Offensive Fuzzing logo

Offensive Fuzzing

CommunityPopular
SnailSploit
offensive-fuzzing

Practical offensive fuzzing methodology covering target identification, fuzzer selection (AFL++, libFuzzer, Honggfuzz, Boofuzz, syzkaller), harness writing, corpus curation, mutation strategies, coverage measurement, and crash triage. Use when setting up or running fuzz campaigns against any target: file parsers, network protocols, kernel drivers, EDR engines, embedded firmware, or language runtimes.

Overview

PublisherSnailSploit
RepositoryClaude-Red
Skill nameoffensive-fuzzing
Stars
6K
Forks
775
Bundled files
Instructions only
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.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

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

Installation

Install the Offensive Fuzzing 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/SnailSploit/Claude-Red.git /tmp/Claude-Red
mkdir -p .claude/skills
cp -r /tmp/Claude-Red/Skills/fuzzing/offensive-fuzzing .claude/skills/offensive-fuzzing
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Offensive Fuzzing 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 Offensive Fuzzing 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 Offensive Fuzzing 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.

Offensive Fuzzing

Fuzzer Types

TypeCoverageSpeedTools
BlackBoxPoorFastPeach, Boofuzz
GreyBoxGoodFastAFL++, Honggfuzz, libFuzzer, WinAFL
SnapshotGoodFastestNyx, wtf, Snapchange
WhiteBoxBestSlowKLEE, QSYM, SymSan
EnsembleBestFastAFL++ + Honggfuzz + libFuzzer

GreyBox sub-variants: Directed (AFLGo, UAFuzz), Grammar (AFLSmart, Tlspuffin), Concolic (QSYM, Driller), Kernel (syzkaller, kAFL, wtf).

Core Workflow

Research target → Choose analyses → Build harness → Seed corpus → Instrument → Fuzz → Triage crashes → Report

1. Research Target

  • Map all input surfaces (files, network, IPC, syscalls, IOCTL)
  • Identify high-value areas: previously patched code, complex parsers, newly added code, input ingestion points
  • For kernel modules: look beyond copy_from_user — DMA-BUF ops, page fault handlers, VM operation structs, allocation callbacks

2. Instrument and Build

bash
# AFL++ (preferred for GreyBox)
CC=afl-clang-fast CXX=afl-clang-fast++ cmake -DCMAKE_BUILD_TYPE=Release .. && make -j

# libFuzzer + ASan/UBSan (C/C++)
cmake -DCMAKE_CXX_FLAGS="-fsanitize=fuzzer,address,undefined -O1 -g" ..

# CmpLog build for hard compares
AFL_LLVM_CMPLOG=1 CC=afl-clang-fast CXX=afl-clang-fast++ make clean all

Windows (MSVC): Project Properties → C/C++ → Address Sanitizer: Yes (/fsanitize=address)

3. Write Harness

libFuzzer (C++):

cpp
#include <cstdint>
#include <cstddef>
extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) {
    parse_or_process(data, size);
    return 0;
}

Honggfuzz HF_ITER (persistent mode — preferred for large targets):

cpp
#include "honggfuzz.h"
int main(int argc, char** argv) {
    initialize_target(); // runs once
    for (;;) {
        size_t len; uint8_t *buf;
        HF_ITER(&buf, &len);
        FILE* s = fmemopen(buf, len, "r");
        target_function(s);
        fclose(s);
        reset_target_state();
    }
}

AFL++ persistent mode (__AFL_LOOP):

cpp
while (__AFL_LOOP(10000)) {
    // re-read input and process
}

macOS IPC (Mach message fuzzing):

c
void *lib_handle = dlopen("libexample.dylib", RTLD_LAZY);
pFunction = dlsym(lib_handle, "DesiredFunction");

4. Build Seed Corpus

  • Pull from target's test suite, bug reports, and real-world samples
  • Web-crawl (Common Crawl) for file formats; filter by MIME type
  • Minimize: afl-cmin -i raw_corpus -o seeds -- ./target @@
  • Trim inputs: afl-tmin -i crash -o crash.min -- ./target @@

5. Launch Fuzzing

AFL++ parallel (primary + secondary with cmplog):

bash
afl-fuzz -M f1 -i seeds -o findings -x dict.txt -- ./target @@
afl-fuzz -S s1 -i seeds -o findings -c 0 -- ./target @@

libFuzzer:

bash
./target_libfuzzer corpus/ -max_total_time=3600 -workers=4

Binary-only (QEMU):

bash
afl-fuzz -Q -i seeds -o findings -- target.exe @@

Snapshot (AFL++ Nyx):

bash
NYX_MODE=1 AFL_MAP_SIZE=1048576 afl-fuzz -i seeds -o findings -- ./target_nyx @@

Ensemble (AFL++ + Honggfuzz sharing corpus):

bash
# Terminal 1
afl-fuzz -M fuzzer1 -i seeds -o sync_dir -- ./target @@
# Terminal 2
../honggfuzz/honggfuzz -i sync_dir/fuzzer1/queue -W sync_dir/hfuzz \
  --linux_perf_ipt_block -t 10 -- ./target ___FILE___

6. Monitor and Unstick

If progress stalls:

  • Enable CmpLog: -c 0 on AFL++ secondaries
  • Add dictionary: -x dict.txt or AFL_TOKEN_FILE
  • Switch to directed fuzzing (AFLGo) targeting specific BBs/functions
  • Use concolic assistance (QSYM, Driller) on hard branches
  • Snapshot the target to increase exec/s
  • AFL_MAP_SIZE=1048576, -L 0 for MOpt scheduler

7. Triage Crashes

bash
# 1. Minimize
afl-tmin -i crash -o crash.min -- ./target @@
# 2. Symbolize
ASAN_OPTIONS=abort_on_error=1:symbolize=1 ./target crash.min 2>asan.log
# 3. Hash + bucket
./cov-tool --bbids ./target crash.min > cov.hash
./bucket.py --key "$(cat cov.hash)" --log asan.log --out triage/

Sanitizer env quick reference:

ASAN_OPTIONS=abort_on_error=1:symbolize=1:detect_stack_use_after_return=1
UBSAN_OPTIONS=print_stacktrace=1:halt_on_error=1
TSAN_OPTIONS=halt_on_error=1:history_size=7
MSAN_OPTIONS=poison_in_dtor=1:track_origins=2

Oracle Selection

Bug ClassOracle
Memory safetyASan, HWASan (AArch64, lower overhead)
Uninitialized readsMSan
ConcurrencyTSan
Undefined behaviorUBSan
Type safetyTypeSan
Heap hardeningScudo Hardened Allocator
Logic bugsDifferential / idempotency oracles
Kernel memoryKASAN, KMSAN, KCSAN
Kernel UBKUBSan (CONFIG_UBSAN_TRAP=y)
CFIKCFI (-fsanitize=kcfi, Clang 18)
Binary-onlyQASAN (QEMU+ASan), DynamoRIO

Property oracle patterns:

  • Idempotency: f(x) == f(f(x))
  • Differential: compare two impls, bucket on output mismatch
  • Invariants: monotonic lengths, checksum equality, schema validation post-parse

Specialized Targets

Kernel (Linux) — syzkaller

json
{
  "target": "linux/arm64",
  "http": ":56700",
  "workdir": "/path/to/workdir",
  "kernel_obj": "/path/to/kernel",
  "image": "/path/to/rootfs.ext3",
  "sshkey": "/path/to/id_rsa",
  "procs": 8,
  "enable_syscalls": ["openat$module_name", "ioctl$IOCTL_CMD", "mmap"],
  "type": "qemu",
  "vm": { "count": 4, "cpu": 2, "mem": 2048 }
}
  • Limit enable_syscalls to deepen coverage on specific subsystems
  • Use syz-extract to pull constants for custom modules
  • Enable CONFIG_KASAN=y, CONFIG_KCFI=y, CONFIG_DEBUG_INFO_BTF=y
  • Use kcov filters and syz_cover_filter to direct coverage
  • Network fuzzing: inject via TUN/TAP + pseudo-syscalls (syz_emit_ethernet)
  • Crash decode: ./scripts/decode_stacktrace.sh vmlinux ... < dmesg.log

syzkaller repro:

bash
syz-execprog -repeat=0 -procs=1 -cover=0 -debug target.repro

EDR / Windows Scanning Engines

WTF snapshot harness skeleton (mpengine.dll / mini-filter):

cpp
g_Backend->SetBreakpoint("nt!KeBugCheck2", [](Backend_t *Backend) {
    const uint64_t BCode = Backend->GetArg(0);
    Backend->Stop(Crash_t(fmt::format("crash-{:#x}", BCode)));
});

FilterConnectionPort fuzzing:

cpp
HANDLE hPort;
FilterConnectCommunicationPort(L"\\PortName", 0, NULL, 0, NULL, &hPort);
FilterSendMessage(hPort, fuzzData, sizeof(fuzzData), NULL, 0, &bytesReturned);

IOCTL fuzzing pattern:

cpp
HANDLE hDev = CreateFile(L"\\\\.\\DeviceName", GENERIC_READ|GENERIC_WRITE, ...);
DeviceIoControl(hDev, ioctlCode, inputBuf, inputLen, outBuf, outLen, &ret, NULL);
  • Take snapshots after initialization, right before parse/dispatch loop
  • Use IDA Lighthouse for coverage visualization
  • Monitor: DRIVER_VERIFIER_DETECTED_VIOLATION (0xc4), IRQL_NOT_LESS_OR_EQUAL (0xa)
  • WinDbg: .symfix; !analyze -v; k; !heap -p -a @rax

Cross-platform mpengine.dll on Linux (loadlibrary + HF_ITER + Intel PT):

cpp
// Bypass Lua VM to avoid stability issues
insert_function_redirect((void*)luaV_execute_address, my_lua_exec, HOOK_REPLACE_FUNCTION);
for (;;) {
    HF_ITER(&buf, &len);
    ScanDescriptor.UserPtr = fmemopen(buf, len, "r");
    __rsignal(&KernelHandle, RSIG_SCAN_STREAMBUFFER, &ScanParams, sizeof ScanParams);
}

Rust

bash
# Full Rust fuzzing pipeline
cargo test                                         # 1. property tests
cargo +nightly miri test                           # 2. UB via interpreter
cargo +nightly careful test                        # 3. runtime bounds checks
cargo fuzz run fuzz_target_1 -- -max_total_time=3600  # 4. libFuzzer crashes
RUSTFLAGS="--cfg loom" cargo test --release        # 5. concurrency (if needed)
cargo fuzz coverage fuzz_target_1                  # 6. coverage report

Focus unsafe blocks on: Vec::from_raw_parts, unchecked indexing, transmute size mismatches, pointer arithmetic, FFI integer truncation.

Embedded / Binary-Only

  • LibAFL: Modular Rust framework; Unicorn engine, snapshot module, LBRFeedback (zero-instrumentation on Intel), SAND decoupled sanitization
  • Retrowrite / QASAN: Binary rewriting for coverage + ASan without source
  • Nautilus: Grammar-based fuzzing for structured formats

Language Ecosystems

  • Go 1.18+: go test -fuzz=Fuzz -run=^$ ./...
  • Python: Atheris (CPython native extension fuzzing)
  • Rust: cargo-fuzz or honggfuzz-rs
  • JS engines: Fuzzilli with extended instrumentation (__builtin_return_address(0) for PC tracking)
  • Wasm runtimes: wasmtime-fuzz, wafl for differential fuzzing across V8/Wasmer/Wasmtime
  • Smart contracts: Echidna, Foundry-fuzz (Solidity); Move-Fuzz (Aptos/Sui)

CI/CD Integration

yaml
- name: Build with afl-clang-fast
  run: CC=afl-clang-fast make -j
- name: Fuzz (smoke, 15 min)
  run: timeout 15m afl-fuzz -i seeds -o findings -- ./target @@ || true
- name: Upload crashes
  if: always()
  uses: actions/upload-artifact@v4
  with:
    path: findings/**/crashes/*

Use ClusterFuzzLite for persistent continuous fuzzing; cache corpora between runs.

Crash Analysis Quick Reference

Linux:

bash
ulimit -c unlimited && sysctl -w kernel.core_pattern=core.%e.%p
gdb -q ./target core.* -ex 'bt' -ex 'info reg' -ex q
addr2line -e ./target 0xDEADBEEF

Windows:

powershell
# Enable local dumps
New-Item 'HKLM:\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps' -Force
# PageHeap
gflags /p /enable target.exe /full

Kernel KASAN/KMSAN:

bash
dmesg -T | egrep -i 'kasan|kmsan' -A 60
./scripts/decode_stacktrace.sh vmlinux /lib/modules/$(uname -r)/build < dmesg.log

Reproducibility: pin CPU governor, disable ASLR only where safe, fix RNG seeds, save input sequences in persistent mode, record binary hashes and sanitizer options with every crash.

Tool Index

ToolUse Case
AFL++General GreyBox, CmpLog, MOpt, Nyx
HonggfuzzIntel PT, crash detection, HF_ITER
libFuzzerIn-process, source available
syzkallerLinux/Windows kernel syscall fuzzing
wtfSnapshot fuzzing, Windows targets
NyxAFL++ snapshot mode (Intel PT)
SnapchangeAWS snapshot fuzzing
LibAFLCustom Rust fuzzing framework
AFLGoDirected fuzzing to target BB/function
kAFLKernel + OS fuzzing
JackalopeBinary coverage-guided (Windows/macOS)
cargo-fuzzRust libFuzzer integration
AtherisPython fuzzing
NautilusGrammar-based fuzzing
AFLTriageAutomated crash triage
afl-covCoverage analysis for AFL++
ClusterFuzzDistributed fuzzing infrastructure

Frequently asked questions

What does the Offensive Fuzzing AI skill do?

Practical offensive fuzzing methodology covering target identification, fuzzer selection (AFL++, libFuzzer, Honggfuzz, Boofuzz, syzkaller), harness writing, corpus curation, mutation strategies, coverage measurement, and crash triage. Use when setting up or running fuzz campaigns against any target: file parsers, network protocols, kernel drivers, EDR engines, embedded firmware, or language runtimes.

Why use Offensive Fuzzing on TypingMind?

Because you install it once and use it with any model. Offensive Fuzzing 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 Offensive Fuzzing in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/SnailSploit/Claude-Red/tree/main/Skills/fuzzing/offensive-fuzzing. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Offensive Fuzzing?

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 Offensive Fuzzing?

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

Is the Offensive Fuzzing AI skill free?

Yes. It is published on GitHub by SnailSploit 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 👇