More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Binary Ninja Reverse Engineering

    1

    Reverse a binary with Binary Ninja's IL ladder: triage in ten minutes, read HLIL for logic, trace data in MLIL's SSA form, and drop to LLIL only when the lifter lies — with evidence on every finding.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Binary Ninja Reverse Engineering

    Binary Ninja Reverse Engineering

    Example session with this skill installed

    A 2 MB utility binary seems to load a config file and then contact a server. I need to know what it sends and under what conditions, with evidence I can check.

    • Read your context and instructions
    • Compiled the binary ninja reverse

    TRIAGE (10 minutes): Hash and architecture recorded. Imports show file I/O, a crypto library, and network calls — the resume fits the suspicion. No exports (executable). Strings show a config path, a server hostname, and a missing-key error message. Segments normal, nothing writable-and-executable.

    READING: Entry-point HLIL shows parse config, load a key, build a request, send it. The config parser is renamed parse_config_header. One variable's origin was unclear; MLIL's SSA chain traces it to a file read of the config path from strings. A condition HLIL renders oddly turns out, in LLIL, to be a length check the lifter inverted — not a branch on the key's presence.

    TYPES: Config struct recovered from field accesses — magic, version, key_len, key. The request function's prototype typed at its call sites fixes a misread parameter that had made the payload length look wrong.

    FINDING (high confidence): The binary reads a 32-byte key from the config file and sends it as the first 32 bytes of the request body, only when the config parses and the key length is 32. Evidence: address of the file read, MLIL trace, config-path string, LLIL-confirmed length check, two call sites. A second analyst re-ran the trace from the finding note alone and confirmed it in under an hour.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Binary Ninja's real advantage is its ladder of intermediate languages, and the skill is knowing which rung a question needs. This skill runs the loop: a ten-minute triage (hash, imports, exports, strings, segments) that frames the whole case, HLIL-first reading with one-sentence summaries and renaming as you understand, MLIL's SSA form for "where does this value come from", and LLIL or disassembly for single blocks the lifter misread.

    Types are treated as documentation — struct layouts from field accesses, enums from small-integer switches, prototypes typed at call sites, since a wrong prototype is the top source of misread HLIL. Every finding that matters carries its address, the IL level it was read at, its evidence, and a confidence grade (high needs disassembly confirmation plus corroboration). Patches are mutations logged on a copy, never on the original. Automation covers bookkeeping — renaming by pattern, dumping cross-references — while judgment stays human. A worked example traces a config key from file read to network send and confirms it in under an hour from the finding note alone.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.

    • Static reading frames hypotheses about runtime behavior; it cannot prove what happens when the binary runs.
    • Packed or obfuscated code must be unpacked first — the IL ladder reads the packer, not the program.
    • The lifter can misread optimized code; suspicious blocks need an LLIL or disassembly check.
    • Huge binaries are triaged and read by call graph, not linearly — full coverage is not the goal.
    • Anti-analysis checks can change observed behavior; a single-run observation is not a verdict.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Disassembly is ground truth but slow. HLIL reads like source and answers most questions; MLIL answers data-flow questions through SSA; LLIL and disassembly are for the single block where the lifter looks wrong. Starting at the bottom rung costs you most of your day.

    Confirmation in disassembly or LLIL plus at least one corroborating call site or string. HLIL-only reads with supporting evidence are medium. A single uncorroborated observation is a lead, not a finding.

    The method transfers — any tool with layered views supports the same ladder logic — but the step-by-step instructions name Binary Ninja's views and Python API, so expect to translate view names and script calls.

    New in Security