Memories

What a project has told the harness to remember, plus the two problems you cannot see from inside a session.

What they are

Claude Code keeps per-project memories as markdown files under ~/.claude/projects/<project>/memory/, with a MEMORY.md index listing them. Each file has frontmatter naming it, describing it and giving it a type, and bodies link to each other with [[name]].

Press M on the map, or:

braids memories
braids memories --project braids
braids memories --json

The two things braids can see that you cannot

Mark	What it means	Why it matters
⊘	in the directory, missing from MEMORY.md	the index is what gets loaded, so nothing ever reads this file
⊕	a `[[link]]` to a memory that was never written	the note meant to point somewhere and does not

Both are invisible from inside a session, which is the point of showing them here. A memory that is not in the index is a note you wrote and the model will never see.

Reading one

↵ opens the memory with its markdown rendered: headings, emphasis, lists, code, links. c jumps to the conversation that wrote it, when the frontmatter recorded one.

Emphasis follows the same two rules a markdown renderer needs to be usable on real notes: 2 * 3 is arithmetic, not italics, and snake_case_names keep their underscores.

Changing them

Key	What it does
`r`	rename, following the name everywhere it appears
`i`	repair the index for this memory
`d`	delete it to the bin
`n / N`	next, previous marked memory
`c`	the conversation that wrote it
`f`	filter the list

A rename is four edits in one: the filename, the name: in the frontmatter, the row in MEMORY.md, and every [[link]] in every other memory that pointed at it. A repair adds a missing row to the index, or reports the loose link it cannot fix by itself, since braids will not invent a memory you have not written.

braids preserves the file mode it found. An earlier version tightened memory files from 0644 to 0600 as a side effect of rewriting them, which is not braids' decision to make.