grandamenium/dream-skill — reverse-engineered prompt

Build me a small Claude Code skill called Dream that keeps the agent memory tidy automatically. I want to be able to run /dream manually, and also have it run on its own after about 24 hours of normal Claude Code use.

The skill should look through the current memory setup, scan recent session transcripts for useful signals like corrections, changed preferences, important decisions, and repeated patterns, then merge those into memory cleanly. It should fix stale or conflicting notes, turn vague dates into real dates, remove references to files that do not exist, avoid duplicates, and keep the main memory index short and easy to skim.

Please include a simple installer, a stop hook that checks whether enough time has passed, and a small test script that proves the consolidation flow works. It should detect common memory layouts, including native Claude Code memory, OpenClaw style memory, or a MEMORY.md in the project root. Look up current Claude Code skill and hook docs online if needed.