Problem

Claude Code's auto-memory system currently has two levels:

1. Global (~/.claude/CLAUDE.md) — applies to every project
2. Project-specific (~/.claude/projects/<path>/memory/) — scoped to a single project directory

There is no way to share memory across a subset of projects. In practice, a lot of knowledge is relevant to multiple (but not all) projects — e.g. shared infrastructure, team conventions, API configurations, deployment patterns, or organizational context that spans several repos.

Current workaround

Duplicating memory entries across project-specific memory directories, or putting everything into global CLAUDE.md even when it only applies to a few projects. Both approaches are fragile and lead to stale or noisy memory.

Proposed solution

Introduce a shared memory directory that can be referenced by multiple projects. For example:

• A configurable list of shared memory paths per project (in .claude/settings.json or CLAUDE.md)
• Or named memory namespaces (e.g. ~/.claude/shared-memory/<namespace>/) that projects can opt into

Example configuration:

{
  "memory": {
    "shared": ["~/.claude/shared-memory/sediwork"]
  }
}

This would allow memories like "Jira project key is SDW", "Bitbucket repo is kupuhealth/frontend", or "team uses pnpm in frontend, npm in backend" to live in one place and be available to all related projects — without polluting the global scope.