Glossary
Terms are grouped by where they belong in the system. Business records are defined in more depth in Shared business records.
The shape of the system
Section titled “The shape of the system”Workspace — The one place you and your agents both make things and see what the other made. It has six kinds of surface. See Your workspace.
Surface kinds — Talk (chat) · See (command views) · Review (the Workbench) · Make (the Studio) · Run (engine cockpits) · Keep (catalogues), plus System screens for whoever runs the workspace.
Pillar — One of the five things the workspace stands on: 01 Agents (who) · 02 Goals, tasks, artifacts (what) · 03 Capabilities (how) · 04 Shared brain (with what) · 05 Engines (what runs without you). See How it works.
Agent loop (also work loop) — The life of one piece of work, told from the agent’s side: Trigger → Workflow → Context → Skill → Brand → Action → Track → Learn, with what it learned feeding what the next request already knows. It runs across the pillars. See How work gets done.
Station — One of the eight places in the agent loop. Each answers one question, is hosted by one or more pillars, and is tied to the files and tables that do it.
Action — The sixth station. An action is one of three kinds: make something, pull a lever in a system, or ask a person.
Foundation (also plane) — What every pillar stands on: identity and access, and data governance. Connections is read as part of the Capabilities pillar. See Foundations.
The WilsonOS Five — The five measurable promises the system is built to keep. See What is WilsonOS? § What you get.
Agents
Section titled “Agents”Agent — A long-running assistant with an identity, a boot contract, and a lane. Brand agents act as the chief operating officer of one brand; cross-cutting agents work across brands or on the platform itself.
Brand agent — The agent assigned to one brand. Scout runs Code Pineapple; Prime runs RiseBlox. Each loads its brand’s context at the start of every session.
Boot contract — The exact set of context an agent loads when a session starts: identity, the operator’s profile, the brand essence, its own recent memory, and its open plate. Defined per agent; nothing else loads “just in case”.
Lane — The area of the system an agent is responsible for. Brand agents deliver business assets; a platform agent builds the app; cross-cutting agents coach and steward.
Run — One execution of a capability against a task: what was invoked, with what inputs, producing which artifacts. A task can involve multiple runs.
Capabilities
Section titled “Capabilities”Skill — An atomic, reusable capability: a method an agent masters right before performing a task (write a Pinterest post, render a brand book). Brand-agnostic; loads the brand stack at run time.
Flow — A multi-step orchestration of skills that produces an outcome (take a brand from “no Pinterest presence” to “posting weekly”). Agents execute flows; flows call skills.
Brand stack (also brand context) — A brand’s products, audience, voice, visual rules, customer language, and constraints. Loaded by skills and flows so the same capability produces on-brand work for different brands.
Connection — An external system reached through an API, MCP server, CLI, or equivalent connector, with its own account scope, permissions, and supported read and action capabilities.
Attribution — The chain linking a task to the skill and version that ran, the run, the artifact it produced, and the review and rating it received. What makes “show me everything this skill produced and how it was rated” answerable.
Memory and context
Section titled “Memory and context”Context — Everything an agent knows before it acts: memory, brand canon, and the boot contract. The third station of the loop; the brand stack is the fifth.
Memory — What persists across sessions: the operator’s profile, brand truth, capability knowledge, operational facts, and each agent’s working notes. Layered, scoped, and loaded on need rather than all at once.
Canon — The authoritative version of a fact or document. Each class of data has exactly one place it canons and exactly one path that writes it.
Recall — The search and boot path that surfaces the right memory for a task without loading everything.
Work and artifacts
Section titled “Work and artifacts”Goal / project / task — Intent, the body of work toward it, and one unit of planned work. Chat messages reference tasks without becoming their storage location.
Deliverable — An artifact marked as a result someone should review, use, or publish — as opposed to a by-product of getting there.
Artifact — A durable output: a file, page, image, video, document. Identified by a stable reference, not by its file name or location.
Version — One explicit state of an artifact. Feedback attaches to a version; a new version does not overwrite the old one’s history.
Registry — The record of what artifacts exist and which version is current. The registry owns identity; storage holds bytes; a name is a label.
Workbench — The review surface: where tasks, deliverables, and artifacts of every type are opened, commented on, rated, approved, and published.
Durable publication — Storing a finished media file in the system’s own durable storage and registering it, as part of finishing the work. Enforced by the publish gate: a task cannot be marked review or done until it has happened. Nobody outside has seen anything.
Publish — Moving an approved artifact version out through a connection (a channel, a store, a site) and recording the external identifier that came back. Distinct from durable publication; a client’s approval never authorises it.
Share — Granting a person a permission — view, comment, approve, download — on a specific version of a deliverable or a collection. A new version never inherits the previous version’s approval.
Collection — A chosen set of deliverables (or specific versions) with a title and an order: the unit you share or publish when one deliverable is not the story. A client delivery page is a collection fixed to delivered versions; a portfolio is one that follows later updates.
Mini app — A small guided surface over one skill: ordered steps, structured inputs, one generation or one skill run at the end. The surface declares inputs and display, never logic.
Studio — The named destination for detailed creation: the Create bar, the Gallery, Elements, Design Systems, and projects. Works on the same tasks and versions as everything else.
Element — A reusable building block a generation can reference — a character, location, prop, art style, voice — kept once and tagged into any scene.
Project (Studio) — A task with a scene tree written onto its artifacts by tags: rounds of generation per scene, the best takes kept, the result assembled into a new version on the same task.
Measurement and learning
Section titled “Measurement and learning”Signal — An observed external event with a source, a date range, and a freshness: an impression, a click, a sale, a rating.
Source · shared record · derived metric · agent analysis — The four steps connected data moves through: what a provider reported, with provenance and freshness; compatible concepts connected where an identifier supports it; a calculation with a named formula over named inputs; an interpretation labelled as judgement. See Connected data.
Contribution — Royalties minus the advertising that produced them, for a title or a brand over an aligned period. Not profit: production, tools, labour, and returns are not in it, and the system never labels an incomplete calculation profit.
Attribution limit — A stated boundary on what a signal can prove. A funnel stage is not proof of a causal conversion.
Capability evaluation — Comparing versions of a skill or flow against explicit criteria, recording which version and inputs produced each result.
Promotion — Deliberately making an improved version the current one, with the evidence and reasoning preserved. Never automatic.
Sharpening — The nightly and weekly processes that distil what was written during work into durable memory and improved capabilities. They cannot capture what was never written.
Say it like this
Section titled “Say it like this”Plain words for everyone. The engineering terms live in each page’s “built on” notes.
| Don’t say | Say | Why |
|---|---|---|
| Shared workspace · agentic workspace | Your workspace | “Shared” clashes with sharing between people; “agentic” names only the agent’s half. |
| Workbench or Studio, for everything | Review happens in the Workbench; making happens in the Studio | Both already mean something precise. |
| A list of screens and routes | Talk · See · Review · Make · Run · Keep | Seventeen screens bury the reader; six verbs do not. |
| Subject | You, or one of your businesses | Same idea, plain words. |
| Context packs · boot bundle · memory | Your brand’s brain | Shared with everyone plugged in, person or agent; every entry shows who added it. |
| Capability · tools · MCP | Capabilities: what it knows how to do and what it can reach | A tool is the door; a connection is the account behind it. |
| Authority ladder · agent trust | Battle-tested skills and flows earn more rope | Trust sits with the skill or flow, not the agent, and you can always veto. |
| Gets you to do your part | Helps you do your part | It focuses you on the right work and helps you finish it; it should not feel like more work. |
| Gets smarter every cycle | Calls it, does it, checks it — the gap is what teaches it | The mechanism is true and different; the outcome is what everyone claims. |
| Ingest → conformed → synthesis | One sentence inside Engines | Plumbing with no story of its own. |
Availability vocabulary
Section titled “Availability vocabulary”Available · Limited · Awaiting verification · Planned — The four labels in What works today, the one page that states availability. Each row also names its evidence type: instruction audit (the rule is present in every flow or configuration), runtime acceptance (a dated test or live run met the criteria), runtime check, live use with no dated record, reported deployment, or written design. See How to read these docs.
Organisation · person · brand · access grant — The identity model: an organisation holds people and brands; a person has credentials; a brand is a workspace; an access grant is a person’s rights on a brand. Agents are none of these — they act on behalf of a person within a brand. See People and teams.
Sources: input/WilsonOS_Overview.html §03, §05, docs/architecture/overview.md, docs/architecture/skills-and-flows.md, docs/architecture/memory.md, docs/DECISIONS.md (ADR-028), WilsonOS Team Brief (prototype), "Our words, not system words" — absorbed here (task 494) · Last verified 2026-10-06

