The durable workflow: install the companion plugin once, then drive everything from inside Claude Code with /install-memory-file, /list-memory-files, /update-memory-file, /remove-memory-file, and /submit-memory-file.

One-time setup 

/plugin marketplace add riboseinc/claude-memory-files
/plugin install memory-files-tools@claude-memory-files

The plugin ships only the five slash commands; install.sh, uninstall.sh, and update.sh are still fetched from main on each invocation (so the installer logic stays current with the catalogue).

The five commands

/list-memory-files [--tag] [--scope] [--team] [--installed]

Browse the catalogue with filters. --installed shows only what’s on your machine. --include-deprecated shows tombstoned entries.

/install-memory-file <slug> [--project <dir>] [--ref <sha>]

Install by slug. The command looks up the file in the catalogue index, shows you the frontmatter (description, type, scope, owners, version, requires-companion warnings), and asks before installing.

/update-memory-file <slug> [--force] [--upstream] [--ref <sha>]

Re-fetch upstream and apply the decision matrix. Aborts if local has been hand-edited (use --force to override). For personal-share (forked) entries, defaults to no-op unless --upstream is passed.

/remove-memory-file <slug>

Uninstall. Shows what will be removed, asks for confirmation. For settings-fragment, the manifest-recorded entries are subtracted and hand-curated entries are preserved.

/submit-memory-file <local-path>

Scaffold frontmatter for a new file (drafted interactively with you), clone the repo as a sibling, run the validator, commit, and open a PR. See Submitting your own file for the walkthrough.

A typical session 

> /list-memory-files --scope universal --tag github

[command shows ~7 entries: github-pr-title-issue-link, github-content-via-file, ...]

> /install-memory-file github-pr-title-issue-link

[command shows frontmatter, asks "install? y/n"]
> y

[command installs; reports target path and manifest update]

> /list-memory-files --installed

[command shows just what's on your machine]

Why both curl and slash commands?

Two different friction curves. The curl one-liner is the lowest-ceremony entry point for "I saw a file referenced in Microsoft Teams and want to try it once"; no plugin install required. The slash commands are the durable workflow for managing multiple installs, surfacing outdated entries, and exposing the submission flow.

Both install paths write to the same ~/.claude/.memory-files-manifest.json. You can curl-install something and remove it via slash command, or vice versa.

See also