Keep the current Linear cycle under control from chat

The agent works with Linear through a single GraphQL endpoint, api.linear.app/graphql: issues, projects, teams, cycles, labels, comments, workflow states and members all live there. A query lists the fields it needs explicitly, so the agent goes after the state of a cycle in one call instead of assembling the picture from a dozen exports: what is not moving, what is overdue, who on the team is carrying the most.

An important trait of Linear is that workflow states belong to teams rather than to the workspace: every team has its own set of states with its own names, and identically labelled columns in two teams remain two different states. So the agent first works out which team an issue belongs to and only then picks a state from that team set, rather than moving the task into a neighbouring team state because the names match.

In writing, the agent files an issue in the right team and project, sets labels, priority and workflow state, posts comments and archives issues — but only if the key carries the Write permission. A key without it reads and counts, and the agent says so up front rather than quietly running into a refusal. After a change it re-reads the issue, so it shows the actual state instead of the request it sent.

What the agent can see is bounded by the key itself: when access was narrowed to specific teams at issue time, the rest of the workspace stays invisible and the answers cover those teams only. Linear returns long selections in cursor-paged chunks, so an issue list is read through to the end rather than cut off at the first hundred. The agent holds the request rate itself, learns about new issues and comments from webhooks, and when permissions fall short it says so directly instead of showing an empty table.