Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add UnboundCompute/security-agent-skills --skill hunting-container-escape-surfacegit clone --depth 1 https://github.com/UnboundCompute/security-agent-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/unboundcompute/security-agent-skills/hunting-container-escape-surface)<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/hunting-container-escape-surface"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/hunting-container-escape-surface/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/unboundcompute/security-agent-skills/hunting-container-escape-surface"><img src="https://agentmods.dev/badge/skills/unboundcompute/security-agent-skills/hunting-container-escape-surface.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00154 | $0.01915 |
| Opus 5.5 | $0.00062 | $0.00766 |
| Sonnet 5.5 | $0.00031 | $0.00383 |
| Haiku 4.5 | $0.00015 | $0.00192 |
Grade A, and why
hunting-container-escape-surface scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Hunting container escape surface: when the container boundary is the security boundary
A container is only an isolation boundary while its security context keeps it one. Treat the container as a box a process might try to climb out of, and the question is which settings hand it a ladder. A privileged pod, an added capability that grants kernel-level power, a host namespace shared in, a writable host path, or a mounted device each turns a container-local compromise into node access, and node access is every other pod on that node. The hunt is not about a bug in the application; it is about the runtime configuration that decides whether a foothold stays contained. You hunt this by inventorying the security context of each workload and finding where a setting lets the process reach the node.
When to use
- Workloads run with elevated security contexts: privileged mode, added capabilities, or host access.
- Pods share host namespaces (host PID, host network, host IPC) or mount host paths and devices.
- A compromise inside a container would be far worse if it could reach the node or neighboring pods.
Scope check
Test escape surface only on clusters and nodes you own or are authorized to assess, on non-production workloads. Demonstrating an escape reaches the real node and can affect other pods, so use a non-production node and stop at proof rather than pivoting. If you can't name the authorization, stop.
The loop
-
Establish the intended isolation first. Name what each workload is supposed to be able to touch: its own filesystem and network, nothing on the node. This is the false-positive killer: a pod that runs unprivileged, drops capabilities, shares no host namespace, and mounts no host path is contained by design, and an elevated setting it genuinely needs and constrains is not automatically a finding. Name the intended boundary, then find where the security context exceeds it.
-
Inventory privileged mode and capabilities. For each workload, read whether it runs privileged and which Linux capabilities it adds. Privileged mode removes most isolation at once; individual capabilities like the ones granting raw device, module, or admin access are nearly as powerful. A privileged or capability-broad container is a candidate escape regardless of the application.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 133 lines · 154 tokens per session scan A 7d3c33c5e405
hunting-container-escape-surface is a skill published in the GitHub repository UnboundCompute/security-agent-skills (5 stars, last pushed 1mo ago), licensed MIT. It adds 154 tokens to every session and 1,915 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-05.
Other skills, from other repositories
cloud-security-engineer
Cloud Security Engineer role bundle for cloud security posture management, IaC review, container security, and identity governance across AWS, Azure, and GCP. Orchestrates cloud-specific posture reviews and zero trust program assessments. Auto-invoked when the user needs help with cloud security configuration, IaC…
container-security
Performs a container and Kubernetes security review against the CIS Docker Benchmark v1.6.0, CIS Kubernetes Benchmark v1.9.0, and NIST SP 800-190. Auto-invoked when reviewing Dockerfiles, Kubernetes manifests, Helm charts, or container orchestration configurations. Evaluates image security, runtime hardening, RBAC…
sandbox-guard
Generate Docker sandbox configurations for safely running untrusted OpenClaw skills. Isolates filesystem, network, and process access.
prodcheck-review
Review this codebase against the prodcheck pre-production checklists — security, performance, scale, integrations, post-launch readiness and people. Use when asked to check whether a project is ready to ship, to audit an area before launch, or to work through a specific checklist. Produces evidence with file:line…
dockerfile-scan
Scan a Dockerfile for insecure build patterns — running as root, unpinned or :latest base images, ADD from remote URLs, piping curl/wget into a shell, hardcoded secrets in ENV/ARG, world-writable chmod 777, and sudo usage. Use when the user asks to "review my Dockerfile", "is this container image secure", "lint my…
analyzing-kubernetes-audit-logs
Parses Kubernetes API server audit logs (JSON lines) to detect exec-into-pod, secret access, RBAC modifications, privileged pod creation, and anonymous API access. Builds threat detection rules from audit event patterns. Use when investigating Kubernetes cluster compromise or building k8s-specific SIEM detection rules.