More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Protocol Reverse Engineering: Binary Formats

    1

    Recover an undocumented protocol from authorized captures: framing, fields, encoding, and state machine, delivered as a message dictionary plus a decoder that runs.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Protocol Reverse Engineering: Binary Formats

    Protocol Reverse Engineering: Binary Formats

    Example session with this skill installed

    We have a smart-home hub on our lab network speaking a proprietary TCP protocol — no docs. We own the device and can capture traffic and trigger actions from its app. Map the protocol for us.

    • Read your context and instructions
    • Compiled the protocol reverse engineering

    PROTOCOL MAP (excerpt) — produced by applying the skill. Scope confirmed first: your device, your lab network, interoperability purpose.

    Framing: every message opens with magic bytes 0xAA 0x55, followed by a 2-byte length (big-endian, header excluded — the interpretation was verified by predicting every message boundary across the full capture, not just the fitting sample), a 1-byte type, the payload, and a 2-byte CRC16 whose variant was identified by recomputation across 40+ messages.

    State machine: connect, then type 0x02 authentication — the server's 0x02 reply carries a challenge and the client answers with a derived response — then ready state. Heartbeats (type 0x01) every 30 seconds. Commands start at 0x10; responses echo the command type with bit 7 set.

    Fields, with confidence labels: the relay-toggle command (0x10) carries a 4-byte device ID. We triggered the relay from the app on cue three times; exactly those 4 bytes changed each time — labeled Confirmed. Two payload bytes in the sensor-report message (0x21) vary with readings but their units are unresolved — labeled Guess, kept out of the decoder's assertions, and listed under open questions in the dictionary.

    Deliverables: a message dictionary covering 8 message types in the standard template, a working decoder (about 60 lines of Python using struct) that parses a fresh capture into dictionary terms, and an evidence pack of hex excerpts paired with decoded forms, sanitized of the session token present in the capture.

    Finding flagged separately: the 0x02 challenge-response reduces to a fixed XOR — effectively no authentication strength. The map documents it; the report recommends disclosure to the vendor.

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

    About this skill

    A proprietary client is talking to a proprietary server, there is no documentation, and you need to understand the conversation — for interoperability, for debugging your own systems, or under a written testing engagement. This skill is the method for recovering the protocol from the traffic itself. It opens with an authorization and scope gate — whose systems, what permission, what the results may be used for — and treats captures as sensitive material throughout, since they routinely contain credentials and tokens. The analysis hunts five answers: framing, fields, encoding, state machine, and protection. Triage separates directions, spots magic bytes and length fields within a dozen messages, reads timing for request/response pairing, and uses controlled client actions, triggered on cue and timestamped, as the fastest correlation signal available. Frame recovery proceeds by alignment: what is invariant across messages is structure, what varies is data; length interpretations must predict every boundary in the capture, checksums are identified by recomputation rather than assumption, and the state machine is drawn as the work proceeds — a protocol you cannot draw, you do not understand yet. Serialization coverage includes schema-free protobuf decoding, gRPC framing, and standardized WebSocket/MQTT payloads; the encryption section starts from key provenance in the client rather than the cipher. Every claimed field carries a confidence label — Guess, Supported, or Confirmed — and the decoder may only rely on the top two. Deliverables are a message dictionary in the supplied template, a working decoder, and a sanitized evidence pack, gated by a simple test: someone else runs the decoder on their own fresh capture and gets sensible output.

    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.

    • Strong, correctly implemented encryption with keys that never touch the client will stop payload analysis; the skill still recovers framing, timing, and state from headers, and says plainly where the wall is.
    • It cannot create authorization: if ownership or scope for a capture is unclear, the correct output is a question, not an analysis.
    • Protocols under active development drift between client versions — the dictionary is versioned to the captures it came from and needs re-verification when the client updates.

    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 24 days ago

    Permissions

    No special permissions declared or detected

    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    Enough to start, with honest limits. Framing, length fields, checksums, and the state machine usually fall out of the capture alone. What you lose is the Confirmed confidence level: without controlled actions, fields stay at Supported, and the dictionary labels them that way. Key provenance for encrypted payloads is also much harder without the client — the skill names that as an open question rather than guessing.

    By the standard the skill enforces: a length interpretation must predict every message boundary in the whole capture, a checksum must recompute correctly on every sample, and a payload field earns Confirmed only when a deliberately triggered action moves it as predicted. Anything weaker is labeled a guess in the dictionary and kept out of the decoder's assertions.

    No. The skill opens with an authorization gate — your own systems, CTF targets, lawful interoperability, or engagements with written scope — and it stops when scope is unclear. Its deliverable is documentation: a message dictionary, a decoder, and a sanitized evidence pack. Findings like weak authentication are documented for the owner or vendor, which is where the worked example ends.

    New in Security