Running Claude Code on Linear tickets

Two different “on a Linear ticket”

One is local. You paste the issue into Claude Code, or you point the session at whatever tools you have installed, and you are the reviewer when it stops to ask. The session ends when the machine sleeps. The key is on that machine. A teammate cannot answer the prompt.

One is Linear’s own cloud agent. Its docs say it uses Claude Code or Codex, writes and tests in a cloud environment, and hands the team a draft pull request. Usage draws on the workspace’s AI credits. That path is real, and it is compared on its own page. Copilot can also be assigned on the same issue.

The third is the one this note walks through. The issue starts an AOSplane session. Claude Code runs in a sandbox on the Anthropic account you connected. Your team is not the first reviewer.

What you connect

1. An Anthropic account, or Bedrock, Claude Platform on AWS, or another compatible provider. Codex is the same shape on an OpenAI account, if that is the harness you want. The gate does not change.
2. The GitHub repository, linked to the project.
3. Linear, and an agent whose harness is Claude Code. CLAUDE.md in the repo still applies.
4. At least one QA check on the repository, if you want Plan first. Without it, the console shows Plan first switched off.

From the issue to the pull request

Start a session from the issue, or hand the issue to the AOSplane agent inside Linear. It answers on the issue. A reply there continues the session. The code does not land because of that reply. Agents cannot push.

With Plan first, the product manager writes a spec from what it can read, and you approve it before Claude Code writes code. Review then reads the diff. QA runs the repository’s commands. Exit code 0 is the pass. The draft pull request opens after they pass, with the spec, the review notes, QA’s results, and the cost of the task. A person merges.

The issue should name the outcome and the boundary. A title like “search is broken” spends the run on guessing. Linear’s own writing guidance says the same thing, and it is good advice on either product. The setup steps that are specific to AOSplane’s Linear connection are on the Linear guide.

The cap before it starts

The run reserves budget against the organization, the project, and the agent. The tightest limit is a hard cap on this ticket’s task. Over it, the task does not start. That is independent of Claude Code’s own session flag and of a workspace credit balance. Both of those are worth setting. They are written up in how to set a spend limit on Claude Code agents.

When the laptop is the right tool

• You are the only reviewer, and you want to steer every step.
• The change is exploratory and you do not want a spec in the way.
• The repository cannot leave the machine. A cloud sandbox is the wrong place for it.

When the ticket is real work for a team, the cloud gate is the one that keeps a teammate from being the first reviewer. Claude Code for teams is that case. Jira is the same gate from the other tracker: Jira tickets.