Skills and memories

Every workspace ships with two native objects, skill and memory. The agent reads skills to know how to do a task and memories to know what it learned about your data. Both are ordinary records: the same REST routes, filters, search, views, permissions and MCP tools your own objects get apply to them. This cookbook creates one of each from the SDK and shows the agent using them.

What you will build

A skill named invoice-followup that the agent follows the next time someone asks it to chase a payment, and a memory attached to one of your companies records that the agent finds when that company comes up. You verify both from the SDK, the CLI, the search endpoint and the app.

What a skill is

A skill is three attributes and, optionally, attached files:

New skills start from a template with the sections ## Overview, ## When to use, ## Instructions, ## Examples and ## What to avoid. Files attached to a skill travel with it: a reference PDF or a CSV template is copied next to the agent when it loads the skill.

What a memory is

A memory is a note the agent keeps across conversations, with a lifecycle:

A few more attributes (links, provenance, the curation timestamps) are maintained by the workspace; you can read them and leave them alone.

There is no owner attribute. A memory is workspace or private like any record, and a private one is visible only to the actor that created it.

Prerequisites

• A workspace, an API key whose role reads and writes skill, memory and your own objects, and @stndrds/client installed.
• One record of your own to attach the memory to. The steps use a companies record named Acme; any object works.

standards schema inspect skill
standards schema inspect memory

import { getStandards } from "./client";

const skills = getStandards().from("skill");

await skills.create({
  name: "invoice-followup",
  description: "Drafts a polite reminder for an overdue invoice. Use when asked to chase a payment.",
  content: [
    "## Overview",
    "Write a short, courteous reminder for an invoice that is past its due date.",
    "",
    "## When to use",
    "- When asked to chase, follow up on or remind about a payment.",
    "",
    "## Instructions",
    "1. Read the invoice record: number, amount, due date, contact.",
    "2. Write three sentences: the invoice, the amount and date, one clear ask.",
    "3. Sign with the sender's first name only.",
    "",
    "## What to avoid",
    "- Never threaten fees or legal action.",
  ].join("\n"),
});

const companies = getStandards().from("companies");
const memories = getStandards().from("memory");

const acme = await companies.eq("name", "Acme").single();
if (!acme) throw new Error("create the Acme company first");

await memories.create({
  name: "acme-prefers-french",
  description: "Acme wants every message in French, addressed to Claire Morel.",
  kind: "preference",
  source: "explicit",
  subjects: [acme.id],
});

curl -X POST https://api.standards.new/v1/records/memory \
  -H "Authorization: Bearer $STANDARDS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "data": { "name": "my-note-on-acme", "description": "Call after 14:00.", "kind": "fact", "source": "explicit" },
    "options": { "visibility": "private" }
  }'

const about = await memories.in("subjects", [acme.id]).fetch();
about.records.map((memory) => memory.name); // ["acme-prefers-french"]

standards records search memory --query french --format table

curl "https://api.standards.new/v1/search?q=french&objects=memory,skill" \
  -H "Authorization: Bearer $STANDARDS_API_KEY"

What the agent adds on its own

When it saves a memory itself, the agent calls save_memory, which first searches for close matches. If it finds any it saves nothing and returns the candidates; the agent then re-calls with updateMemoryId to update one, or force: true to create anyway. After a conversation, the workspace extracts new memories as private records, some flagged for review, with source: "extracted"; a nightly pass promotes the ones several people agree on and writes source: "corroborated" or curated. You never have to run any of that. As long as your own writes set source, the attribute tells what was saved by hand from what the workspace learned.

What you cannot do

• Change or delete the native objects' attributes, or name your own object after one of them.
• Exceed 100 characters on name or 160 on description; the API answers SCHEMA_VALIDATION_FAILED.
• Call save_memory, search_memories, get_memory or get_skill over MCP: they are agent-only tools and never appear in tools/list. Over MCP, read and write the two objects with search_records, get_record, create_record and update_record like any object.
• Set a memory's owner: ownership comes from the key that created it.

Next steps

• Records: every filter used above, including in on a relation.
• Objects and records: visibility, workspace versus private.
• Tools: what an MCP client can do with skill and memory.
• Agent workflow: the same objects from Claude Code.