13.3.0 — The Library grows: memory you can tidy, promotion, tools, storage and repeats

A snapshot as of 13.3.0 (released 2026-11-21). It will go stale as the app moves on — that is expected, and the living reference is the guide each section links to.

Upgrade:

npx mulmoterminal@latest

Nothing to configure. Most of this release is in the Library (More features → Library). Its row of kinds under the search box now reads Conversations · Memory · Repeats · Tools · Storage.

New features

Memory: see what can be tidied, and fix it in place

In 13.2.0 the Memory kind could only be read. Now it can be changed — and it tells you what is worth changing.

The Memory kind: a memory open in the editor

Promote a conversation to a memory, rule, plan, issue or skill

Open a conversation in the Library and press ⋯ at the bottom. Under Promote to are A memory, A CLAUDE.md rule, A plan, An issue and A skill.

A conversation's ⋯ menu with Promote to and its five targets

Pressing one closes the Library and opens a Claude cell in plan mode, in the directory the conversation ran in. It runs the bundled mulmoterminal-library skill, which reads the conversation, drafts what it would write (for a rule it asks which CLAUDE.md; for a skill, whether to make a new one or improve one you have) and writes nothing until you approve the plan. Every file it changes is backed up first under ~/.mulmoterminal/library/backups/, and it tells you how to undo each change (#3366).

Repeats: the instructions you keep typing

Repeats lists what you typed 3 times or more, in 2 conversations or more, in the last 90 days — read from Claude Code’s prompt history. Slash commands, ! lines, pasted text, automated runs and very short replies are left out, and the same words written slightly differently are grouped.

The Repeats kind: repeated instructions on the left, one open on the right

Pick one to see how else you typed it and the conversations you typed it in lately (click one to open it). Make a rule or Make a skill opens Claude in plan mode with the mulmoterminal-library skill, in the directory you typed it in most. It proposes the rule or skill and writes nothing until you approve (#3393).

Tools: what Claude Code loads in a directory

Tools asks for a Directory (one a conversation ran in) and lists the instruction files Claude Code loads there, in the order it loads them: the organization’s, yours, each CLAUDE.md and CLAUDE.local.md from the root down, the project’s rules, and the files they import with @path — or AGENTS.md where there is no CLAUDE.md. Pick one to read it beside the list (#3367).

The Tools kind: the instruction files for one directory, and the settings laid over each other

Read only: nothing here writes a file.

Storage: what Claude Code’s folder takes on disk

Storage shows each entry of ~/.claude, largest first, with the total. Pick projects/ to see each project, largest first, with the ones from temporary directories marked temp; type to narrow them. It also says what cleanupPeriodDays is, and so how long Claude Code keeps conversation transcripts.

The Storage kind: the entries of ~/.claude by size, and projects/ open

It is measured in the background, again when the last measurement is over an hour old, or when you press Measure again. Nothing is removed (#3384).

More: Feature reference — Library.

Edit a skill’s SKILL.md

In More features → Skills, one of your own skills or a project’s now has an Edit button. It opens the whole SKILL.md: ⌘Enter saves, Esc cancels. As with memory, a save is refused if the file changed since you opened it. Plugin skills and the ones MulmoTerminal ships stay read only. A project’s skill is in that project’s .claude/skills, which is usually committed (#3387). More: Feature reference — Skills viewer.

Bring everything since last time from another terminal

In a cell, Tools → Talk to another terminal… lists the other terminals. Each row now has a list icon beside Bring … last turn here: it brings every exchange since the last time this directory received one from that terminal, prompts and replies, without tool output. It is pasted, not sent — you press Enter. The first time, it brings as much as fits (#3392). More: Feature reference.

The last weeks, in Usage

More features → Usage now opens with The last weeks: what twelve weeks of conversations cost, how many there were and how many PRs they made — by week, by project and by agent. It is added up from the Library’s index; a conversation counts in the week it started (#3390). More: Token usage.

The GitHub board follows your open terminals

In the GitHub view, Open the board (experimental):

More: GitHub — the board.

More agents’ turns can be brought over and read

More: Agents.

What looks different

Under the hood