---
title: "Dashboard guide"
description: "Use the environment map, workflows, run evidence and working settings."
canonical: "https://agenttwin.si/docs/dashboard/"
---

# Dashboard guide

Use the environment map, workflows, run evidence and working settings.


## Environment

Start with **Connect project** and `npx agenttwin init`. Agent-analysis and sync instructions expand when needed. The environment map appears after discovery or explicit resource selection. It shows those resources with logos and coverage labels; no demonstration services are preselected.

Drag empty canvas space to pan, drag cards to move them, or use zoom, Reset view and Arrange. Search the selected-resource list to find a tool, inspect its coverage and remove it. Removal can be undone. Manual catalog selection is a declaration, not account access or provisioning.

Once connected, Environment shows the project name and resource map. **View all workflows** opens the complete searchable candidate inventory.

## Workflows

Review discovered entrypoints and dependencies. **Prepare workflow** creates a preparation task for your coding agent using the selected candidate. Package real application files and reviewed operation mappings with the CLI, then upload the resulting bundle to execute. Saved configurations store fixtures, faults and assertions separately from application files. Candidates with recorded evidence link to their run results.

## Runs

Run history contains loaded account results and any `run.json` files you explicitly import. Filter by repository, workflow or outcome; search by test name, revision or run ID. On phones, use **Select test** to switch results. Active executions and executions without a loaded receipt appear separately above the history; completed tests are not duplicated there.

- **Summary** identifies the workflow, test configuration, repository, tested revision, runner and command. It explains the result and lists each expected outcome beside its observed value. Failed checks appear first. Failure scenarios show whether an injected condition actually activated.
- **Recorded steps** shows observed service activity in sequence, including writes and activated failures. It is not a complete trace of every application step.
- **Run history** lists other tests of the same workflow and repository. Select a baseline to compare checks, revisions and final collection state. Different test configurations, starting state, source digests or fault settings are highlighted; a comparison does not establish the cause of a regression.
- **Technical details** contains execution output, state changes, runner metadata and the original receipt.

Use **Copy fix prompt** for a failed test or **Copy review prompt** for a passed test. The task includes the workflow, available source entrypoint, tested revision, checks, recorded activity and output. Give it to your coding agent in the original repository, review the fix, rerun the same test and sync the new receipt. Local CLI execution is the default; use a cloud machine only when needed. You can also preview or download the task, export a result, or export the selected comparison.

A passed result covers its recorded checks, not every behavior of the workflow. A failed check explains what differed; the underlying cause may still require inspecting source and logs. Missing repository or Git information is shown as **Not recorded**, not inferred. Imported files remain in the current browser session and are not uploaded. The API currently lists the latest 100 account jobs; Settings controls how many receipts load per refresh, and later refreshes fetch unloaded receipts.

## Account menu

Click the initials in the top-right corner to open account actions: Account & settings, View runs and Disconnect account. Connection status reflects the current tab. Escape or a click outside closes the menu. Disconnect is available only when an account is connected.

## Settings

| Setting | Effect |
| --- | --- |
| Workspace name | Changes the dashboard label and initials in this browser |
| Default execution limit | Suggested duration in coding-agent preparation tasks; each reviewed bundle keeps its own execution limit |
| Refresh automatically | Polls account discovery and active jobs while the tab is visible; manual Refresh workspace remains available |
| Compact lists | Reduces workflow and receipt row spacing |
| Reduce motion | Disables transitions and smooth scrolling; system preferences are also respected |
| Account connection | API URL, token visibility, Connect account, Refresh workspace and Disconnect |

Use **Save preferences** to persist changes or **Reset defaults** to restore browser preferences. Preferences are not shared team settings. Account tokens stay in memory and are cleared on reload or disconnect; they are not saved with preferences.

## Navigation and storage

The top navigation contains Home, Environment, Workflows, Runs and Settings. The logo returns to Home. Home uses the full-width artwork layout. Other views have a collapsible project sidebar; closing it leaves a full-width content column. Narrow screens use the top navigation and a short repository row. Documentation uses a book icon; infrastructure implementation labels do not occupy the sidebar.

Manual declarations, map layout and dashboard preferences persist in browser storage. Imported discovery and receipts stay in memory. Synced discovery and submitted run evidence persist on the account’s API store. Sync sends metadata; executing a reviewed bundle uploads the selected application files. No production OAuth session or production account is cloned.

[Open the dashboard](/dashboard/). The compatibility dashboard is available at [the API origin](https://agenttwin-api.212.56.40.125.sslip.io/dashboard/).

## Failure scenarios

In **Workflows → Advanced: saved configurations and failure scenarios → Edit configuration**, select healthy execution or **Timeout after write**, choose a service collection, and set the response delay. Apply the scenario to the draft, save or export it, and rebuild your application bundle before uploading. Previously reviewed bundles are not changed by this editor.

Timeout after write commits a write before delaying its response. Your application decides whether to retry. Inspect fault activation and final-state assertions in Runs to detect duplicate effects. Other fault types are not currently implemented.

Settings includes a results-per-refresh limit for larger workspaces. Account changes clear transient resource selections, receipts, prepared uploads and upload approval. Server quotas and organization access policies are not browser preferences.

## Guided workflow

Home shows the next action for your connected project, its detected resources, workflow candidates, recent results, active work and failing assertions. The workflow assistant opens from a compact dock at the bottom of Home. The artwork fades behind the workspace and carries “What should your agent test?”. Customize lets you choose which overview cards appear; those choices and your company/name preferences are saved in this browser. Workflows shows ten candidates per page and searches the complete inventory. Selecting a workflow opens a focused preparation guide with a task for your coding agent and the upload step. Saved JSON configurations and failure controls remain under **Advanced: saved configurations and failure scenarios**. Preparing application files still happens in your repository; dashboard chat cannot do it automatically.

## Repositories and chat

Sync each repository with the same AgentTwin account. The sidebar lists synced repositories and the selected repository’s workflow candidates; selecting a repository changes its environment, saved contracts and chat context. Repository names identify inventories, so use distinct names for separate products. Run history currently spans the account. The sidebar is hidden in Settings.

Open the chat dock to see its connection status. Connect your account in Settings before sending a message. Chat explains discovery metadata, saved contracts and selected run evidence; use Workflows to prepare and execute code.

## CLI tests and Git activity

Test reviewed bundled Node workflows through `npx agenttwin run --file path/to/agenttwin.world.json` from the original repository. This uses in-process HTTP twins by default. Review output and fixtures, then run `npx agenttwin sync` to send discovery, Git metadata and CLI results to the website. CLI execution uses your machine and is not a VM isolation boundary.

Select `--runtime hosted` or `--runtime e2b` explicitly for a workflow that needs a cloud machine. PostgreSQL and Redis worlds currently require cloud execution. There is no automatic VM provisioning for discovery, chat or dashboard viewing.

Home shows recorded test results, adapter activities and failed checks. Git activity shows up to 100 commits from 14 days, total repository commits, branch and HEAD. Changed workflow entrypoint paths are flagged since the previous sync. Runs retain the tested SHA and source digests; older passing tests do not prove a newer commit passes. This does not yet trace transitive changes across every workflow.

### Workflow review

Workflows separates discovered code paths from recorded test evidence. Filter by activity, or choose **Not tested**, **Needs attention**, **Retest changes**, or **Last test passed**. Each card shows its dependencies and preparation needs; expand **Source and setup** to inspect its entrypoint and open resource setup.

Test status only uses receipts with a matching workflow ID and recorded repository. Missing repository context cannot establish coverage. A changed entrypoint or newer Git revision marks a previously passing workflow for retesting. Counts reflect loaded results, not every historical test.

Choose **Prepare a test** to preview, copy or download a task for your coding agent. The task includes the latest recorded result when available. Review the generated files and assertions, run locally, then sync. Expand cloud execution only when the workflow requires a cloud machine.

### Time-of-day artwork

Home artwork spans the full width, extending behind the assistant and project banner before fading at the top of the workspace cards. Its height adjusts with the layout. It uses your browser’s local time: a bay sunrise from 5am–noon, sunny Golden Gate from noon–5pm, and the moonlit coast from 5pm–5am. It checks again when you return to the tab and once a minute while open. No location permission is needed. **Customize → Preview background** lets you inspect each scene; previews last for this tab session, and new visits use the automatic schedule.