AI code review catches structural issues — null derefs, leaks, races — about 65% of real defects. The other 35% are intent violations: bugs only catchable when you know what the code is supposed to do. Quality Playbook derives behavioral requirements from your codebase AND your docs (specs, issues, chat history), then drives a six-phase review against them. Finds the bugs that look right but aren't.
1 skill
Orchestrate Quality Playbook runs across multiple repos from one Claude Code session. The harness skill drives a disk-backed state machine via in-session ScheduleWakeup polling: each tick dispatches QPB worker subagents (Task tool), tails their heartbeats, advances the state machine, prints a status table, and reschedules — until every run reaches a terminal state or you write a STOP file. No external scheduler, no long-running daemon; the same ScheduleWakeup primitive the v1.5.7 watcher has run reliably for weeks.
1 skill