[FEATURE] Auto-memory should support user-scoped entries, not only project-scoped

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Auto-memory is always stored per-project under ~/.claude/projects/<slug>/memory/, where the slug is derived from the cwd Claude was launched in. This means a memory saved while running Claude from ~ will not be loaded when Claude is later launched from ~/some-project. This is fine for project-specific facts, but it is the wrong default for user-specific preferences. A memory like "user prefers shell, Node, or Ruby over Python for ad hoc scripts" is a fact about the human, not about any one codebase. Scoping it per-project means the preference silently fails to apply in other projects, and the user gets inconsistent behavior across cwds without any indication of why.

If the assistant said "I cannot remember preferences across projects," users would compensate by repeating themselves. Instead, the assistant confirms it saved the memory, the user believes they have taught Claude a global preference, and then a week later in a different repo Claude does the opposite. That is a worse UX than having no memory at all, because it quietly violates the user's mental model.

Proposed Solution

Let the assistant mark a memory as user-scoped or project-scoped at save time. Store user-scoped memories in a single cwd-independent location (for example ~/.claude/memory/) and load them in every session regardless of where Claude was launched. The existing per-project directory continues to hold project and reference memories. A reasonable heuristic for the assistant: memories of type: user and most of type: feedback should default to user-scoped, while type: project and type: reference should default to project-scoped. The assistant can ask when genuinely ambiguous.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

1. Run Claude from ~. Tell it "I do not want to use Python for scripts, prefer shell or Node."
2. Claude saves a feedback memory under ~/.claude/projects/-Users-me/memory/.
3. Later, run Claude from ~/some-repo. Ask it to write a quick scraper.
4. Claude reaches for Python, because the memory lives in a different project slug.

Additional Context

_No response_