What problem does it solve? Building new memory features for the hippo system without a disciplined evaluation loop leads to regressions and wasted hours on expensive LoCoMo benchmark runs. This Skill enforces a strict eval pyramid so every feature is proven on fast micro fixtures before touching slower benchmarks. ## Core Features & Use Cases - Tiered eval pyramid: Runs a ~30s micro benchmark on every change, a 5-10 minute stratified LoCoMo smoke test before PRs, and reserves the full LoCoMo run as a release gate only. - RED-GREEN TDD loop: Writes a failing micro fixture first, implements the smallest diff to pass it, then checks all existing fixtures for regressions. - Proactive branch and review management: Handles dirty WIP safely, creates feature branches, and spawns a senior-code-reviewer sub-agent for plan review on non-trivial features. - Use Case: You want to add the ACC EVC-adaptive recall mechanic from RESEARCH.md. The Skill picks the feature, writes a failing fixture, implements it, validates against micro and LoCoMo smoke benchmarks, and commits one logical change. ## Quick Start Run the hippo-feature skill with a feature name like acc-evc to build that memory mechanic from RESEARCH.md through the micro-eval TDD loop.
