Storing things is the easy half. This is the field guide to using them — each item type, in plain English, with the laziest workflow that works. Every section has a ▶ mini-demo — press play and watch it happen.
Live connection (MCP) — the recommended way. The agent reaches into your library on demand, and the items you mark with / show up as slash‑commands — always your latest version. Set it up once per tool on the Connect page; it’s all most people need.
Installed commands (optional) — npx -y @aistoragedepot/mcp pull writes items into each tool’s commands folder as typed /commands files. You don’t need it for slash commands — that’s what marking with / does — but pulled files need no token, work offline, and cover tools that don’t show MCP prompts. They stay frozen until you re-run pull.
See the same item twice in a slash menu? It’s both marked / and pulled as a file — (user) is the frozen copy, (MCP) is live. Pick (MCP) — it’s always current — or skip pulling things you’ve marked.
Everything here is free for individuals — 2,500 MCP / API pulls a month, 30,000 on Plus. Team and Company plans add shared libraries, roles, and controls.
A reusable message with [fields] — “Draft a cold outreach email from [Sender] to [Recipient] about [Topic].”
In the app: open it, click each field, fill, hit Copy result — or Open in ChatGPT/Claude to open your AI with it prefilled.
As a slash command, two ways to feed it. The precise way: give the values in order, each wrapped in quotes — /your-prompt "first value" "second value" — the quotes mark where each field starts and ends, and your editor shows you which field you’re filling as you type.
The lazy way: skip the quotes and just brain-dump in plain English on the same line — don’t try to remember the fields. The AI maps your words onto them and asks you for anything you missed. Typing the command with nothing after it works too: the AI interviews you, one question at a time.
A method the AI follows — “how to do a proper code review”, “how to write a conventional commit”. No blanks to fill; it’s knowledge, not a form.
Over MCP, don’t type a command — just ask: “review this diff the way my code-review skill says” and the agent goes and reads the skill mid-conversation.
As a slash command: pressing it pastes the instructions into chat and the AI follows them on whatever you’re doing. Two ways to get one: mark the skill with / in the app (a live (MCP) command), or install a local copy with pull.
Skills can use other skills. There’s no special syntax — a step that names another skill (“write the commit by following the commit-message skill”) makes the agent read that skill and follow it for that step. Keep each skill small and single-purpose, and let bigger skills orchestrate. The prebuilt library’s Ship-it checklist example shows the pattern in three lines.
Instructions a coding agent should always follow — TypeScript standards, review etiquette, “never use any”.
Main move: Download. Grab the rule and drop it into your project or user file — AGENTS.md, .cursor/rules/*.mdc — and from then on your tool loads it every session, so the agent follows it automatically.
Over MCP: “follow my team’s TypeScript rules” — the agent fetches and applies them without the file ever landing in the repo.
Anything the AI should be able to look up — architecture notes, API conventions, onboarding guides, cheatsheets.
Over MCP: “check my deployment doc before answering” — the agent reads the current version on demand. No copy-paste, never stale.
Ready-to-use mcp.json snippets — which MCP servers a project should have. Download drops the file where it belongs; your whole team pulls the same tool setup from the shared library.
The shape a finished output should take — an ADR, a PRD, release notes, a postmortem. Where a Doc describes (“here’s how this works”), a Template prescribes (“emit it like this”). Reference one from a prompt or skill with @aisd:<name> and its body is inlined when the AI reads it — so the output lands in exactly that shape.
The team trick: copy a prebuilt Template into your team library and edit it — anything referencing it by name picks up your version instead of ours, no forking. The prebuilt library ships ADR, PRD, release-notes, and postmortem Templates to start from.
The catch-all for anything that isn’t one of the other six — an ignore file (.gitignore, .dockerignore), a shell script, a snippet, a one-off note. Store it, tag it, version it like everything else.
The trick: each File keeps its own name and extension, so a download lands as the real .gitignore or deploy.sh — ready to use, not a renamed .md. Imported ignore files land here automatically.
Prompt → mark it /, then slash command + brain-dump (or fill fields in-app) · Skill → ask for it by name, or mark it / and type it · Rule → download into the repo (or ask the agent to pull it in over MCP) · Doc → tell the agent to read it · Template → reference it from a prompt or skill with @aisd:<name> — the AI shapes its output to match · Config → download into place · File → download — it lands with its real name & extension.
And everything, of every type, can be copied, downloaded as a real file, shared with a link, or published to your team.
Take your first items from the prebuilt library’s Start-here examples, fill in a prompt’s fields and open it in your AI, browse the prebuilt library, and see version history at work — with pause, step-back, and step-forward controls. Watch the tour →
Sixty seconds per chapter, cursor does the clicking: stock the team library, pull a copy to improve, sync it back without duplicates, and branch off a new item when your version becomes its own thing. Watch the walkthrough →