What problem does it solve? Routines that write nothing force every run to re-learn the same facts, repeat the same mistakes, and re-ask the owner the same questions. This Skill turns a stateless routine into one where each run writes a dated memory shard and the next run reads it back before acting. ## Core Features & Use Cases - Memory contract definition: Specifies the routine, its shard name and location, run cadence, and a write-what list of at most five categories (decisions, observations, open questions). - Retention and supersede rules: Dated entries, SUPERSEDED marking on contradicted findings, and a strict no-silent-deletion policy so the shard stays a memory rather than an unbounded archive. - Read-back verification: Validates the contract by requiring two writes that a later run demonstrably read and acted on, covering both steady-state and finding runs. - Use Case: A weekly vendor-watch routine keeps re-summarizing the same competitor changelog. Apply this Skill to give it a shard where each run records dated findings and open questions, so the next run opens by stating what it is acting on from last week. ## Quick Start Give my weekly vendor-watch routine a memory shard contract defining what it writes, where it writes, and how the next run reads it back.
