Guides
Task-shaped walkthroughs for daily use. Each guide states what you need before you start, the steps to take, the result you should see, and where to go next. Use Concepts when you need the mental model and Reference when you need exact behavior.
Set up your team
| Task | Guide |
|---|---|
| Create agents, then bind them to a project workspace | Create a workspace and agents |
| Change members or the leader | Edit members and the leader |
Add the generated AGENTS.md block and .ensemble metadata | Initialize a workspace |
Work day to day
| Task | Guide |
|---|---|
Send a group or direct message, use $Name, or reference a file with @path | Route a group or direct message |
| Assign and invoke reusable guidance | Use a skill |
| Focus and page through work, then delegate, reassign, retry, restart, cancel, or open a native session | Use the Work pane |
| Resolve the sticky summary or the full approvals inbox | Resolve an approval |
Configure and protect
| Task | Guide |
|---|---|
| Set identity, main and backup models, Skillbox, and save behavior | Configure an agent |
| Add, edit, delete, or promote saved knowledge | Manage memory |
| Choose a recovery backup or data export, then restore or roll back | Back up or restore |
Operator procedures
These existing pages cover recovery and upgrade paths:
| Task | Guide |
|---|---|
| Back up or restore durable state | Backup and restore |
| Upgrade from a prerelease | Upgrade |
| Reverse a package or state upgrade | Package rollback |
| Recover a broken workspace or stuck work | Incident runbook |
| Find a key by context | Keyboard reference |
| Understand the leader, work, and budgets before delegating | Concepts |
Recipes
Recipes apply Ensemble to a real workflow. Each recipe states the goal, the actors, example prompts, what to watch in the Work pane, expected output, and pitfalls. Start with the recipe index, or go straight to a workflow:
| Workflow | Recipe |
|---|---|
| Split one research question across two agents and combine the findings | Run parallel research |
| Review one change through separate correctness, security, and test lenses | Run a multi-agent code review |
| Turn an issue into a scoped implementation ticket | Turn an issue into a ticket |
| Use one agent's skill without leaking it to the team | Use an agent-scoped skill |
Writing a guide
Follow the page template. Lead with the task, keep one action per step, show expected output where it matters, and end with the next guide. Put architecture decisions and verification records behind a Go deeper link, not in the steps.