Browser Exploitation V8 logo

Browser Exploitation V8

OrganizationPopular
yaklang
browser-exploitation-v8

Browser and V8 exploitation playbook. Use when exploiting JavaScript engine vulnerabilities including JIT type confusion, incorrect bounds elimination, and V8 sandbox bypass to achieve renderer RCE and sandbox escape in Chrome/Chromium.

Overview

Publisheryaklang
Repositoryhack-skills
Skill namebrowser-exploitation-v8
Stars
2.2K
Forks
292
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by yaklang on GitHub. Read the source before you install it.

Installation

Install the Browser Exploitation V8 AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/yaklang/hack-skills.git /tmp/hack-skills
mkdir -p .claude/skills
cp -r /tmp/hack-skills/skills/browser-exploitation-v8 .claude/skills/browser-exploitation-v8
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Browser Exploitation V8 in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Browser Exploitation V8 on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Browser Exploitation V8 is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

SKILL: Browser / V8 Exploitation — Expert Attack Playbook

AI LOAD INSTRUCTION: Expert V8/Chrome exploitation techniques. Covers V8 compilation pipeline, JIT type confusion, addrof/fakeobj primitives, ArrayBuffer corruption, WASM RWX pages, V8 sandbox (pointer compression), and Chrome sandbox escape overview. Distilled from ctf-wiki browser sections, Project Zero research, and CTF competition patterns. Base models often confuse V8 object representation details and miss the pointer compression barrier.

0. RELATED ROUTING

Advanced Reference

Load V8_EXPLOITATION_PATTERNS.md when you need:

  • Detailed exploitation patterns and code templates
  • Heap layout manipulation and GC interaction
  • V8 sandbox bypass techniques
  • Object map confusion patterns

1. V8 ARCHITECTURE

Compilation Pipeline

JavaScript Source
    ↓ Parser
  AST (Abstract Syntax Tree)
    ↓ Ignition
  Bytecode (interpreted, profiling)
    ↓ Sparkplug (non-optimizing baseline, V8 ≥ 9.1)
  Baseline code (fast startup)
    ↓ Maglev (mid-tier, V8 ≥ 10.2)
  Mid-optimized code
    ↓ TurboFan (optimizing JIT)
  Optimized machine code (with speculative optimizations)
    ↓ Deoptimization (if speculation fails)
  Back to Ignition bytecode

Key V8 Concepts

ConceptDescription
Tagged pointersSMI (Small Integer): value << 1, HeapObject: ptr | 1
Pointer compressionV8 ≥ 8.0: objects addressed via 32-bit offset from cage base (4GB sandbox)
Maps (Hidden Classes)Define object shape: property names, types, offsets
Elements kindsInternal array type: PACKED_SMI_ELEMENTS, PACKED_DOUBLE_ELEMENTS, PACKED_ELEMENTS, etc.
Write barrierGC bookkeeping when heap pointers are written
Garbage collectionOrinoco GC: minor (Scavenge) and major (Mark-Compact)

Object Representation (64-bit, pointer compression)

HeapObject in V8 heap (compressed):
  +0x00: Map pointer (compressed, 32-bit offset)
  +0x04: Properties/Hash
  +0x08: Elements pointer (compressed)
  +0x0C: Length (for arrays)
  +0x10: Inline properties or backing store data

2. COMMON V8 BUG CLASSES

Bug ClassDescriptionExample
JIT Type ConfusionTurboFan assumes wrong type after optimizationSpeculative type guard eliminated, wrong operation applied
Incorrect Bounds EliminationJIT removes array bounds check based on wrong range analysisCheckBounds node eliminated → OOB access
Prototype Chain ConfusionOptimization assumes stable prototype, mutations invalidatePrototype change after optimization → wrong property access
Turbofan Reduction BugIncorrect strength reduction or constant foldingInteger overflow in range analysis
Race ConditionSharedArrayBuffer + worker thread raceType confusion via concurrent modification
Off-by-one in BuiltinBoundary error in built-in function implementationString/Array bounds
Typer BugIncorrect type range computation in TurboFanTyper says value is in [0, N] but can be N+1

Triggering JIT Optimization

javascript
function vuln(arr) {
    // ... vulnerable code path ...
}
// Force optimization by calling many times
for (let i = 0; i < 100000; i++) {
    vuln(arr);
}
// Or use V8 intrinsics (d8 only):
%OptimizeFunctionOnNextCall(vuln);
vuln(arr);

3. EXPLOITATION PRIMITIVES

addrof — Leak Object Address

javascript
// Goal: get the raw heap address of a JavaScript object
// Method: type confusion between object array and float array
// If we can confuse PACKED_ELEMENTS array with PACKED_DOUBLE_ELEMENTS:
// - Write object reference to element of object array
// - Read same element as double from confused float array
// - Float bits = compressed pointer of the object

function addrof(obj) {
    // Setup depends on specific bug
    // Typically: trigger type confusion so array reads obj ref as float
    object_array[0] = obj;
    return ftoi(confused_float_array[0]);  // float-to-int conversion
}

fakeobj — Create Fake Object Reference

javascript
// Goal: create a JS reference to an arbitrary heap address
// Method: reverse of addrof — write float (raw pointer bits) to float array,
//         read from confused object array → treated as object reference

function fakeobj(addr) {
    confused_float_array[0] = itof(addr);  // int-to-float conversion
    return object_array[0];                 // now a "pointer" to addr
}

Building Arbitrary R/W from addrof + fakeobj

javascript
// 1. Create a Float64Array with known layout
let rw_array = new Float64Array(0x100);
let rw_array_addr = addrof(rw_array);

// 2. Fake a Float64Array object at controlled address with modified backing_store
// 3. Corrupt backing_store pointer to target address
// 4. Read/write through the fake Float64Array → arbitrary R/W

function read64(addr) {
    // Set fake array's backing_store = addr
    write_to_fake_backingstore(addr);
    return fake_float64array[0];
}

function write64(addr, value) {
    write_to_fake_backingstore(addr);
    fake_float64array[0] = value;
}

4. OOB READ/WRITE VIA CONFUSED ARRAY BOUNDS

When TurboFan incorrectly eliminates bounds checks:

javascript
function trigger(arr, idx) {
    // TurboFan thinks idx is always < arr.length
    // But due to bug, idx can exceed bounds
    return arr[idx];  // OOB read
}

// OOB read adjacent memory (next heap object's metadata)
// OOB write to corrupt next object's map/elements/length

What's Adjacent in V8 Heap?

Objects are allocated sequentially in V8's young generation (new space). By controlling allocation order:

javascript
let arr1 = new Array(0x10);    // spray object
let arr2 = new Float64Array(0x10);  // target: adjacent to arr1
// OOB from arr1 can reach arr2's metadata
// Corrupt arr2's length → unconstrained OOB on arr2

5. ARRAYBUFFER ARBITRARY R/W

ArrayBuffer's backing store is a raw pointer to allocated memory. Corrupting it gives absolute memory R/W.

javascript
let ab = new ArrayBuffer(0x100);
let view = new DataView(ab);
// If we can overwrite ab's backing_store pointer:
// ab.backing_store = target_addr
// view.getFloat64(0) → reads 8 bytes from target_addr
// view.setFloat64(0, val) → writes to target_addr

V8 Sandbox (Pointer Compression) Impact

Since V8 ≥ 8.0 (pointer compression) and V8 sandbox (≥ 11.x):

  • ArrayBuffer.backing_store is a sandbox pointer (within the V8 cage, 4GB region)
  • Cannot directly point outside the V8 cage
  • Need sandbox escape to get full process memory access

6. WASM RWX PAGE

WebAssembly JIT code is placed on RWX (Read-Write-Execute) pages on some platforms.

javascript
// Allocate WASM module → JIT compiles to RWX page
let wasm_code = new Uint8Array([0x00, 0x61, 0x73, 0x6d, ...]);
let mod = new WebAssembly.Module(wasm_code);
let instance = new WebAssembly.Instance(mod);
// instance.exports.func → points to RWX page

// If we can find and write to this page:
// 1. addrof(instance) → find WASM instance object
// 2. Follow pointers: instance → jump_table_start → RWX page
// 3. Use arbitrary write to overwrite RWX page with shellcode
// 4. Call instance.exports.func() → executes shellcode

Modern Chrome: W^X enforcement means WASM pages are either RW or RX, not RWX simultaneously. JIT code is written in RW mode, then switched to RX. Exploitation requires finding a write window or using JIT spray.


7. V8 SANDBOX

Architecture (V8 ≥ 11.x)

Process Virtual Address Space:
┌──────────────────────────────────────┐
│  V8 Sandbox Cage (4GB region)        │
│  ├── V8 Heap (JS objects)            │
│  ├── ArrayBuffer backing stores      │
│  ├── WASM memory                     │
│  └── External pointer table          │
├──────────────────────────────────────┤
│  Process memory outside cage         │
│  ├── libc, Chrome code               │
│  ├── Stack                           │
│  └── Other allocations               │
└──────────────────────────────────────┘

Sandbox Escape Vectors

VectorMethod
External pointer tableCorrupt entries in the external pointer table to reference arbitrary addresses
WASM code pointerOverwrite WASM function entry to jump to controlled shellcode
JIT code corruptionWrite to JIT code page via race condition or confused pointer
Mojo IPC (Chrome)Exploit Chrome IPC to attack browser process from compromised renderer
Backing store seal bypassFind type confusion to get unsandboxed pointer

8. CHROME SANDBOX ESCAPE (OVERVIEW)

After renderer RCE (via V8 exploit), the process is still sandboxed. Full compromise requires:

StageTargetExample
Renderer exploitV8 / Blink DOMType confusion → shellcode
IPC/Mojo bugChrome IPC layerUse-after-free in Mojo interface
Browser process exploitPrivileged browser processCode execution outside sandbox

Mojo interfaces (Chrome's IPC) expose attack surface: find UAF or type confusion in Mojo message handlers.


9. TOOLS

bash
# V8 debugging
d8 --allow-natives-syntax exploit.js  # Enable V8 intrinsics (%DebugPrint, etc.)
d8 --trace-turbo exploit.js           # Dump TurboFan IR
d8 --print-opt-code exploit.js        # Print optimized machine code

# Turbolizer: visual TurboFan IR graph
# Chrome DevTools Memory panel: heap snapshots

# Build V8 for debugging
git clone https://chromium.googlesource.com/v8/v8.git
gclient sync
gn gen out/debug --args='is_debug=true v8_enable_sandbox=false'
ninja -C out/debug d8

10. DECISION TREE

V8 vulnerability identified
├── Bug type?
│   ├── JIT type confusion → trigger optimization, confuse array element kinds
│   ├── Bounds check elimination → OOB read/write on array
│   ├── Typer bug → incorrect range leads to OOB
│   └── Builtin bug → direct memory corruption primitive
├── Build primitives
│   ├── Can confuse object array ↔ float array?
│   │   └── addrof + fakeobj → arbitrary R/W within V8 heap
│   ├── OOB on array?
│   │   └── Corrupt adjacent object (length/backing_store) → expand to full R/W
│   └── Direct write primitive?
│       └── Target WASM instance or ArrayBuffer metadata
├── V8 sandbox enabled?
│   ├── YES (modern Chrome) →
│   │   ├── R/W limited to V8 cage (4GB)
│   │   ├── Need sandbox escape: external pointer table corruption,
│   │   │   WASM code pointer overwrite, or Mojo bug
│   │   └── Then proceed to shellcode execution
│   └── NO (older V8, CTF, d8) →
│       ├── Corrupt ArrayBuffer backing_store → absolute R/W
│       └── Overwrite WASM RWX page → shellcode
├── Code execution method
│   ├── WASM RWX page available? → write shellcode, call WASM func
│   ├── JIT code writable? → overwrite JIT code
│   └── ROP needed? → corrupt stack or return address
└── Full browser exploit chain
    ├── Stage 1: V8 bug → renderer RCE
    ├── Stage 2: Mojo IPC bug → browser process compromise
    └── Stage 3: OS-level escalation (if needed)

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Browser Exploitation V8 AI skill do?

Browser and V8 exploitation playbook. Use when exploiting JavaScript engine vulnerabilities including JIT type confusion, incorrect bounds elimination, and V8 sandbox bypass to achieve renderer RCE and sandbox escape in Chrome/Chromium.

Why use Browser Exploitation V8 on TypingMind?

Because you install it once and use it with any model. Browser Exploitation V8 is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Browser Exploitation V8 in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/yaklang/hack-skills/tree/main/skills/browser-exploitation-v8. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Browser Exploitation V8?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Browser Exploitation V8?

As many as you like. As long as a model supports skills, you can use Browser Exploitation V8 with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Browser Exploitation V8 AI skill free?

Yes. It is published on GitHub by yaklang under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇