Guide · Agent skill
How to use the Binary Ninja reverse engineering agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The binary-ninja-reverse skill is a set of instructions that teaches an AI agent how to analyze compiled programs using Binary Ninja, a commercial reverse engineering tool. It is written for authorized analysis of software you are permitted to examine. After reading this guide you will know how to install the skill, ask for analysis in plain language, and set boundaries on what the agent may change.
Binary Ninja Reverse Engineeringzhaoxuya520/reverse-skill · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the ZIP from this page and unzip it so you have the binary-ninja-reverse folder with SKILL.md and agents/openai.yaml.
- 2Place it or upload it. Copy the folder to ~/.claude/skills/ for personal use or to a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Install Binary Ninja. Install Binary Ninja manually with a valid Vector 35 license, since it is commercial software and is not bundled with the skill.
- 4Confirm scope. State that the target is authorized and in scope before asking the agent to act on it.
- 5Ask for analysis. Make a plain request that names Binary Ninja, such as listing imports or decompiling a function in HLIL.
- 6Review the output. Check the reported addresses, IL level, strings and imports, and reproduction steps, and cross-check important conclusions.
What the skill makes the agent do
An agent skill is a folder with a SKILL.md file that tells the agent when and how to perform a task. This skill covers analysis in Binary Ninja, including reading HLIL, MLIL and LLIL representations, listing strings, imports and exports, following cross-references, working with types, reviewing patches, and automating through the Python API.
The skill asks the agent to work in a fixed order. It confirms the repository case scope is ready, records the architecture, entry points, imports, exports and file hash, then reads decompiled logic. Findings are reported with addresses, function names, the IL level used, supporting strings or imports, confidence, and reproduction steps.
Install the binary-ninja-reverse skill
Download the ZIP for this skill from this page and unzip it. The skill folder contains agents/openai.yaml and SKILL.md. For a personal setup, place the unzipped skill folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.
In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.
Binary Ninja itself is not included. It must be installed manually with a valid Vector 35 license.
Set up Binary Ninja before you start
The skill assumes Binary Ninja is installed and that you hold a valid license. Install the Binary Ninja side through its Plugin Manager or from a reviewed source. If Binary Ninja cannot open the target, the skill falls back to other tools such as ghidra-reverse, ida-reverse or radare2.
- Required: Binary Ninja installed manually with a valid Vector 35 license.
- Optional: Node.js with npx, to run the pinned binary-ninja-mcp@1.0.0 bridge.
- Optional: the fosdickio/binary_ninja_mcp Binary Ninja plugin, for MCP use.
- Optional: an MCP-capable client that you have configured yourself.
Trigger the skill with natural requests
You do not need special syntax. Name Binary Ninja when you want it, or describe the analysis you need. The skill is meant for requests where Binary Ninja's intermediate languages help with data-flow, or where you want a cross-check against IDA, Ghidra or radare2.
State that the target is authorized and in scope. The skill assumes a confirmed target and does not verify permission on its own.
- Use Binary Ninja to list the imports and exports of this binary and summarize the entry point.
- Decompile the function at this address in HLIL and explain what it does, citing the strings it uses.
- Trace the callers of this function and show me the cross-references to the config string.
- Cross-check my IDA findings on this function with Binary Ninja's MLIL SSA view.
Use the optional MCP bridge safely
The MCP bridge is optional and only used when you explicitly request it. The skill names the community project fosdickio/binary_ninja_mcp, which is GPL-3.0 and is not an official Vector 35 component. Its plugin metadata version is 1.1.0 and it needs Binary Ninja build 4000 or later.
For the stdio bridge, the skill pins the published package version rather than using an unbounded one. Register the command only in the MCP client you chose. The bridge is not ready until Binary Ninja is running, a binary is open, and the localhost plugin endpoint responds. The Binary Ninja HTTP listener should stay on 127.0.0.1:9009, and network exposure is off by default.
npx -y binary-ninja-mcp@1.0.0 --host 127.0.0.1 --port 9009Safety notes: scripts, network and changes
This skill runs scripts and uses the network, so review what you enable. The bridge runs locally, but it is a third-party plugin and should not be exposed on a shared network.
The skill treats patches, prototype changes and saved-file writes as mutations. The agent performs them only when you ask, and it is expected to work on a copy and keep the original artifact. High-impact conclusions should be checked against a second evidence source or another disassembler.
- Work on a copy of the binary before patching or saving database changes.
- Do not enable the MCP bridge on a shared network.
- Confirm the target is authorized before analysis begins.
- Expect the agent to discover the live tool list rather than assume every upstream function exists.
What people use it for
- Listing imports, exports and representative strings to get an overview of an unfamiliar binary.
- Reading decompiled HLIL for a specific function and tracing its callers and callees.
- Using MLIL SSA to follow data flow when HLIL is hard to follow.
- Cross-checking IDA, Ghidra or radare2 findings with an independent Binary Ninja view.
- Teaching students how to read compiled code with a structured, evidence-based workflow.
Frequently asked questions
Is the Binary Ninja reverse engineering skill free?
The skill itself is MIT licensed. Binary Ninja is commercial software that must be installed manually with a valid Vector 35 license, so you need to obtain that license separately. The optional MCP bridge plugin is a GPL-3.0 community project.
What do I need to run this skill?
You need Binary Ninja installed with a valid Vector 35 license. Node.js with npx is optional and only needed for the pinned MCP bridge. An MCP-capable client is also optional and must be configured by you.
Where do I put the skill files?
Unzip the downloaded ZIP into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can upload it under Settings → Capabilities → Skills.
Does the skill work without Binary Ninja?
Not for Binary Ninja analysis. The skill says to fall back to ghidra-reverse, ida-reverse or radare2 when Binary Ninja is unavailable or cannot open the target.
Why won't the MCP bridge connect?
The bridge is not ready until Binary Ninja is running, a binary is open, and the localhost plugin endpoint responds. Check that the HTTP listener is on 127.0.0.1:9009 and that the command is registered in your chosen MCP client.
Is it safe to let the agent patch a binary?
The skill asks the agent to treat patches, prototype changes and saved-file writes as changes that happen only when requested, and to work on a copy while keeping the original. Review every change yourself.
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 security researchers, students and analysts who already have a Binary Ninja license and are analyzing software they are authorized to examine. It is not a fit for anyone without a license or permission, or for anyone who wants to expose the MCP bridge on a shared network; in those cases choose a different tool such as Ghidra or radare2 with its own skill.
Download the skill