dexmachina 0.1.0

Android pentesting tool manager — install, update, and version-lock your toolkit

DexMachina

Android pentest environment manager — install, sync, diagnose, and repair your entire mobile security toolkit from one CLI.

DexMachina is evolving from a tool manager into a full Android penetration environment: one command to get adb, frida, jadx, apktool, and the rest of your kit installed, version-locked, and working together. It solves dependency hell — tools like objection, r2frida, and frida-tools all require the exact same frida version, and apktool needs a compatible JDK.

Why DexMachina

Most Android pentest setups grow from shell history, old notes, and a handful of one-off install scripts. DexMachina turns that into a repo-local, reproducible environment: curated profiles, version pins, a lockfile, Frida runtime isolation, device readiness checks, and doctor/fix when the setup drifts.

It is not a replacement for MobSF, Corellium, Burp, or manual reverse-engineering judgment. It is the glue layer that gets a consistent workstation and rooted emulator/device ready faster than rebuilding the same adb/frida/objection/jadx setup for every engagement.

Trust and Support

• Trust model: THREAT_MODEL.md
• Vulnerability reporting: SECURITY.md
• Contributing and adding tools: CONTRIBUTING.md
• Release process: docs/RELEASE.md
• Branch protection checklist: .github/BRANCH_PROTECTION.md
• Release history: CHANGELOG.md

Support matrix

Host	Status	Notes
Ubuntu/Debian Linux	Supported in CI	Primary development target; apt helpers expect Debian-style systems.
macOS Intel/Apple Silicon	Supported in CI	Homebrew-backed tools require brew.
Windows native	Unit-tested in CI	Core CLI paths are tested; some upstream tools have limited Windows support.
WSL2	Not officially supported yet	Often works for host-only commands, but USB/emulator workflows need manual validation.

Known good device setups

Setup	Status
Stock AVD	CI smoke-tested for adb/emulator basics
Rooted AVD	CI integration job boots an AVD and runs device ready + frida-ps -U
Genymotion	Expected to work, needs published validation
Corellium	Expected to work, needs published validation

Install

git clone https://github.com/Codeblin/dexmachina.git
cd dexmachina
pip install -e ".[dev]"

Run via the installed entrypoint or as a module:

dexmachina --help          # full ASCII banner + commands
dexmachina status          # compact banner + tool table
python -m dexmachina doctor

Set DEXMACHINA_NO_BANNER=1 or pass --no-banner to suppress the ASCII art (useful for scripts/CI).

Use it as a self-contained pentest environment (recommended)

Turn any repo into a portable Android pentest kit. Tools are downloaded into ./.dexmachina/tools/, version-locked, and added to your PATH on demand — the heavy binaries stay gitignored, while the config and lockfile are committed for reproducibility.

cd my-engagement-repo

dexmachina init                 # create dexmachina.toml + .gitignore + .dexmachina/tools
dexmachina up                   # install the profile + set up frida (one command)
dexmachina console              # interactive pentest REPL (recommended)
dexmachina shell                # …or a raw subshell with every tool on PATH

Inside dexmachina shell, tools like jadx, apktool, objection, and frida are directly on your PATH. Type exit to leave.

Pentest a rooted emulator — full walkthrough

Assuming a rooted emulator/device is running with USB debugging (e.g. a Genymotion / rooted AVD / Corellium instance, visible via adb devices):

# 0) one-time: build the dynamic-analysis kit (adb, frida, objection, …)
dexmachina up --profile dynamic

# 1) drop into the interactive console
dexmachina console

Then drive the engagement from the DexMachina console:

dexmachina [no-device | no-target]> devices          # auto-selects a single device
dexmachina [emulator-5554 | no-target]> ready        # push + start frida-server
dexmachina [emulator-5554 | no-target]> apps owasp   # find your target (filter)
dexmachina [emulator-5554 | no-target]> target com.example.app
dexmachina [emulator-5554 | com.example.app]> status # device + frida + target summary
dexmachina [emulator-5554 | com.example.app]> proxy 10.0.2.2:8080   # route HTTPS to Burp
dexmachina [emulator-5554 | com.example.app]> hook   # SSL pinning + root bypass (spawn)
dexmachina [emulator-5554 | com.example.app]> objection             # interactive explorer
dexmachina [emulator-5554 | com.example.app]> logcat com.example.app   # watch logs
dexmachina [emulator-5554 | com.example.app]> exit

The console keeps live session state (selected device + target app), so you don't retype serials/packages. Every verb maps to the same engines the flat CLI uses (device ready, bypass, objection, …) — just faster to chain.

Console commands

Verb	Description
devices / use <serial>	List devices / select the active one
apps [filter]	List installed apps (running ones marked)
target <package>	Set the app under test (accepts substrings)
status	Device + frida runtime + frida-server + target
ready	Push & start frida-server matching local frida
hook [--spawn] / bypass [ssl|root|all] [--spawn]	Attach SSL/root bypass to a running target; optionally spawn
objection [args]	Open Objection's explorer on the target
proxy <host:port> / proxy off	Set/clear the device global HTTP proxy
logcat [filter]	Stream device logs
screenshot [file.png]	Capture the screen to a local PNG
adb <args> / adbshell	Run adb / open an interactive adb shell
pull / push	Copy files off/onto the device
ps / run <tool> [args]	List processes / run any DexMachina tool
clear · help · exit	Housekeeping

Reproduce the exact kit on another machine:

dexmachina lock                 # write dexmachina.lock.toml (commit this)
# … teammate clones the repo …
dexmachina restore              # install the locked tools + frida runtime

Quick Start

# One-command environment (auto-inits a repo-local workspace inside a git repo)
dexmachina up --profile dynamic

# See / inspect profiles
dexmachina profile list
dexmachina profile show static

# Drive an engagement interactively
dexmachina console              # pentest REPL with live device + target state

# Put every installed tool on PATH
dexmachina shell                # subshell (easiest)
dexmachina env                  # or print the PATH snippet for your shell

# Get a single tool, downloaded + on PATH + verified
dexmachina get jadx

# Get a device frida-ready in one shot (runtime + push-server + verify)
dexmachina device ready

# Check what's installed vs latest (use --offline to skip network)
dexmachina status --offline

# Diagnose / auto-repair
dexmachina doctor
dexmachina fix --bootstrap

When to run up

Run dexmachina up when creating an environment, after cloning an engagement workspace, or after changing its profile, pins, or lockfile. It is safe to run again: tools and Frida environments that already satisfy the configuration are skipped. Use dexmachina update for upgrades and --force only when you intentionally want a reinstall.

For a normal daily pentest session, you usually only need:

dexmachina device ready          # reconnect/restart the matching frida-server
dexmachina console               # resume device + target workflow

Commands

Get started

Command	Description
dexmachina init [--profile NAME] [--force]	Create a repo-local workspace (config + .gitignore + .dexmachina/tools)
dexmachina up [--profile NAME] [--no-frida] [--yes]	Build the environment: install a profile + set up frida + write lockfile
dexmachina console [--device SERIAL]	Interactive pentest REPL with live device + target state
dexmachina shell	Open a subshell with every installed tool on PATH
dexmachina profile list / show <name>	List profiles or inspect a profile's tools
dexmachina lock	Write dexmachina.lock.toml from the current kit
dexmachina restore [--yes]	Install tools/frida exactly as recorded in the lockfile

Environment

Command	Description
dexmachina status [--category NAME] [--offline]	Table of installed vs latest versions
dexmachina install <tool> [--version X.Y.Z] [--force]	Install tool + dependencies
dexmachina get <tool>	Install a tool, put it on PATH, and verify it's runnable
dexmachina install --all	Install entire registry (topological order)
dexmachina update [tool] [--all] [--force]	Update to latest; pin groups update atomically
dexmachina pin <tool> <version>	Lock tool/pin-group version in config
dexmachina unpin <tool>	Remove version lock
dexmachina env [--frida-only]	Print PATH setup for all tools (or just the frida venv)
dexmachina doctor	Environment health report
dexmachina fix [--dry-run] [--yes] [--bootstrap] [--aggressive]	Diagnose and auto-repair issues

Device & arsenal

Command	Description
dexmachina device list	List connected ADB devices
dexmachina device ready [--device SERIAL]	Ensure frida runtime + push frida-server + verify with frida-ps -U
dexmachina push-server [--device SERIAL] [--no-start]	Push frida-server matching local frida
dexmachina info [tool]	Tool catalog, or detail card for one tool
dexmachina arsenal	List runnable tool CLIs (ready / missing)
dexmachina run <tool> [args…]	Run a tool via explicit dispatch
dexmachina <tool> [args…]	Direct dispatch (e.g. dexmachina frida -U)
dexmachina config	Show dexmachina.toml
dexmachina config set <key> <value>	Update a setting

Profiles

Profiles are curated tool bundles for one-command setup with dexmachina up --profile NAME:

Profile	Tools
minimal	adb, frida, frida-tools
dynamic	frida stack, objection, r2frida, medusa
static	jadx, apktool, smali/baksmali, dex2jar, enjarify
traffic	mitmproxy, apk-mitm
scanners	MobSF, nuclei, drozer
full	everything auto-installable in the registry

Repo-local layout

dexmachina init produces a portable, reproducible workspace:

my-repo/
├── dexmachina.toml         # committed — settings + profile + pins
├── dexmachina.lock.toml    # committed — exact installed versions (dexmachina lock)
├── .gitignore              # auto-updated to ignore downloaded tools
└── .dexmachina/
    ├── tools/              # downloaded CLIs/jars (gitignored)
    └── cache/              # PyPI/GitHub cache (gitignored)

Frida runtimes live in per-version venvs under ~/.dexmachina/venvs/ (machine-specific, recreated by dexmachina use / dexmachina restore), so they're never committed.

Configuration

~/.dexmachina/dexmachina.toml (or ./dexmachina.toml in the project directory):

[settings]
adb_path = "adb"
java_path = "java"
install_dir = "~/.dexmachina/tools"
auto_push_frida_server = false

[pins]
frida = "16.1.4"   # pins the entire frida group

[ignored]
tools = ["ghidra", "wireshark"]

Frida pin group (important)

The frida runtime (pip package frida) uses the version you care about for device hooks — e.g. 17.11.0.

frida-tools and objection have their own pip version numbers (e.g. 14.9.0, 1.12.5) but must be compatible with the frida runtime. DexMachina never installs frida-tools==17.11.0 — that was the old bug.

Command	What it does
dexmachina use 17.11.0	nvm-style — create/select isolated venv for this runtime
dexmachina use latest	Use latest frida release
dexmachina sync frida	Align global pip: exact frida + upgrade companions
dexmachina pin frida 17.11.0	Save preference in config (then use or sync)
dexmachina versions frida	Show active, pinned, installed, PyPI releases
dexmachina env	Print PATH snippet for active venv

After dexmachina use X, run the dexmachina env output in your shell, then dexmachina push-server.

Arsenal — run tools through DexMachina

Tools with a CLI are registered as native commands:

dexmachina arsenal              # ready vs missing
dexmachina frida --version
dexmachina frida-ps -U
dexmachina objection explore
dexmachina jadx --help
dexmachina run mitmproxy        # explicit form

Resolution order: active frida venv → ~/.dexmachina/tools/*/bin → system PATH.

Tool Categories

• Dynamic Analysis — frida, objection, r2frida, medusa
• Static Analysis — jadx, apktool, smali, radare2, ghidra (manual)
• Traffic Interception — mitmproxy, apk-mitm, Burp (manual)
• Device & ADB — adb, scrcpy, pidcat, androguard
• APK Manipulation — uber-apk-signer, apksigner (manual)
• Automated Scanners — MobSF, drozer, nuclei
• Data & Storage — binwalk, trufflehog, gitleaks
• Network — nmap, tcpdump, wireshark (manual)

Manual-only tools appear in status and doctor with install instructions but are not auto-installed.

dexmachina fix

Three-phase repair flow: diagnose → plan → apply → re-check.

Safe by default (from doctor findings):

Issue	Action
Frida pin group mismatch	Sync or reinstall entire group (respects [pins])
frida-server not running	push-server
adb missing	Install platform-tools
Broken pip/github install	Reinstall tool

Flags:

Flag	Effect
--dry-run	Show plan only
--yes / -y	Skip confirmation
--bootstrap	Install missing core env tools: adb, frida, jadx, apktool, mitmproxy, …
--aggressive	Also update all outdated registry tools
--only frida,server	Limit to categories

Manual steps (Python upgrade, Java, Node.js, no device connected) are listed but never auto-applied.

Impact levels

Each fix shows an Impact column — how disruptive the change is, not a security rating.

Level	Meaning
Low impact	One tool or device; easily reversed (install adb, push frida-server, reinstall one tool)
Medium impact	Multiple packages or version bumps together (sync frida pin group, --aggressive updates)
High impact / manual	DexMachina cannot apply automatically (upgrade Python, install JDK, plug in a device)

Every dexmachina fix run prints a legend panel explaining these before the plan. Use --no-legend to hide it.

Roadmap: Pentest Environment

DexMachina aims to become a batteries-included Android pentest environment:

• Today — tool registry, version pinning, doctor, fix, frida-server push
• Next — environment profiles (minimal, dynamic, full), PATH/setup shell hook, workspace templates
• Future — bundled configs (mitmproxy certs, MobSF docker compose), project scaffolds, CI-ready headless mode

GitHub API

Release versions are fetched from the GitHub Releases API with a 1-hour cache in ~/.dexmachina/cache/. Set GITHUB_TOKEN to avoid rate limits.

Development

pip install -e ".[dev]"
pytest

License

MIT