decision-recommenderlisted

# Decision Recommender An L3 Truth-Evolution skill. Where `learning-loop-engine` surfaces patterns across time, this skill turns a specific question — "should we ship this? should we add this gate? should we accept this risk?" — into a document the team can read, argue with, and decide on. The output is NEVER a single answer. It's **always** a set of options with their real trade-offs, including "do nothing" as option zero, and a recommendation with an explicit confidence level. The failure mode this skill is designed against: AI-assisted decision tools that confidently produce one "correct" answer and make the human feel bad for questioning it. That's not a decision, that's an opinion delivered with extra steps. Real decisions involve trade-offs, and a recommendation without trade-offs is either obvious (in which case the human didn't need help) or wrong (in which case the tool made it harder). ## When You're Invoked - **PIPELINE-4 step 2 (conditional)** — only when a decision is explicitly requested by a human or another skill. Not auto-invoked on every run. This is slow-loop, high-context, human-in-the-loop territory. - **On demand** as `/vibeflow:decision-recommender <findings-path> [--type <t>]`. - **From `learning-loop-engine`** when a pattern has escalated to `urgent` and the team needs an options-based framing, not just a one-liner recommendation. - **From `release-decision-engine`** when the release verdict is `CONDITIONAL` and the team needs to dec