Memory features: project-level, cross-project, and account-level persistence

Problem

Claude Code starts fresh every session with no memory of past interactions. There is no built-in way for it to remember what happened in previous sessions — what was tried, what failed, what decisions were made, or what the user's preferences and standards are.

Currently, users compensate by manually maintaining files like ISSUES.md, PROJECT_STATUS.md, and CLAUDE.md to give Claude context it should already have. This is a manual workaround for a missing system capability.

Proposed Solution

Memory at three levels:

1. Project-level memory

Remember past sessions within a specific project. What was attempted, what failed, what issues were identified, what architectural decisions were made. This should persist automatically across sessions in the same project directory without the user needing to maintain manual files.

2. Cross-project memory

Carry learned preferences and patterns across different projects. If a user has a consistent standard of quality, recurring feedback patterns, or general preferences (e.g., "break concepts down more granularly," "don't over-engineer"), those should be accessible regardless of which project is open.

3. Account-level memory (browser integration)

Bridge Claude Code with the user's Claude browser account. If the user discussed an idea, explored a concept, or established context in a conversation on claude.ai, Claude Code should be able to access that context. The user's interactions with Claude are currently siloed between browser and CLI — they shouldn't be.

Use Case

I'm working on a project where the same feedback comes up repeatedly across sessions (e.g., "not broken down enough," "too much clutter," "missing the conceptual core"). I have to re-explain this every session or maintain manual files. Claude Code should learn from past sessions and carry that forward automatically.