learning-loop-enginelisted

# Learning Loop Engine An L3 Truth-Evolution skill. Where every L2 skill looks at a SINGLE run and says "here's what's wrong right now", this skill looks across TIME and says "here's the pattern, here's what it's been costing you, here's what to change". The output isn't a gate verdict; it's an improvement plan. L3 skills are the slow loop. They run once per sprint (or on demand when a production incident needs a post-mortem), they take longer, and they produce recommendations the team evaluates rather than gate-blocks on. A team that runs learning-loop-engine weekly tends to catch structural quality drift months earlier than a team that only runs per-release gates — that's the bet L3 is making. ## When You're Invoked - **PIPELINE-6 step 1** — first step of the release pipeline, before the release gate runs. Surfaces recurring patterns the release decision should know about (not block on — L3 recommendations are informational to the gate). - **PIPELINE-7 step 1** — first step of the retrospective pipeline. Runs on demand at sprint boundaries. - **On demand** as `/vibeflow:learning-loop-engine [--mode <m>] [--since <sha>]`. - **From `release-decision-engine`** when a post-release incident needs the "what did we miss" trace. ## Input Contract The skill runs in one of four distinct modes. Each mode has its own input contract. ### Mode 1: `test-history` Default mode. Analyzes the project's accumulated test reports. | Input | Required | Notes | |-------|-------