Key Concepts

Wardian has a few connected parts. This page names them in the same way you see them in the app, so you can tell what you are opening, changing, or sharing.

The App Layout

Wardian window
├── Left sidebar
│   └── Tools such as Explorer, Source Control, Command, and Settings
├── Workbench
│   └── Pane (a split region)
│       └── Tab (an open view)
│           └── Content such as an agent, Dashboard, Inbox, or Library
└── Right roster
    └── Watchlists, teams, and the agents you can monitor or target

┌──────────────┬───────────────────────────────┬────────────────┐
│ Left sidebar │ Workbench                     │ Right roster   │
│              │ ┌───────────┬───────────────┐ │ Watchlist: All │
│ Explorer     │ │ Pane 1    │ Pane 2        │ │ ├─ Build team  │
│ Command      │ │ [Agent A] │ [Dashboard]   │ │ │  ├─ Agent A  │
│ Settings     │ │           │               │ │ │  └─ Agent B  │
│              │ └───────────┴───────────────┘ │ └─ Agent C     │
└──────────────┴───────────────────────────────┴────────────────┘

Term	What it means
Left sidebar	Tools that act on the current selection without replacing what is open in the center.
Workbench	The central area where you arrange the views you are using.
Pane	One region of the Workbench. Split a pane to work side by side.
Tab	One open view inside a pane. Move, close, or reopen tabs without stopping an agent.
Right roster	The persistent list used to monitor agents and choose the agents that tools should target.

How an Agent Is Composed

An agent is a working instance of a coding assistant, not a terminal tab or a row in the roster. It brings together the context for this task, a reusable setup, and a live provider session.

Agent
├── This instance
│   ├── Name, agent ID, provider, and model
│   └── Provider choices such as approvals and sandboxing
├── Task context
│   ├── Workspace: the main project folder
│   ├── Project instructions and files in that workspace
│   ├── Additional folders it may read (optional)
│   └── Worktree: an optional isolated Git checkout for branch work
├── Reusable practice
│   ├── Skills
│   │   ├── Shared with your user profile
│   │   ├── Given to the agent's class
│   │   └── Assigned directly to this agent
│   └── Class: a reusable role for new agents
│       ├── Instructions (AGENTS.md)
│       └── Skills for that role
├── Live work
│   ├── Provider session and conversation
│   ├── Terminal and Chat: two views of that work
│   └── Conversation archive, when enabled
└── Result and observation
    ├── Status, current activity, and telemetry
    ├── Inbox items for finished work, updates, approvals, or actions
    └── Roster, teams, watchlists, and Workbench tabs that refer to the agent

Some Wardian objects compose with an agent without becoming part of it:

Prompt          → a saved message sent to one or more agents
Automation        → a reusable process that can coordinate agent work
Team/watchlist  → a working set of agents for people to monitor or target
Inbox/evidence  → records the outcome of agent or automation work
Workbench tab   → one presentation of a running agent

Those objects are deliberately separate. A class changes the reusable setup for new agents; a prompt is an on-demand message; an automation is a process that may use agents. Closing a tab or removing an agent from a watchlist does not stop its running session.

Term	What it means
Agent	A named working instance with task context, selected reusable practice, and a provider session.
Provider and model	The coding tool the agent runs, plus the model and provider-specific choices it uses.
Class	A reusable role for new agents, with shared instructions and skills. It does not automatically change existing agents.
Prompt	A saved message you choose to send when the same request comes up again. It is not permanent agent setup.
Skill	Reusable guidance or a procedure. A skill can be scoped to your profile, a class, or one agent.
Workspace	The main project folder and its local context. It can have additional readable folders.
Worktree	An optional separate Git checkout, usually used when an agent needs its own branch. Moving an agent to one starts a fresh provider session.
Session	The agent's current provider process and conversation. Closing its tab does not stop the session.
Terminal and Chat	Two ways to view the same agent: the provider's live terminal or Wardian's structured messages and tool activity.
Telemetry	Live health and activity information such as status, query count, uptime, CPU, and memory.

Tailorability Slope

By tailorability, we mean how much you can adapt Wardian to the way you work. A genuine slope trades a little more learning and effort for a little more control at each step. You can stop anywhere, use something another person has made, or move back down when a simpler option is enough.

More tailoring power
      │                                      Build a new custom tool
      │                                      or integration (advanced)
      │                                Compose an automation blueprint
      │                          Define a reusable role: Class
      │                    Write a reusable procedure: Skill
      │              Save a repeated request: Prompt
      │       Adjust an existing agent's workspace, model, or setup
      │ Use an existing agent, project, and built-in tools
      └──────────────────────────────────────────────────────────────
        Less learning and effort             More learning and effort

The point is not to turn every task into an automation or every user into a programmer. Use the smallest adaptation that solves the problem. Move upward only when it cannot express the change you need.

When you need	Smallest useful move	What becomes reusable
A task that fits an existing setup	Start an agent with the right workspace, class, and provider	Nothing yet; use what exists.
A small task-specific adjustment	Change the agent's workspace, model, or direct setup	The agent's saved setup.
A request you keep rewriting	Prompt	A message you can send again without making it permanent setup.
A procedure that should be followed consistently	Skill	Guidance that can be deployed to one agent, a class, or your user profile.
A role with a stable way of working	Class	Shared instructions and skills for new agents in that role.
A process with connected steps	Automation blueprint	A design that can produce separate automation runs.
A need none of those pieces can express	A custom tool or integration	A new capability; this is an advanced, higher-effort step.

Project files, instructions, results, and evidence stay alongside this slope. They give each step the local context it needs, rather than becoming another kind of reusable setup.

Reusable Work

The Library keeps the items you may want to use again. They serve different purposes:

Library
├── Prompts       → saved messages you send to selected agents
├── Skills        → reusable guidance deployed to an agent or class
├── Classes       → reusable starting setups for new agents
└── Automations     → saved automation designs that can start automation runs

A prompt is for a request you want to send again. A skill is for instructions or a procedure an agent should have available while it works. A class combines general instructions and skills for a repeatable role. An automation coordinates work across steps or agents.

Automation Runs

Automation blueprint (saved design)
├── Trigger or manual launch
├── Run (one execution of that design)
│   └── Nodes (the individual steps)
└── Results and history in Automations and Inbox

Use an automation blueprint when a process has repeatable steps. A run is one specific time you execute it; changing a blueprint does not rewrite a run that has already started.

Where to Go Next

• UI Overview explains the Workbench and sidebars in detail.
• Agents explains monitoring and arranging agents.
• Library explains prompts, skills, classes, and automations.
• Automations explains building, launching, and monitoring automation runs.