small agent config file that disables implicit invocation. It ships no scripts and carries no exploit code. The instructions tell the agent to inspect kernel, runtime and namespace state of a sandbox target the user supplies, which means the agent would run commands on that target itself. The skill replies in Simplified Chinese unless asked for English and runs only after the orchestrator skill is active.