hypnguyen1209/offensive-claude / skills/container-k8s-escape
container-k8s-escape
FlaggedUse when breaking out of a container or escalating inside Kubernetes — runc/BuildKit CVEs, privileged/capability/cgroup misconfig escapes, NVIDIA GPU toolkit escape, K8s RBAC abuse, kubelet RCE, ingress/admission-controller RCE, node-to-cluster pivot
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.9 KB |
| references/ingress-admission-attacks.md | markdown | 6.9 KB |
| references/k8s-rbac-escalation.md | markdown | 7.5 KB |
| references/node-host-pivot.md | markdown | 7.5 KB |
| references/nvidia-gpu-escape.md | markdown | 6.7 KB |
| references/privileged-misconfig-escape.md | markdown | 9.1 KB |
| references/runtime-cve-escapes.md | markdown | 9.3 KB |
| scripts/escape_enum.sh | script | 4.5 KB |
| scripts/k8s_rbac_audit.py | script | 7.9 KB |
| scripts/kubelet_exec.py | script | 5.6 KB |
| scripts/nvidiascape_build.sh | script | 4.0 KB |
| scripts/release_agent_escape.sh | script | 4.1 KB |
| scripts/runc_cwd_escape.py | script | 5.5 KB |
scripts/k8s_rbac_audit.py
View on GitHub#!/usr/bin/env python3
"""k8s_rbac_audit.py - Audit Kubernetes RBAC for privilege-escalation and escape primitives.
Walks ClusterRoles/Roles (and their bindings) and flags subjects that hold dangerous permissions:
verb/resource wildcards, escalate/bind/impersonate, pods/exec|attach, secrets get/list, nodes/proxy
(kubelet-exec path), token/SA creation, and webhook-config tampering. Works against any kubeconfig or
an in-pod ServiceAccount token.
USAGE:
python3 k8s_rbac_audit.py --kubeconfig ~/.kube/config --dangerous
python3 k8s_rbac_audit.py --server https://API:6443 --token "$SA_TOKEN" --insecure
python3 k8s_rbac_audit.py --in-pod # auto-load the mounted SA token/CA
python3 k8s_rbac_audit.py --kubeconfig ~/.kube/config --who-can 'create clusterrolebindings'
Backends (auto-detected): the `kubectl` CLI if present (preferred), else direct REST via urllib.
Dependencies: Python 3.6+ stdlib; optional kubectl. Read-only (only issues GET/list + auth can-i).
Authorized engagements only.
"""
import argparse
import json
import os
import shutil
import ssl
import subprocess
import sys
import urllib.request
SA_DIR = "/var/run/secrets/kubernetes.io/serviceaccount"
# (verb, resource) patterns that constitute an escalation/escape primitive -> human note.
DANGEROUS = [
(("*", "*"), "WILDCARD verb+resource == effectively cluster-admin"),
(("escalate", "*roles*"), "can grant perms beyond own (bypasses escalation guard)"),
(("bind", "*roles*"), "can bind self/SA to cluster-admin"),
(("create", "clusterrolebindings"), "can bind self/SA to cluster-admin"),
(("create", "rolebindings"), "can bind within namespace"),
(("impersonate", "*"), "can impersonate users/groups/SAs (system:masters)"),
(("*", "secrets"), "read all secrets (SA tokens, TLS, cloud creds)"),
(("get", "secrets"), "read secrets (SA tokens / creds)"),
(("list", "secrets"), "list secrets cluster/ns wide"),
(("create", "pods"), "schedule privileged/hostPath pod or steal a stronger SA"),
(("*", "pods/exec"), "exec into higher-priv pods, steal their tokens"),
(("create", "pods/exec"), "exec into higher-priv pods, steal their tokens"),
(("*", "pods/attach"), "attach to higher-priv pods"),
(("get", "nodes/proxy"), "nodes/proxy GET -> kubelet WebSocket exec on ANY pod (bypasses API audit)"),
(("*", "nodes/proxy"), "nodes/proxy -> kubelet exec (bypasses API audit)"),
(("create", "serviceaccounts/token"), "mint tokens for higher-priv SAs (TokenRequest)"),
(("create", "serviceaccounts"), "create SAs"),
(("*", "validatingwebhookconfigurations"), "disable admission / install credential-stealing webhook"),
(("*", "mutatingwebhookconfigurations"), "install cluster-wide mutating implant"),
(("create", "certificatesigningrequests"), "mint client certs (CSR approval path)"),
(("*", "daemonsets"), "DaemonSet -> root on every node"),
]
def kubectl_get(kctx, resource):
cmd = ["kubectl"] + kctx + ["get", resource, "-A", "-o", "json"]
try:
out = subprocess.run(cmd, capture_output=True, text=True, timeout=60)
if out.returncode != 0:
# cluster-scoped resources reject -A; retry without it
cmd = ["kubectl"] + kctx + ["get", resource, "-o", "json"]
out = subprocess.run(cmd, capture_output=True, text=True, timeout=60)
return json.loads(out.stdout) if out.stdout.strip() else {"items": []}
except Exception as e:
print(f"[!] kubectl get {resource} failed: {e}", file=sys.stderr)
return {"items": []}
def rest_get(server, token, insecure, path):
ctx = ssl.create_default_context()
if insecure:
ctx.check_hostname = False
ctx.verify_mode = ssl.CERT_NONE
else:
ca = os.path.join(SA_DIR, "ca.crt")
if os.path.exists(ca):
ctx.load_verify_locations(ca)
req = urllib.request.Request(server.rstrip("/") + path,
headers={"Authorization": f"Bearer {token}"})
with urllib.request.urlopen(req, context=ctx, timeout=30) as r:
return json.loads(r.read().decode())
def matches(rule_verbs, rule_res, want_verb, want_res):
def m(have, want):
if have == "*" or want == "*":
return True
if want.startswith("*") and want.endswith("*"):
return want.strip("*") in have
return have == want
return any(m(v, want_verb) for v in rule_verbs) and any(m(r, want_res) for r in rule_res)
def scan_roles(roles, kind):
findings = []
for item in roles.get("items", []):
name = item["metadata"]["name"]
ns = item["metadata"].get("namespace", "-")
for rule in item.get("rules") or []:
verbs = rule.get("verbs", [])
resources = rule.get("resources", []) or rule.get("nonResourceURLs", [])
for (dv, dr), note in DANGEROUS:
if matches(verbs, resources, dv, dr):
findings.append((kind, ns, name, f"{dv} {dr}", note,
",".join(verbs)[:40], ",".join(resources)[:40]))
return findings
def main():
ap = argparse.ArgumentParser(description="Kubernetes RBAC privilege-escalation auditor")
ap.add_argument("--kubeconfig")
ap.add_argument("--context")
ap.add_argument("--server")
ap.add_argument("--token")
ap.add_argument("--insecure", action="store_true")
ap.add_argument("--in-pod", action="store_true", help="load mounted SA token/CA/server")
ap.add_argument("--dangerous", action="store_true", help="print dangerous-permission findings (default)")
ap.add_argument("--who-can", help="like 'create clusterrolebindings' -> roles granting it")
args = ap.parse_args()
if args.in_pod:
with open(os.path.join(SA_DIR, "token")) as f:
args.token = f.read().strip()
args.server = "https://%s:%s" % (os.environ.get("KUBERNETES_SERVICE_HOST", "kubernetes.default"),
os.environ.get("KUBERNETES_SERVICE_PORT", "443"))
use_kubectl = shutil.which("kubectl") and not (args.server and args.token)
kctx = []
if use_kubectl:
if args.kubeconfig:
kctx += ["--kubeconfig", args.kubeconfig]
if args.context:
kctx += ["--context", args.context]
# quick self check
whoami = subprocess.run(["kubectl"] + kctx + ["auth", "can-i", "--list"],
capture_output=True, text=True)
print("[*] Current identity capabilities (auth can-i --list):")
print(whoami.stdout[:2000] or whoami.stderr)
def get(resource, rest_path):
if use_kubectl:
return kubectl_get(kctx, resource)
return rest_get(args.server, args.token, args.insecure, rest_path)
croles = get("clusterroles", "/apis/rbac.authorization.k8s.io/v1/clusterroles")
roles = get("roles", "/apis/rbac.authorization.k8s.io/v1/roles")
findings = scan_roles(croles, "ClusterRole") + scan_roles(roles, "Role")
if args.who_can:
try:
wv, wr = args.who_can.split(None, 1)
except ValueError:
print("[!] --who-can format: '<verb> <resource>'"); return 2
print(f"\n[*] Roles granting '{wv} {wr}':")
for item in (croles.get("items", []) + roles.get("items", [])):
for rule in item.get("rules") or []:
if matches(rule.get("verbs", []), rule.get("resources", []), wv, wr):
md = item["metadata"]
print(f" {md.get('namespace','-')}/{md['name']}")
return 0
print(f"\n[*] {len(findings)} dangerous permission grant(s):")
print(f" {'KIND':12} {'NS':14} {'ROLE':32} {'PRIMITIVE':34} NOTE")
for kind, ns, name, prim, note, _v, _r in sorted(set(findings)):
print(f" {kind:12} {ns:14.14} {name:32.32} {prim:34.34} {note}")
print("\n[*] Next: resolve which subjects hold these via (cluster)rolebindings, then chain "
"(see k8s-rbac-escalation.md). nodes/proxy -> kubelet_exec.py.")
if __name__ == "__main__":
sys.exit(main())
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |