Coding agents … are good at coding. They know bash like the back of their non-existant hand. They know a million programming languages. They can open and read an entire manual without complaining in seconds. No human can do what an AI can do.

Coding agents make incomprehensible mistakes though. I asked it to research GPU performance and it gives me a useless plot. I use it to build a web app and every other run I’m asking it to fix what it previously broke. It soon became hard to see the higher picture like I am suppose to when I’m wacking down mis-aligned commits after every run.

I think the AI just lacks context. Context about our project goals, current feature, and its current task.

Billion dollar companies now exist just for building the correct human agent interface. Thus, we got our VS Code forks, then cli agents, now agent orchestrators. I think the next interface is a project management board. In that case, let’s bootstrap our workflow to be driven from the gold standard of project management software, linear.

linear provides an mcp to interact their app. However, it is clearly not built for agents and my gosh vibe coding takes a lot of tokens already. This project is building a leaner cli that wraps the mcp server so agents stop wasting 20% of its context wrangling linear and possibly messing up.

This project is an axi with inspiration from gh-axi.

linear-axi is designed with an agent’s usage pattern in mind. It will store the linear project we are working on. Then base the issues, milestones, or documents displayed on our current project. We’ll also try filter and sort the most relevant issues.

Demo

Installation

Installing the Agent SKill:

npx skills add AnonymousMorris/linear-axi -g

Installing the cli:

npm install -g linear-axi

Links

npm package: https://www.npmjs.com/package/linear-axi

github: https://github.com/AnonymousMorris/linear-axi