Fluent Commerce Logo
Docs

AI-Assisted Dashboard and Manifest Editing

Essential knowledgeCore knowledge to understand Fluent Commerce

Changed on:

19 Aug 2026

Overview

[Warning: empty required content area]

Key points

  • Describe a desired dashboard change in plain language, with enough detail for an assistant to draft a manifest change
  • Review a generated manifest draft for correctness without needing to read raw JSON fluently
  • Distinguish between draft changes ready to use as-is and those requiring developer follow-up
If you completed the UX Framework Knowledge Track, you've already been introduced to what a manifest is. In short, a manifest is a file that tells the dashboard what to show and where to show it. When you ask AI to implement a dashboard change, it will draft changes to this file.

Describing a Dashboard Change in Plain Language

The three things a good request specifies:
  • What to add, remove, or change (a field, a widget, a layout section, etc.)
  • Where it should appear on the dashboard
  • What data it should show
The more context and information you are able to provide the agent, the more likely it will be able to produce something you are able to use. Vague prompts will often cause them to make assumptions, which can increase the chances of error.
Example request structure
"Add a(n) `object` to `dashboard name`, that shows `information`, positioned `in location`."The request does not have to explicitly following the structure above, but it is recommended to add as much detail as possible. Some sample requests can include:
  • `Add a dashboard tile above the table showing the order count`
  • `Add a mystique component that shows a Google Map on a card. Assume the data input is a connection of orders. Show an icon for the fulfillment collection address of each order and zoom accordingly.`

Reviewing a Generated Manifest Draft

You are not expected to read raw JSON fluently. However, you should still verify that the generated results match what you're asking for. A manual review by a developer is always the most optimal, but there are ways to minimize errors when you are not one yourself.
1. Preview the rendered result
The most reliable check available to a non-technical reviewer is visually inspecting the component/change that you've asked for in a sandbox environment. 
2. Ask the assistant to translate its own draft back to plain language
You can ask for an explanation of what the agent has done, using a prompt such as "Before I approve this, explain in plain English what this manifest will show, where it will appear, and what data source each part pulls from." If the plain-language explanation doesn't match what was requested, you should review your prompt.
3. Ask targeted yes/no verification questions
Rather than eyeballing the structure, ask specific questions the assistant can answer directly: "Does this reference the order's status field, or something else?" / "Is this pulling from the fulfillment entity or the order entity?" 
4. Compare against a known-working example, side by side
If a similar dashboard element already exists and works correctly, ask the assistant to point out what's different between the new draft and that known-good example. Differences are much easier to spot in a comparison than to judge in isolation. You can also request that the agent verifies its changes against best practices by calling the Search MCP server.
5. Ask the assistant to flag its own assumptions
Prompt: "What assumptions did you make in generating this, and what would you be unsure about?" This can help identify where the draft is most likely to be wrong.
6. Look for odd values
Without any JSON literacy, you can still reasonably flag a draft that looks unusually short, has obviously placeholder-looking values, or repeats the same value in multiple spots where variety was expected.