Printing Press Output Review

Internal review layer that catches plausibility bugs rule-based checks miss. Runs after a CLI prints, scanning sampled command output for substring-match false positives, silently dropped sources in aggregations, broken ranking weights, and format issues like malformed URLs or mojibake. You don't invoke this directly,it's called by the main printing-press and polish skills during their diagnostic loops, runs in a forked context to keep agent chatter isolated, and surfaces everything as warnings in Wave B (never blocks shipping). The agent gets live-check JSON plus research context and judges what a human would notice in five minutes of hands-on testing. Think of it as automated dogfooding for the stuff grep can't encode.