Docs / Working with Your Library

Working with Your Library

Day-to-day tasks: creating and organizing items, keeping history, importing what you already have, and getting your data out.

Create and edit items

Click + New ▾ and choose a type (prompt, skill, rule, doc, template, config, file — or a custom type). Give it a title and a body; for prompts, wrap the fill-in parts in [brackets]. Every item has a format and a file extension that come from its type.

Description — an optional one-liner your agent sees in the slash-command and resource lists (and it becomes the frontmatter of a downloaded command file).
Field defaults — give a field a default with =, e.g. [Environment = staging]. It pre-fills in the app and makes that field optional over MCP. A field with no default stays a literal [Field] until filled.
Include another item — @aisd:<slug> in a body inlines that item's current content when read or exported, so a thin command can reuse a shared template. Resolves to your nearest readable copy; grab an item's handle from the chip by its title.
Override our shape with yours — because a reference resolves to your nearest copy (personal → team/company → ours), copying a Template into your team library and editing it makes every item that references it — including our prebuilt skills — emit your house format. Keep the title the same as ours and your copy wins; delete it and it falls back to ours. Our ADR, Release Notes, Postmortem, and PRD templates are set up for exactly this.

Edits save a new version (see below), so you can always go back.

Organize and find

Categories & tags — group and label items. Manage categories (rename, recolor, delete) from the categories menu; deleting one moves its items to Uncategorized. Company-wide categories are managed by company admins.
Favorites & pins — favorite the ones you love; pin the ones you want stuck to the top regardless of sort.
Sort & drag — sort by recency, name, type, or version. Drag cards to reorder for the moment (even into the empty space below the list to send one to the end) — the sort dropdown switches to Custom (drag) while you do, and one click snaps back to the real sort.
Search — by title, content, author, who last edited it, or your notes.
Tabs — open libraries as tabs you can drag to reorder and pin the ones you live in.

Notes and comments

Context that isn't the content — and that your AI never sees.

The 📝 and 💬 counts sit by the version badge on any open item — click one to jump to that thread.

Private notes — keep as many notes as you like on any item you can open, including read-only and prebuilt ones: "why this exists," "works in Claude, not GPT," a TODO. They're private to you and never sent to your AI (only the item's body is). Each note shows as a compact block and only becomes an editor while you're editing it. Find your annotated items under Notes in the sidebar, or search My notes.
Team comments — on a shared team or company item, anyone who can see it can add to a comment thread the whole team reads. Edit or delete your own; team/company admins can moderate. Comments are never sent to the AI either — they're for the humans discussing the item.
Curator notes — some prebuilt items carry read-only Notes from AIStorageDepot — a short "use this one vs that one" pointer where two items look alike (e.g. a review prompt vs a review skill). You can read them; only we can write them, the same way the prebuilt library itself is read-only. Also never sent to the AI.
Comments travel when you copy — pull an item into your library and its comments come along. On a personal copy they show read-only (so a pulled prebuilt item keeps its "Notes from AIStorageDepot"); copy into a team and they stay a live, postable thread. Your private notes are yours alone and never copied.

Deactivate (set an item aside)

Not using an item right now — a work-in-progress skill, or a shared item you don't need? Deactivate it (the 💤 toggle). It dims out of your way in the app and your AI tools skip it (both the live MCP path and pulled commands). Deactivation is per-user — only you are affected, nobody else. Reactivate (🔊) any time to bring it back.

Follow items (🔔)

Care about an item someone else maintains? Follow it (the 🔔 toggle) and changes land in your alerts bell in the header — edits, trash/restore, deletion, and new comments by others. Alerts are in-app only, per-user, and work on any item you can see, including shared and prebuilt ones. The alert survives even if the item is later deleted, so you always know what happened.

Version history

Every change to an item is tracked with full history — what changed, when, and who did it (each version row is stamped with the person's name, so "Synced from Platform · by Alice" reads like a story). Open any item's history to restore a previous version in one click — an overzealous edit is never a disaster. If two people edit the same item at once, the app warns instead of silently overwriting, and shows who else is in the item.

Copies that stay current

When you copy an item from a shared library (a team, the company library, or the prebuilt one), your copy remembers where it came from. If the original moves ahead, your copy flags it — at two speeds:

The ⬆ update chip on the list card applies the latest in one click (your current content stays in history, so it's always undoable).
The Update available badge in the detail view opens a diff to review first, then apply.

A library with several stale copies gets an update-all bar.

Not ready to take an update? Skip this version in the review dialog: the badge and your AI tools' "newer version" notices go quiet until the original changes again. Publishing an edited copy back to a shared library updates the original there rather than creating a duplicate.

And when a family of copies grows, the Lineage tab in the detail view draws the whole tree: the original at the top, every copy you can see beneath it, each box showing whose library it lives in, its version, who last touched it, and whether it's ✓ in sync, ⬆ behind, or ⏸ skipping the current version. Click any copy in the tree to jump straight to it. Copies in libraries you can't access appear as a discreet "private library" placeholder — you see the shape of the family, never its private contents — and a very large family says plainly when the view is clipped.

Trash and recovery

Anything you delete goes to the Trash for 30 days, restorable in one click. Emptying the trash — or deleting an item "forever" — is permanent and takes its version history with it.

Import what you already have

Bring in scattered prompts without retyping:

Files or a .zip — drag them in.
A GitHub repo — point us at it and we pull the contents. Private repos work too: paste a GitHub token with Contents: read on that repo. It is used for that one import and never stored, so you paste it again next time.

Everything lands auto-detected as the right type (an mcp.json becomes a Config, an AGENTS.md becomes a Rule, a markdown file with template in its name becomes a Template, and so on) so you just review and save. A Claude Code command file (or any file using $1 / $ARGUMENTS with YAML frontmatter) is converted on the way in — its positional args become [fields] and its description is lifted — so it lands as a first-class prompt.

Download and export

Your data is never locked in:

Download any item as a clean Markdown, JSON, or YAML file (named title.ext). Drop it into your repo as the file your tool reads — AGENTS.md, .cursor/rules/*.mdc, mcp.json. File-type items download with their own extension (.gitignore, .sh, whatever you set) — see Item Types → Files.
Export a .zip of many items at once.
Pull everything over the API and MCP.
A company owner can export every library across the whole organization as one .zip, foldered exactly as it appears in the app.

Give any item a public read-only link. Strangers see a clean view with no account needed; teammates who sign in open the real item with their normal permissions. Links stay active until you revoke them (revocation is immediate).