Guide · Agent skill
How to use the Android APK Malware Static Analysis agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains what the analyzing-android-malware-with-apktool agent skill does, how to install it for Claude, and how to ask an agent to statically triage a suspicious Android APK. It is written for analysts and students who have not used agent skills before. By the end you will know how to install the skill, what to prepare before running it, and how to read the report it produces.
Android APK Malware Static Analysismukul975/Anthropic-Cybersecurity-Skills · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the skill ZIP from this page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ for a project.
- 3Or upload in Claude apps. Alternatively, go to Settings → Capabilities → Skills in the Claude apps and upload the ZIP.
- 4Install the requirements. Install Python 3.9+, androguard and apktool, and optionally jadx, inside an isolated analysis environment.
- 5Ask for a triage. Describe the task and name the APK file, for example asking for a full static triage of sample.apk.
- 6Review the report. Read the JSON-style risk report and treat its findings as indicators that need expert review.
What is an agent skill and what does this one do?
An agent skill is a folder of instructions and scripts that an AI agent loads when a task matches the skill's description. This skill covers static analysis of Android APK files that may be malware. Static analysis means the sample is inspected without being run.
The agent uses androguard to read the app's manifest, permissions, activities, services, broadcast receivers and providers. It scans for suspicious API calls such as reflection, crypto, SMS and telephony, and detects dynamic code loading patterns such as DexClassLoader and Runtime.exec. It also extracts hardcoded URLs, IP addresses and other indicators from strings.
Install the skill in Claude
Download the ZIP of the skill from this page. Then choose one of the following options.
For a personal setup, unzip the file into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Project: unzip into .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Install the requirements before you run anything
The skill needs Python 3.9 or newer, the androguard Python library and apktool for resource decompilation. jadx is optional and is used for Java source recovery. Run the analysis inside an isolated environment such as a virtual machine or sandbox, and only use sample APK files you are authorised to examine.
- Python 3.9+ with androguard
- apktool (resource decompilation)
- jadx (optional, Java source recovery)
- An isolated VM or sandbox
- Sample APK files to analyse
How do you trigger the skill with a request?
You do not need to type the skill name. Describe the task in plain language and mention the APK file. The agent matches your request to the skill's description, which covers statically triaging suspicious APKs and building mobile malware detection rules.
Requests that work well name the file and the kind of answer you want, such as a permission review, a component list or a full report.
- Triage this suspicious APK statically and tell me the risk level: sample.apk
- List the permissions and components declared in sample.apk
- Check sample.apk for dynamic code loading and reflection calls
- Find hardcoded URLs and IP addresses in sample.apk
What the script can do and how to run it
The skill's script agent.py accepts one APK and one mode. The modes are permissions, manifest, apis, strings and full. If you omit the mode, full analysis runs by default.
The permission check flags 20 dangerous permissions, including SEND_SMS, READ_CONTACTS, BIND_DEVICE_ADMIN and BIND_ACCESSIBILITY_SERVICE. The risk level is CRITICAL at 8 or more dangerous permissions, HIGH at 5 or more, MEDIUM at 2 or more, and LOW below 2.
- permissions: permission risk assessment
- manifest: activities, services, receivers and providers
- apis: suspicious API calls
- strings: URLs, IPs and indicators
- full: complete report (default)
python agent.py sample.apk permissions
python agent.py sample.apk manifest
python agent.py sample.apk apis
python agent.py sample.apk strings
python agent.py sample.apk full
python agent.py sample.apkRead the report with care
The output is a structured JSON-style report with permission analysis, component listing, suspicious API calls, network indicators and a risk score. The report also maps findings to MITRE ATT&CK mobile techniques.
Treat the risk score as a heuristic. A high score is a reason to investigate further, not proof of malicious intent. Static analysis cannot show what the sample does at runtime, so confirmation needs expert review and, where appropriate, other methods.
- Risk levels come from counts of dangerous permissions
- Findings are indicators to review, not forensic evidence
- The skill does not make network requests
Safety notes for scripts and network use
The skill runs scripts, so review agent.py and the requirements before you run it on a sample. The skill does not use the network, according to its metadata. Still, keep analysis in an isolated environment so that a malicious sample cannot reach your main machine or network.
Only analyse apps you have authority to examine. This is a reverse-engineering workflow and should not be used on live production apps you do not own or have permission to test.
- Runs scripts: yes
- Uses network: no
- Work in a VM or sandbox
- Do not analyse apps without authority
What people use it for
- A SOC analyst triages an APK attached to a phishing report before deciding whether to escalate.
- A security engineer reviews which dangerous permissions a malware family requests to write detection rules.
- A student practises identifying obfuscation, dynamic code loading and reflection in a sample APK.
- An analyst extracts hardcoded URLs and IP addresses from a sample to build threat hunting queries.
Frequently asked questions
How do I install the androguard agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What do I need to run this APK analysis skill?
You need Python 3.9 or newer, the androguard Python library and apktool for resource decompilation. jadx is optional and is used for Java source recovery. The source recommends an isolated environment such as a VM or sandbox.
Is this skill free to use?
The skill is published under the Apache-2.0 licence, as listed on this page. Check the licence file in the skill for the full terms. The source does not mention pricing.
Does the skill run the malware sample?
No. The skill performs static analysis, which inspects the APK without executing it. It cannot prove what a sample does at runtime.
How is the risk level calculated?
The permission check counts dangerous permissions from a list of 20. CRITICAL is 8 or more, HIGH is 5 or more, MEDIUM is 2 or more, and LOW is below 2. Treat the result as a heuristic that needs expert review.
What are the alternatives to this skill for APK analysis?
The source does not compare this skill with other tools. Analysts often use apktool and jadx directly for manual decompilation. This skill is most useful when you want an agent to run the androguard checks and produce a structured report.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits SOC analysts triaging suspicious APKs, security engineers writing Android malware detection rules, and students learning static analysis. It is not suitable for dynamic analysis, for confirming runtime behaviour, or for forensic evidence without expert review. Use it inside an isolated environment and only on samples you are authorised to examine.
Download the skill