Claude Memory Manager — Development skill for Claude Code

Memory Manager

A desktop app that draws your Claude Code memory bank as a game-style radial skill tree — a glowing hub, colour-coded sectors fanning outward, hexagonal nodes you can drag, search and prune.

It exists because a memory bank is a context budget you cannot see. Files accumulate, the index quietly stops matching the folder, and the first sign of trouble is a session that has clearly forgotten something. This makes the bank a thing you can look at.


*The `~/.claude` level. Each leg is a category root fanning out to its members; the rings and numbers on the tool nodes are real usage scanned out of session transcripts, and the amber badges are rule violations. Double-click a project to drill into it.*

What it actually does

• Browse at three levels, all the same graph. ~/.claude → a project → that project's memories. Double-click to drill in, breadcrumb back out. There is no list view anywhere, on purpose.
• Read a real memory folder — frontmatter and [[wiki-links]] become nodes and edges. Links to memories you have not written yet show up as ghost nodes.
• Write it back without damaging it. Export is round-trip safe: frontmatter the app does not model is preserved verbatim in its original key order, dates are written back wherever the file already kept them, and the existing folder is backed up to a timestamped subfolder first. Pinned by a test that asserts zero frontmatter keys are dropped.
• Rules and limits. Secret patterns block a save or export; size limits (CLAUDE.md too long, index too long, bank too large) raise amber badges on the offending node. Rules are plain JSON, inherited built-in → user → project → bank, and every effective rule shows where it came from and what it replaced.
• Show what a session will actually load. A bank with memor