What problem does it solve? Long-running agent workflows lose context between task runs, causing repeated mistakes, rediscovery of known constraints, and inconsistent decisions. This Skill keeps workflow memory files current so each Compozy PRD task run starts with the right durable context. ## Core Features & Use Cases - Memory Loading and Updating: Reads shared workflow memory and current task memory before code changes, then updates them as objectives, decisions, and learnings evolve. - Promotion Discipline: Applies a three-question test to decide which facts move from task memory into shared cross-task memory. - Compaction: Rewrites noisy or repetitive memory files into short factual bullets while preserving decisions, risks, and handoffs, guided by references/memory-guidelines.md. - Use Case: During a multi-task PRD implementation, an agent records a discovered API rate limit in shared memory so every subsequent task respects it without rediscovering the constraint. ## Quick Start Read the shared workflow memory and current task memory files at the provided paths, keep them updated during this task, and compact them if flagged before finishing.
