amruth-sn/kong — reverse-engineered prompt

Build me a tool called Kong that acts like an AI reverse engineering assistant for binaries. I want to point it at a compiled binary, especially stripped or somewhat obfuscated ones, and have it run a mostly hands off pipeline that figures out what functions do, renames ugly autogenerated function names into human readable ones, recovers likely types, parameters, structs, and calling conventions, and writes the improved results back into Ghidra as well as exporting a JSON report.

It should use Ghidra directly so the analysis has real program context like call graphs, cross references, strings, and decompiled code, instead of just dumping raw text into an LLM. Have it analyze functions in a smart order so lower level helpers get named before callers. Include signature matching for known library or crypto functions, a cleanup and synthesis pass to make names consistent, some deobfuscation help, support for OpenAI and Anthropic, and basic token cost tracking. A simple setup flow and one command to run everything would be great. Look up current docs online if you need to.