STScott Thurmaninscottthurman.hashnode.dev·May 12 · 4 min readFour .cursorrules files that turn Cursor chat into a focused code reviewerCursor's chat is great at writing code. It is mediocre at reviewing it. The default behavior is conversational - "looks good, here are some suggestions" - which is the wrong mode for a review. A revie00
STScott Thurmaninscottthurman.hashnode.dev·May 11 · 4 min readprpack v0.2.0: autonomous review mode landsThe first version of prpack had one job: pack the right review context and get out of the way. That was deliberate. The failure mode I kept seeing was not "the model is bad at code review." It was "th00
STScott Thurmaninscottthurman.hashnode.dev·May 11 · 9 min readAI PR review tools in 2026: where prpack fitsAI PR review is not one category anymore. In 2026, the tools split into three buckets. The first bucket is the GitHub or GitLab reviewer bot. It comments on a pull request, tries to catch bugs, and ma00
STScott Thurmaninscottthurman.hashnode.dev·May 11 · 5 min readAdd an LLM PR reviewer to your GitHub Actions in 5 minutes (no API key)Most "AI code review" GitHub Actions need either a paid platform account, an OpenAI/Anthropic API key sitting in your repo secrets, or both. Both are friction. The first costs ongoing money; the secon00
STScott Thurmaninscottthurman.hashnode.dev·May 11 · 5 min readA reproducible test: Claude reviews the same PR twice, with and without contextMost of what I read about LLM code review is "we tried it, it works." That's not enough to be useful. I want to know: same model, same PR, two different review contexts - does the output actually chan00
