Add SmartBench - Evidence-constrained code diagnosis

[xianyu-sheng]

SmartBench is a code diagnosis workbench that combines deterministic static analysis with evidence-constrained LLM review.

Key features:

• Language-neutral semantic analyzers (CFG/ICFG, resource lifecycle, state machines)
• Evidence-gated architecture: static analyzers own facts, LLM proposes hypotheses, validators decide
• Real upstream bugs found and verified: stunner #89, qscan #22, sniproxy #203, Robyn #1432
• Reproducible benchmark corpus with 20 before/after cases
• Python 3.10+, MIT license

Why add it:

• Proven results: found real bugs in production open-source projects
• Unique architecture: evidence-constrained to reduce false positives
• Complete verification chain: static audit → runtime reproduction → local fix → upstream submission

GitHub: https://github.com/xianyu-sheng/SmartBench

[mre]

Thanks for submitting SmartBench. It is under six months old and currently has 1 star and one contributor. Please resubmit once eligible and add the uses-llm tag.