Bob Review Coach

Code review is the slowest human bottleneck in most development workflows. A reviewer opening a pull request has to hold several unrelated concerns in their head at once: is this a security risk, is the logic correct, will this be maintainable later. In practice one concern crowds out the others, and real defects slip through. Bob Review Coach takes a git diff, either pasted directly or fetched from a public GitHub pull request, and runs it through three specialist reviewers concurrently. Each returns findings tagged by severity (blocking, suggestion, nitpick) with a plain-English explanation of why it matters and a concrete suggested fix. Reviewing a real two-file pull request from the stripe-node repository, it surfaced ten findings in six seconds against an estimated 28 minutes of manual review. Among them: a lexicographic string comparison of Node version numbers that would misclassify Node 10 as newer than Node 19, exactly the class of subtle bug a human reviewer skims past. Every line was written through IBM Bob IDE in Agent mode. Bob scaffolded the project, recovered from its own scaffolding errors without instruction, restructured the inference layer when the architecture changed, and implemented the parallel specialist design, rate-limit recovery, and diff truncation logic. Session summaries for every task are in the bob_sessions folder.