Practitioner details an LLM-assisted reverse engineering workflow
Original title: LLM assisted reverse engineering
DeepSeek v4.1 Flash drives Ghidra and radare2 through agent skills, with notes and hypothesis logs when analysis stalls.
A reverse engineer describes replacing manual analysis with agent skills around DeepSeek v4.1 Flash, Ghidra via an MCP bridge, and faster CLI tools radare2 and kuna. The agent records findings in Markdown, then debugs, logs, or follows strings when disassembly stalls. Python with UV is used for probes and Rust for implemented changes. The author says recent successes would have taken years by hand.
Unconfirmed: Claims of multi-year reverse-engineering successes in six months are the author's and are not independently verified.
- Primary model is DeepSeek v4.1 Flash; Codex with Luna or Terra is occasional.
- A public Ghidra skill includes 22 scripts for targeted decompilation.
- Hypothesis debugging logs one testable claim at a time in a CSV.
- The post says agents still need human steering when they narrow too early.
- AI models
- DeepSeek v4.1 Flash
From the original article
6 min read at slowchronicle.com
I have been hacking both hardware and software for 2 decades. Previously manually. But in 2026 with the help of AIs the proccess is greatly sped up.
By ioma8 · slowchronicle.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.