Notes, boards and cloud machines for app development.
Browser extensions built with a coding agent
A coding agent builds a browser extension from a brief specifying an action and its permissions. The built folder downloads for loading unpacked in a browser on a separate computer. The action can then be checked against the brief. Extension-store publishing is outside this workflow.
See plans and usage pricing
Browser actions and execution context
A first extension can focus on an action such as saving the current link, highlighting text or reading a value from a page. The brief identifies where the action runs and the browser permissions it needs. This provides a target for the coding agent and a behavior to test when the extension is loaded in a browser on a separate computer.
A kanban card holds the action, permissions and target page. A popup, a content script and a background script are different places extension code can run. Specifying the execution context keeps that decision in the brief alongside the behavior the extension is expected to provide.
Manifest permissions and implementation
Manifest permissions and exact behavior define the implementation task. An example is a toolbar button that saves the current page title and URL to a list. That gives the agent a manifest entry and function to write, and supplies a concrete action for review after the build is loaded.
The codespace runs Linux with a terminal and VS Code in the browser. Claude Code, Codex CLI, opencode, Cursor CLI, Antigravity and Qwen Code work there with supplied sign-ins or API keys. Build tooling needed by the extension project can be installed on the machine.
Unpacked builds in an external browser
This workflow does not run a browser session inside the codespace, and it does not publish to an extension store. The built extension folder downloads to a separate computer and loads unpacked from the browser’s extensions page. Testing there checks the action against the original brief.
The codespace is billed by the hour while it runs. A snapshot supports stopping the machine and returning later; snapshot storage is billed monthly in credits. Problems found in the loaded extension can be recorded on the board as the next task, with the original action and permissions still available for reference.
Questions about browser extensions built with a coding agent
- What should a first version of a browser extension include?
- One action and its target page can provide an initial scope, such as highlighting text or saving the current link. The brief identifies required permissions and whether the action belongs in a popup, content script or background script.
- How do I try the extension once it is built?
- Download the built extension folder from the codespace to your own computer and load it unpacked from your browser's extensions page. This workflow does not run a browser session inside the codespace, and it does not publish to an extension store.
- Which coding agents can write the extension?
- Claude Code, Codex CLI, opencode, Cursor CLI, Antigravity and Qwen Code are coding-agent options, alongside other installed agents. Agent access requires sign-ins or API keys.