Dev uses Opus 5.5 with Lean to formally verify Claude Agent SDK, shipping 16 bug-fix PRs
jimmykoppel · x · 2026-09-23
bcherny used Opus 5.5 with Lean to formally verify the Claude Agent SDK: a few short prompts yielded 16 PRs fixing bugs and race conditions. He also combines Lean and TLA+ to probe data flow, concurrency and state management, noting Claude excels at both even for non-experts.
Mike Knoop adds that while formal verification is becoming practical for security, it doesn't automatically build human understanding—which he sees as the bigger alignment problem.
More from coding & agent
- Anthropic's Opus 5.5 guide: hand over whole tasks and stop micromanaging — xiaohu · 2026-09-23
- Claude Opus 5.5 lands: beats Fable 5.1 on coding benchmarks, cuts API prices 60% — xiaohu · 2026-09-23
- Opus 5.5's three changes: run long tasks autonomously, skip 'think harder' — xiaohu · 2026-09-23
- Yoav Goldberg: some tasks just need deterministic rules — agents can write them — yoavgo · 2026-09-23
- Devs migrate from GPT-6 Astra to Opus 5.5 as coding model race churns — rudrank · 2026-09-23
- Goldberg Follows Up: Design Bespoke Extractors for Variables in Tailored Predictors — yoavgo · 2026-09-23