consolidate-memorylisted

# Consolidate Memory Scan the current conversation, extract what's worth keeping across future sessions, and write it to the project's auto-memory system. ``` read context → classify signals → draft memories → write files → update MEMORY.md ``` ## 1. Compute the memory directory ```bash MEMORY_DIR="$HOME/.claude/projects/$(pwd | sed 's|^/||; s|/|-|g')/memory" ``` Verify it exists: `ls "$MEMORY_DIR"`. If the directory doesn't exist, the project has no memory system — tell the user and stop. ## 2. Check what's already documented Before scanning for signals, grep the CLAUDE.md files to know what's already written down — anything captured there does NOT need a memory file: ```bash # Find all CLAUDE.md files in the project tree find . -name "CLAUDE.md" -not -path "*/node_modules/*" | head -10 # Then grep them for relevant terms as you identify candidate learnings grep -l "git clone\|TUI\|bootstrap" CLAUDE.md */CLAUDE.md 2>/dev/null ``` If a learning is already stated in any CLAUDE.md, skip it. Memory files are for what code/docs can't capture — behavioral preferences, incident context, team decisions, non-obvious constraints. ## 3. Scan for signals worth keeping Read the entire visible conversation. Extract only what a future Claude instance couldn't derive from the code, git history, or CLAUDE.md files. **High-value signals:** - User corrections — "no, not that", "stop doing X", explicit pushback on an approach you chose - Validated non-obvious choices — user accept