More screenshots
Works with the AI Agents you already use
Ghidra Reverse Engineering: Headless Analysis
Turn an unknown binary into a written behavior profile with Ghidra's free decompiler: five-minute triage, string-and-import hunting, function reconstruction, and headless batch runs.
$5
Ghidra Reverse Engineering: Headless Analysis
Example session with this skill installed
Our file server flagged an unknown executable dropped in a shared folder. We pulled it (never ran it), plus 30 similar files from the same week. Give us a behavior profile and tell us which of the 31 deserve a deep dive.
- Read your context and instructions
- Compiled the ghidra reverse engineering
SAMPLE — invoice_032.exe, 1.2 MB, SHA-256 recorded in the evidence log and cited on every claim below. Context: dropped in a finance share; not executed on any host.
TRIAGE (5 MIN): High-entropy .text section. Strings show a browser profile path ending in Login Data, HTTPS endpoints, a database-open import, and one base64 blob. First impression, written before import: credential stealer with an encrypted config.
HEADLESS BATCH (31 SAMPLES): analyzeHeadless with a fixed analyzer set; the post-script emitted one JSON record per sample (hash, imports, flagged strings, entry function). 27 records cluster as one known updater family — imports and strings match within noise, closed as benign with the diff attached. 4 outliers share the browser-path string. Those get GUI time; the 27 do not.
GUI FINDINGS — invoice_032.exe:
- walk_browser_profiles @ 0x140002c80 — CONFIRMED: enumerates browser profile directories; cross-referenced from the Login Data string.
- exfil_post @ 0x140003f10 — CONFIRMED: calls the OS data-protection decryption API, then posts the result. Credential theft, decrypted with the victim's own OS keys.
- decrypt_config @ 0x140004a00 — CONFIRMED: constant tables match the ChaCha20 sigma; the base64 blob is a server config decrypted at runtime.
- Keylogging — SUSPECTED ONLY: a window-hook import exists, but no function wires it to the exfil path. Named as unconfirmed, flagged for sandbox follow-up.
VERDICT: Credential stealer with encrypted configuration. The 3 remaining outliers share the stealer markers and go to full analysis next. Nothing was executed at any point.
Connects securely to your tools. The creator never sees your data.
About this skill
A field routine for investigating binaries you are authorized to analyze, using Ghidra's decompiler as the primary lens. It begins with an authorization and lab-safety gate, then a five-minute no-tools triage (hashes, strings, section entropy, a written first impression). The interactive routine imports with intent, hunts behavior backwards from the strings and imports a binary cannot hide, and reconstructs one function at a time, renaming and commenting so the decompiler output becomes readable evidence. The headless half of the skill turns one-off analysis into a pipeline: analyzeHeadless with a fixed analyzer set, a post-script contract that emits one JSON record per sample, per-sample failure logging, and run-to-run diffing so a queue of fifty samples resolves into the four that deserve your afternoon. A mode decision table tells you when to stay in the GUI, when to go headless, and when to stop static work and hand off to dynamic analysis. A confidence rule separates confirmed behaviors (decompiled logic plus an independent signal) from suspected ones, and a decompiler-honesty section names exactly where Ghidra's output lies: guessed struct layouts, callback signatures, jump tables, and inlined crypto. Includes a behavior profile template and a complete worked example that profiles a credential stealer from four functions.
- It directs analysis and structures the writeup; you run Ghidra in your own environment and bring back the output.
- Packed samples must be unpacked first (or analyzed as an unpacked memory image); the routine says so rather than pretending the stub is the program.
- Dynamic behavior — live network traffic, injection, anti-debug — is handed off to a sandbox or debugger; static findings are labeled accordingly.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
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