Navigate any binary.
AI-assisted reverse-engineering pipeline. Drop a binary, get a structured triage report with MITRE ATT&CK mapping, IoCs, and a SOC-actionable verdict.
Built on:
- Ghidra — NSA's reverse-engineering suite (analysis engine)
- libghidra — typed Python SDK for headless Ghidra
- ghidrasql — SQL over a Ghidra program database
- ghidrasql-skills — Claude Code agent skills
- DearPyGui — GUI
- reveal.js — demo deck output
GUI (DearPyGui)
|
v
orchestrator.py ----> Anthropic Claude (Agent SDK)
| |
| | tool calls (ghidrasql HTTP)
v v
headless Ghidra <--- LibGhidraHost HTTP :18080
(analyzeHeadless + LibGhidraHeadlessServer.java)
Phases (each = focused prompt, JSON output):
- Triage — funcs/imports/strings overview, pack detection, suspicion score
- Capability — MITRE ATT&CK technique mapping with evidence
- IoC — URLs, IPs, mutex, regkey, paths (xref-validated)
- Verdict — clean / suspicious / vulnerable (BYOVD) / malicious / unknown + SOC action recommendation
Final output: report.json (schema in schema/report.schema.json) plus an HTML reveal.js deck for demos.
Prereqs:
- Windows 10/11 or Linux
- Python 3.12+
- Ghidra 12.1+ (https://github.com/NationalSecurityAgency/ghidra/releases)
- JDK 21 (for Ghidra)
- LibGhidraHost extension built and installed in Ghidra
- An Anthropic API key (or Claude Code with the ghidrasql-skills plugin)
git clone <this-repo> theseus
cd theseus
python -m venv .venv && source .venv/bin/activate # or .venv\Scripts\activate on Windows
pip install -r requirements.txt
cp .env.example .env # fill in ANTHROPIC_API_KEY, GHIDRA_HOMECLI:
theseus run path/to/binary.exe # single binary, prints + writes report.json
theseus run binary.exe --deck report.html # also generate reveal.js demo deck
theseus serve --port 8090 # HTTP API for SOC ingestGUI:
theseus guiDrop a binary on the window. Watch the pipeline tabs fill. Open the deck when done.
# Push verdict to a SOAR webhook
theseus run binary.exe --webhook https://soar.internal/ingestJSON output is schema-stable. See schema/report.schema.json.
- crackmes.one — legal RE practice binaries
- https://github.com/lallouslab/vulnerable-drivers-ex