Figma Specs 與 Figma agent 如何將上下文轉化為實作計劃
Workflow lab: Turning context into a plan with Specs and the Figma agent
Figma 推出 Specs 開放測試版,讓團隊在畫布上集中整理產品需求、設計與工程上下文,並由 Figma agent 生成供編碼智能體使用的實作計劃。
Welcome to Workflow lab, where we present a sample workflow
Workflow lab: Test ideas with code layers and the Figma agent
using Figma products and tools.
When teams are going from prototype to production faster than ever, it’s easy for important context to get lost along the way. Written requirements, design details, and engineering guidance are critical to making sure teams are sprinting toward the right outcomes, but these inputs often end up scattered across separate documents and links.
Now, Specs
Introducing Specs: Markdown as a medium
We’re launching Specs, a container for your context that lives on the Figma canvas.
provides a shared container for context right on the canvas, alongside the work it describes, where teammates and agents can contribute together. In this workflow example, an engineer uses Specs and the Figma agent From making designs more expressive and automating busywork, to exploring more possibilities on the canvas, here are six ways to push designers further with the Figma agent.
6 easy ways to use the Figma agent right now
The problem
The team at Pickup, a fictional marketplace shopping app, is scoping a new delivery feature. The lead engineer is taking stock of the designs and technical decisions they’ll need to build it: There’s a PRD, a diagram in FigJam, and a Figma file containing designs, code layers
Design together with code layers on the canvas
, and components on the canvas. Across all of these sources, it’s not clear which details are missing or who still needs to give input. The engineer needs to connect the dots.
A code layer is an interactive prototype on the canvas backed by running code.
Starting with the data
The engineer combs through the PRD and notices that the team plans to pilot the new delivery feature in a single city, but haven’t identified which one yet. The engineer doesn’t want to start implementing code until they know which city’s map they’re building for, so they decide to look into where Pickup’s demand is highest based on the app’s usage data. That data lives in Hex, the team’s data analytics platform, but the engineer wants designers and other teammates to be able to view and react to the data in a shared space.
Connectors let the agent access information from a tool outside Figma.
The engineer turns to the Figma agent to help add the data to the canvas in the form of a spec. They open the agent, add the Hex connector to the conversation, and prompt: Go find where we have the most usage of Pickup and create a spec breaking that down. Also, update this map code layer to show usage data for the top cities.

The agent produces a spec that contains a table with app usage and order volume by city, along with code layers showing maps for each of the top four cities. The engineer and their collaborators review the Hex data and the screens on the canvas. The team decides to launch in Bengaluru, India, the city that clearly has the highest density of customers and demand.

Gathering context on the canvas
The engineer opens the team’s technical spec. It explains how the app will store information about deliveries, how the team will make the feature available to customers, and how they’ll monitor it for problems after launch. They then drag in an existing FigJam diagram showing the delivery process and what the seller, buyer, and driver each do.

Elsewhere on the canvas, designers are reviewing the final screens. The team has added annotations on the frames to call out key details, and both design and engineering approved the designs.

The engineer also includes an interactive version of the Bengaluru map as a code layer so the agent can see how it should work—for example, how nearby items group together and how they animate as someone moves around the map.
Lastly, the engineer reviews a new component from the design systems team. A code layer shows how the component works, and a written spec beside it explains when to use it instead of the existing component.
The engineer selects the technical spec, annotated design frames, interactive map code layer, new component, and design system guidance on the canvas, then opens the Figma agent. They prompt the team’s previously configured build-implementation-plan skill—a reusable set of instructions for the agent. The agent creates a new spec that brings those inputs together into a plan for a coding agent to build the feature, including how it should look and behave.

Reviewing the plan and handing it off
Before handing the plan off to a coding agent, the engineer reviews it to make sure the details are right, expanding the spec to its full width. They check that the spec includes the team’s design system, testing and observability requirements, rollout details, and a table outlining what needs to be done to build and ship the feature, along with references back to the specs, designs, and code layers it draws from.

The engineer spots a detail in the spec that isn’t quite right, leaves a comment on the spec with the problem, then prompts the Figma agent: Read my comment and reply to it with a solution. The agent comments back with an edit for the engineer to review. Reading on, the engineer notices another detail that needs adjustment, but this one requires input from a designer on the team, who the engineer tags in a comment. The designer receives a notification, jumps into the spec, and makes the change in real time.

The path to production
Clicking on the spec, the engineer copies a link to share it and opens Codex. They paste the link and prompt: Implement this feature. Through Figma’s MCP
The TL;DR on MCP: Why context matters and how to put it to work
, Codex can read not just the spec, but also the annotated design frames, code layers, design system components, and other context that went into the plan—all with just one link. All of that context helps Codex build the delivery feature, and once implemented, the team can review and test the feature before rolling it out to users.
Specs is in open beta on all plans and available on web, desktop, and mobile. Get started here, and learn more in the help center.
來源:Figma AI / Agent / Make · figma.com

