knowledge-dreamlisted

# Knowledge Dream Synthesize scattered project learnings into durable knowledge — inspired by Claude Code's autoDream memory consolidation engine. ## Concept AI agents accumulate context across sessions: decisions made, patterns discovered, bugs fixed, architecture choices. Without consolidation, this knowledge scatters and decays. The Dream is a periodic pass that gathers recent signal and writes it into persistent, well-organized files that future sessions can orient from quickly. ## Where Memory Lives Memory is just markdown files in your project. No special infrastructure needed. ``` your-project/ .claude/ memory/ ← Agent memory lives here (Claude Code convention) MEMORY.md ← Index: what's known, pointers to topic files architecture.md decisions.md patterns.md team-conventions.md docs/ ← Or here, if you prefer project docs CLAUDE.md ← Project-level context (Claude Code reads this automatically) ``` **Any directory works.** The skill doesn't depend on a specific path. Pick a convention: - `.claude/memory/` — Claude Code's native memory dir - `docs/knowledge/` — if you want it in project docs - `.ai/memory/` — generic, works with any agent - `CLAUDE.md` / `AGENTS.md` — single-file for small projects ## The 4-Phase Process ### Phase 1: Orient Scan the current knowledge landscape before changing anything. ``` 1. List the memory directory — what topic files exist? 2. Read the index file (MEM