Feature Request: Auto-load project-level .claude/memory/ files (git-tracked team memory)

Problem

Currently, Claude Code auto-loads memory from the user-level path (~/.claude/projects/<hash>/memory/MEMORY.md) but does NOT auto-load from a project-level .claude/memory/ directory inside the git repo.

This means:

• Auto-memory is per-user and cannot be shared with the team via git
• Project-level .claude/memory/MEMORY.md files exist in repos but are never seen unless manually read
• Teams end up duplicating memory across individual user-level stores, or relying on CLAUDE.md which serves a different purpose (instructions vs learned knowledge)

Proposed Solution

Auto-load .claude/memory/MEMORY.md (and optionally other .md files in .claude/memory/) from the project directory (git repo), similar to how CLAUDE.md and .claude/rules/*.md are auto-loaded.

This would enable:

• Team-shared memory: Known issues, debugging notes, architectural decisions tracked in git
• Persistent context: Bug lists, known issues, and patterns that survive across sessions and team members
• Separation of concerns: CLAUDE.md = instructions, .claude/memory/ = learned knowledge

Current Workarounds

1. Put everything in CLAUDE.md (bloats the instruction file)
2. Use .claude/rules/*.md (semantically wrong - rules are instructions, not memory)
3. Add "read .claude/memory/known-issues.md at start" to CLAUDE.md (manual, fragile)
4. Duplicate content in user-level memory for each team member (not shareable)

Suggested Behavior

• Auto-load .claude/memory/MEMORY.md from project root (same 200-line limit as user-level)
• Optionally auto-load other .md files in .claude/memory/ (like .claude/rules/ does)
• Project-level memory loads in addition to user-level memory (not replacing it)
• Project-level memory is read-only for Claude (Claude writes to user-level, team curates project-level)