SKILL: Bug Identification

Metadata

• Skill Name: bug-identification
• Folder: offensive-bug-identification
• Source: https://github.com/SnailSploit/offensive-checklist/blob/main/bug-identification.md

Description

Systematic bug identification methodology: source code review patterns, black-box testing strategies, taint analysis, dangerous function hunting, data flow tracing, and automated scanning setup. Use for code audits, bug bounty triage, or building vulnerability identification pipelines.

Trigger Phrases

Use this skill when the conversation involves any of: bug identification, code review, taint analysis, dangerous functions, data flow, source audit, black box, vulnerability identification, static analysis, code audit, bug hunting

Instructions for Claude

When this skill is active:

1. Load and apply the full methodology below as your operational checklist
2. Follow steps in order unless the user specifies otherwise
3. For each technique, consider applicability to the current target/context
4. Track which checklist items have been completed
5. Suggest next steps based on findings

---

Full Methodology

Bug Identification

Overview

Bug identification is the process of discovering potential vulnerabilities in software through various techniques including static analysis, dynamic analysis, and fuzzing. This document outlines methodologies and tools for effective vulnerability research.

For practical exploit development, see Exploit Development.

flowchart TD
    BugId["Bug Identification"]

    %% Main Methods
    Static["Static Analysis"]
    Dynamic["Dynamic Analysis"]
    Fuzzing["Fuzzing"]
    AI["AI-Assisted"]

    %% Static Analysis Methods
    CodeReview["Manual Code Review"]
    RevEng["Reverse Engineering"]
    PatchDiff["Patch Diffing"]
    StaticTools["Static Analysis Tools"]
    SBOM["Supply Chain Analysis"]

    %% Dynamic Analysis Methods
    DebugTrace["Debugging/Tracing"]
    DBI["Dynamic Binary Instrumentation"]
    Taint["Taint Analysis"]
    SymExec["Symbolic Execution"]
    Snapshot["Snapshot Analysis"]

    %% Fuzzing Methods
    DumbFuzz["Dumb Fuzzing"]
    SmartFuzz["Smart Fuzzing"]
    EvoFuzz["Evolutionary Fuzzing"]
    LLMFuzz["LLM-Guided Fuzzing"]

    %% AI Methods
    LLMTriage["LLM Crash Triage"]
    MLPattern["ML Pattern Recognition"]
    AutoVariant["Automated Variant Analysis"]

    %% Connections
    BugId --> Static
    BugId --> Dynamic
    BugId --> Fuzzing
    BugId --> AI

    Static --> CodeReview
    Static --> RevEng
    Static --> PatchDiff
    Static --> StaticTools
    Static --> SBOM

    Dynamic --> DebugTrace
    Dynamic --> DBI
    Dynamic --> Taint
    Dynamic --> SymExec
    Dynamic --> Snapshot

    Fuzzing --> DumbFuzz
    Fuzzing --> SmartFuzz
    Fuzzing --> EvoFuzz
    Fuzzing --> LLMFuzz

    AI --> LLMTriage
    AI --> MLPattern
    AI --> AutoVariant

    %% Combinations
    Taint -.-> Fuzzing
    SymExec -.-> Fuzzing
    RevEng -.-> Fuzzing
    AI -.-> Fuzzing
    AI -.-> Static

    class BugId primary

Vulnerability Research Methodology

Phase 1: Reconnaissance

• Target Enumeration: Identify version, dependencies, configuration
• Attack Surface Mapping: List all input vectors, APIs, protocols
• Documentation Review: RFCs, specifications, developer docs
• Prior Art Analysis: CVE database, exploit-db, bug trackers

Phase 2: Static Analysis

• Source Review: If available, focus on parsing/validation code
• Binary Analysis: Reverse engineering with Ghidra/IDA
• Patch Diffing: Compare vulnerable vs patched versions
• SBOM Analysis: Check third-party component vulnerabilities

Phase 3: Dynamic Analysis

• Behavioral Analysis: Monitor syscalls, network, file I/O
• Debugging: Trace execution paths with controlled input
• Instrumentation: Coverage-guided exploration
• Taint Analysis: Track input propagation

Phase 4: Fuzzing

• Corpus Generation: Create valid seed inputs
• Harness Development: Isolate target functionality
• Coverage Monitoring: Identify untested code paths
• Crash Triage: Classify and prioritize findings

Phase 5: Exploitation

• Primitive Development: Convert bug to reliable primitives
• Mitigation Bypass: Defeat ASLR, DEP, CFG, etc.
• Payload Development: Create working exploit
• Weaponization: Package for real-world use (if authorized)

Attack Surface Identification

Before diving into specific bug hunting techniques, it's essential to understand where to look for vulnerabilities.

Windows User Mode

• Shared Memory
• RPC
• Named Pipes
• File & Network IO
• Windows Messages
• For authentication-related vulnerabilities, see Windows Auth

Kernel

• Device Drivers
  • Many third-party software with drivers to target
  • Can accept arbitrary user input via the IOCTL interface
  • Also performs actions when we open,close handles to it
• OS
  • Drivers that handle hardware and user input
  • Intercepts/transitions from user to kernel
• Modern Linux interfaces (hotspots)
  • io_uring: SQE size/offset confusions, submission/completion race windows, kernel copy‑sizes derived from user buffers
  • userfaultfd: cross‑thread write‑what‑where and TOCTOU primitives during fault handling
  • seccomp user‑notifier: confused‑deputy patterns in broker processes; notifier time‑of‑check vs time‑of‑use gaps
• Hyper-V & VTL Interfaces – On many modern Windows 11 systems (especially 24H2 on supported hardware), Virtualization‑Based Security and VTL1 are enabled or easily enabled by policy. Treat the hypervisor surface (e.g., hvix64.exe and synthetic MSRs) as a common kernel target, and verify VBS/HVCI status on the host before assuming defaults.

Drivers

• DriverEntry: registers for any callbacks, setup structure, etc
• I/O Handlers: handlers that get called when a process attempts to open,close,etc the driver, IOCTL allows driver functionality to be called from user processes
• Practical triage example (CVE‑2025‑8061):
  • IOCTL handlers that accept a fixed‑size struct and pass a user‑controlled PHYSICAL_ADDRESS directly to MmMapIoSpace
  • then memcpy out/in mapped memory (sometimes via wrappers that swap src/dst) indicate physical memory read/write primitives.
  • Similarly, unguarded MSR read/write paths yield RDMSR/WRMSR primitives.
• See the Lenovo LnvMSRIO.sys case study in windows-kernel.md

eBPF & XDP

• BPF helpers and verifier: pointer leaks, verifier bypass, JIT bugs
• User‑entry vectors: bpf() syscall, privileged pods in Kubernetes, Cilium datapath
• Tooling: bpftool, verifier logs, bpftrace scripts for quick triage
• CO‑RE skeletons (bpftool gen skeleton) simplify packaging portable tracing probes.
• BPF LSM hooks allow low‑overhead coverage feedback on security‑critical kernel paths; export events with trace_pipe.

Container & Micro‑VM Surface

• Namespace/cgroup escapes, device‑mapper abuse, races in snapshotting backends (e.g., overlayfs)
• Micro‑VM hypercalls in Firecracker, CloudHypervisor, Kata Containers
• For detailed container exploitation techniques, see Container

Cloud‑Native & IAM Bugs

• Misconfigured IAM policies, privilege‑escalating API actions (AWS sts:AssumeRole, Azure Golden SAML)
• SSRF paths into metadata services (169.254.169.254, IMDSv2 bypass techniques)
• Race conditions in managed control‑plane components (Kubernetes API server, AWS Lambda workers)
• Kubernetes Attack Vectors: look at kubernetes for a deeper checklist
• Serverless Vulnerabilities:
  • Lambda layer poisoning
  • Function URL authentication bypass
  • Event injection through SQS/SNS/EventBridge
  • Cold start race conditions

Network / Transport Protocol Parsers

• QUIC / HTTP/3: coalesced frames, reorder/timing corner cases; verify against RFC 9000 (QUIC) and RFC 9114 (HTTP/3)
• HTTP/2: stream state machine desync; flow‑control integer edge cases (RFC 7540)
• gRPC / Protobuf: length truncation across language FFI, map/list coercion; see gRPC framing and protobuf varint rules
• GraphQL: input coercion and resolver recursion limits; check GraphQL spec for type coercion semantics

WebAssembly Runtimes

• WASM JIT optimization bugs in V8, Wasmtime, Wasmer
• WASI sandbox escapes through host‑call interfaces
• Typed‑Func‑Refs, GC, Tail‑calls, Memory64 expand type/bounds confusion surface. See the WebAssembly proposals status page for current rollout and engine adoption.
• Checklist:
  • validate table element types/import signatures/hostcall marshalling
  • fuzz mixed 32/64-bit memories.
  • Fuzzing tip: compile native libs to WASM for fast, deterministic mutation cycles

Browser / JS Engine Exploitation

Modern V8 Architecture (2024-2025)

V8 now uses a multi-tier JIT pipeline with distinct exploitation characteristics:

• Ignition (Interpreter): Bytecode interpreter; rarely targeted directly
• Maglev (Mid-tier JIT): Introduced Chrome 115+; simpler IR than TurboFan
• TurboFan (Optimizing JIT): Aggressive optimization; traditional exploitation target
• Turboshaft: New IR replacing TurboFan internals; different optimization patterns create new bug classes
  • Type lattice changes affecting confusion bugs
  • Maglev → Turboshaft transition paths expose state inconsistencies
  • Node-based to block-based IR transition

V8 Maglev Exploitation

• Integer overflow in Maglev's fast-path arithmetic
• Corrupted HeapNumber backing store via Maglev bounds check bypass
• Map/ElementsKind confusion in polymorphic inline caches

WebAssembly JSPI (JavaScript Promise Integration)

• Stack Heap Spray: Suspended WASM stacks allocated on heap; predictable layout
• Type Confusion: WebAssembly.Suspending wrapper type mismatch
• Info Leak: Stack pointers exposed through Promise resolution chains
• Sandbox Escape: JSPI bridges JS/WASM boundary; bypass traditional WASM isolation

Spectre-BHB Browser Mitigations

• Chrome 120+: Site Isolation per-frame; shared array buffer restrictions
• Firefox 122+: Process-per-site with BHI fences in JIT trampolines
• Safari 17.4+: WebKit JIT speculation guards on type checks

Site Isolation Plus

• Frame-level process isolation: Each cross-origin frame in separate process
• Cross-origin memory protection: Hardware-backed memory isolation
• New IPC attack surface: Mojo interface exploitation required for escapes
• Renderer → Browser requirements: Need Mojo race or type confusion
• New Info-Leak Requirements:
  • Traditional SharedArrayBuffer + Atomics timing attacks less reliable
  • Need alternative side-channels: CSS timing, WebGL shader execution, AudioContext
  • Cross-origin info leaks require chaining multiple primitives

Practical Browser Exploitation Workflow

1. Target Selection:

   • V8 Maglev for Chrome/Edge (faster development cycle = more bugs)
   • JSC for Safari (less scrutiny than V8)
   • SpiderMonkey for Firefox (IonMonkey/Warp still viable)

2. Primitive Development:

   • addrof: Leak object addresses (info leak)
   • fakeobj: Craft fake object (type confusion)
   • arbread/arbwrite: Arbitrary memory access
   • shellcode: RWX page or WASM JIT abuse

3. Sandbox Escape:

   • Mojo IPC race conditions
   • GPU process exploitation via WebGL
   • Utility process TOCTOU (Chrome's new architecture)

4. Post-Exploitation:

   • Chrome: Target browser process via Mojo
   • Safari: XPC service exploitation for sandbox escape
   • Firefox: Target parent process via IPC

Firmware & Embedded

• UEFI DXE driver flaws, BMC web console auth bypass, ECU/CAN message injection
• BLE & Zigbee stack overflows, heap exploits in btstack, lwIP

macOS / Apple‑Silicon Kernel

• IOKit user‑client input validation, IOMFB allocator corner‑cases
• Hypervisor.framework fuzzing with hv_fuzz

Mobile Platforms (iOS/Android)

iOS 17+ Exploitation

• PAC Bypass: Pointer Authentication Code bypass via signing gadgets
• PPL Bypass: Page Protection Layer exploitation for kernel r/w
• Secure Enclave: SEP exploitation via malformed Mach messages
• Neural Engine: ANE kernel driver attack surface

Android 14+ Exploitation

• MTE (Memory Tagging): Probabilistic bypass with tag collisions
• GKI (Generic Kernel Image): Vendor hooks as attack surface
• Scudo Hardening: Heap exploitation with hardened allocator
• Hardware Attestation: Keymaster/StrongBox TEE attacks

Cross-Platform Mobile

• Flutter: Dart VM type confusion, FFI boundary issues
• React Native: JavaScript bridge serialization bugs
• Unity: IL2CPP memory corruption, native plugin vulnerabilities

Supply Chain Attack Surface

Package Manager Vulnerabilities

• Dependency Confusion: Internal vs public package name conflicts
• Typosquatting: Similar package names (numpy vs numpi)
• Manifest Manipulation: Lock file poisoning, version pinning bypass
• Build-time Injection: Malicious install scripts, post-install hooks

CI/CD Pipeline Analysis

• GitHub Actions: Workflow poisoning via PR from forked repos
• Jenkins: Groovy script injection, plugin vulnerabilities
• Docker: Build argument exploitation, base image substitution
• Secrets Exposure: Environment variables in build logs, artifact leakage

AI & LLM Application Security

• Prompt‑injection, sandbox boundary escapes, hidden‑channel data exfil
• See AI Security for a deeper checklist

Confidential‑Computing / TEE Surface

• Intel TDX: diff tdx.ko or tdx_psci.c between kernel LTS branches to spot new GPA→HPA validation checks.
• AMD SEV‑SNP: look for unchecked VMGEXIT leafs in PSP firmware; sevtool --decode helps locate IDA entry points.
• Arm CCA / RMM: analyze SMC handlers inside Realm Management Monitor (RMM) EL3 firmware.
• Cloud offerings (Azure CCE, Google C3): focus on paravirtualised MMIO and attestation report flows exposed to guests.
• For TEE-specific exploitation, see Secure Enclaves

GPU & vGPU Surface

• HGX HMC (verify CVE/advisories): research indicates malformed NVLINK‑C2C packets can corrupt HMC register space; confirm against vendor advisories for the specific platform.
• vGPU manager IOCTLs: diff nvidia‑vgpu‑mgr monthly; watch VGPU_PLUGIN_IOCTL_GET_STATE and similar calls for unchecked buffers.
• LeftoverLocals info‑leak: contiguous VRAM allocations can leak data from prior tenants in multi‑tenant AI clusters.

Hardware Security Attack Surface

Side-Channel Analysis

• Power Analysis: DPA/SPA attacks on cryptographic operations
• Electromagnetic (EM): Near-field probing of processor emissions
• Timing Attacks: Cache timing, branch prediction analysis
• Acoustic: Key extraction via CPU sound emissions

Fault Injection

• Voltage Glitching: Brown-out attacks on secure boot
• Clock Glitching: Skip instruction execution
• Laser Fault Injection (LFI): Targeted bit flips
• EM Pulse Injection: Wider area fault induction

Hardware Implants & Supply Chain

• PCB Modification: Added components, trace rerouting
• Firmware Backdoors: UEFI/BMC persistent implants
• Hardware Trojans: Malicious logic in ICs
• DMA Attacks: PCIe, Thunderbolt, FireWire exploitation

EDR Driver Vulnerability Research

Common vulnerability types in EDR drivers

• Authorization bypass issues
• Memory corruption in IOCTL handlers
• Race conditions in driver communication
• Improper input validation
• For detailed EDR analysis techniques, see EDR

Research methodology

1. Identify accessible driver interfaces
2. Reverse engineer IOCTL/message handlers
3. Analyze authorization mechanisms
4. Test for input validation flaws
5. Look for race conditions and memory corruption

Tools for driver analysis

• IDA Pro / Ghidra for reverse engineering
• WinDbg for dynamic analysis
• Process Monitor for behavior analysis
• Custom fuzzing tools for interface testing

Quick triage rubric (post‑crash)

• Buffer overflow vs UAF: check access type, allocation lifetime, and red‑zones (ASan/KASAN reports)
• Integer issues: trace size/length and allocation math; look for truncation/casts
• Logic bugs: unexpected state transitions without memory errors; validate auth/flags
• Info‑leaks: uninitialized reads, OOB reads, pointer/string formatters

Coverage‑first recon checklist

• Produce one baseline coverage run (e.g., drcov, Intel® PT, or Lighthouse import)
• Identify cold paths reachable from attacker inputs
• Seed corpus: include minimal valid examples that traverse target parsers
• Enable lightweight oracles (ASan/UBSan/KASAN) where feasible to maximize signal

Static Analysis Methods

Static analysis examines code without execution to identify potential vulnerabilities.

Manual Code Review

• Installing the target application and examining its structure
• Enumerating the ways to feed input to it
• Examine the file formats and network protocols that the application uses
• Locating logical vulnerabilities or memory corruptions
• For Windows-specific techniques, see Windows Kernel
• For Linux-specific techniques, see Linux

Patch Diffing

Patch diffing compares vulnerable and patched versions of binaries to identify security changes.

What is Patch Diffing

Patch diffing is a technique to identify changes across versions of binaries related to security patches. It compares a vulnerable version of a binary with a patched one to highlight the changes, helping to discover new, missing, and interesting functionality across versions.

Benefits

• Single Source of Truth: Without a CVE blog post or sample POC, a patch diff can be the only source of information to determine changes and deduce the original issue.
• Vulnerability Discovery: While understanding the original issue, you may discover additional vulnerabilities in the troubled code area.
• Skill Development: Patch diffing provides focused practice in reverse engineering and helps build mental models for various vulnerability classes.

Challenges

• Asymmetry: Small source code changes can drastically affect compiled binaries.
• Finding Security-Related Changes: Security patches often include other changes like new features, bug fixes, and performance improvements.
• Minimizing Noise:
  • Diff the correct binaries to avoid analyzing unrelated updates
  • Reduce the time delta between compared versions
  • Use binary symbols when available to add precision to comparisons

Tools

• IDA Pro with plugins like DarunGrim and Diaphora
• BinDiff Works with analysis output from IDA or Ghidra
• Ghidriff: Ghidra binary diffing engine
• Radare2 (radiff2)
• Ghidra Version Tracking Tool
• Ghidra 11 built-in Partial Match Correlator

Patch Diffing Workflow

The process of patch diffing typically follows these steps:

1. Preparation

   • Create a diffing session
   • Load binary versions (vulnerable and patched)
   • Ensure binaries pass preconditions
   • Run auto-analysis on both binaries

2. Evaluation

   • Run correlators to find similarities
   • Generate associations between binaries
   • Evaluate matches between functions
   • Accept matching functions
   • Analyze differences until sufficient understanding is reached

3. Function Analysis

   • Identify new functions: Functions in the patched binary with no match in the original
   • Identify deleted functions: Functions in the original binary with no match in the patched version
   • Identify changed functions: Functions that exist in both versions but have been modified
   • Focus on functions with security relevance (often indicated by their names or based on CVE descriptions)

4. Interpreting Results

   • New functions often indicate added security checks or validation
   • Changed functions may show modified logic for handling edge cases
   • Correlate changes with public CVE information when available
   • Remember that patches are not necessarily atomic - multiple issues may be fixed in one update

When using Ghidra's Version Tracking:

• Use "Show Only Unmatched Functions" filter to identify new or deleted functions
• Look for functions with a similarity score below 1.0 to find modified functions
• Examine the modified functions to understand what security checks were added

Starting with Ghidra 11 (December 2024) a built-in Partial Match Correlator covers most PatchDiffCorrelator use-cases; install the plugin only if you need bulk-mnemonics scoring.

Case Study: 7‑Zip Symlink Path Traversal

• Target: 7‑Zip 24.09 (vulnerable) → 25.00 (fixed)
• File of interest: CPP/7zip/UI/Common/ArchiveExtractCallback.cpp
• High‑signal edits: absolute‑path detection and link‑path validation for WSL/Linux symlinks converted on Windows.

Minimal security‑relevant diff (simplified):

-bool IsSafePath(const UString &path)
+static bool IsSafePath(const UString &path, bool isWSL)
{
  CLinkLevelsInfo levelsInfo;
-  levelsInfo.Parse(path);
+  levelsInfo.Parse(path, isWSL);
  return !levelsInfo.IsAbsolute
      && levelsInfo.LowLevel >= 0
      && levelsInfo.FinalLevel > 0;
}

+bool IsSafePath(const UString &path);
+bool IsSafePath(const UString &path)
+{
+  return IsSafePath(path, false); // isWSL
+}

-void CLinkLevelsInfo::Parse(const UString &path)
+void CLinkLevelsInfo::Parse(const UString &path, bool isWSL)
{
-  IsAbsolute = NName::IsAbsolutePath(path);
+  IsAbsolute = isWSL ? IS_PATH_SEPAR(path[0]) : NName::IsAbsolutePath(path);
  LowLevel = 0;
  FinalLevel = 0;
}

Root cause (logic):

• Linux/WSL symlink data containing a Windows‑style path (e.g., C:\...) was treated as relative by the Linux absolute‑path check, setting linkInfo.isRelative = true.
• SetFromLinkPath prefixed the symlink’s zip‑internal directory when building relatPath, letting IsSafePath(relatPath) pass despite an absolute Windows target.
• A subsequent “dangerous link” guard checked _item.IsDir; non‑directory symlinks skipped the validation.
• Result: symlink creation to arbitrary absolute Windows paths; extracted files written into the link target.

Practical triage checklist:

• Search this file for: IsSafePath, CLinkLevelsInfo::Parse, SetFromLinkPath, CloseReparseAndFile, FillLinkData, CLinkInfo::Parse, _ntOptions.SymLinks_AllowDangerous.
• Verify absolute‑path detection across OS semantics (Linux vs Windows) and that relative/absolute status cannot be desynced by mixed‑style paths.
• Ensure “dangerous link” checks run for both files and directories; avoid _item.IsDir short‑circuiting validation for file symlinks.
• Confirm IsSafePath evaluates the final target path after concatenations; normalize before validation.

Quick repro (Windows, developer mode or elevated):

• Create zip structure:
  • data/link → symlink to C:\Users\<USER>\Desktop
  • data/link\calc.exe → payload file
• If link is extracted first, subsequent writes follow the symlink into the absolute target directory.

Apple Patch Diffing