Turn Read AI meeting action items into Linear issues

Build me an agent workflow that turns engineering-flavored action items from Read AI meetings into Linear issues, so bugs and feature requests captured on calls actually reach our backlog instead of getting lost between the meeting and the next standup.

Trigger: a Read AI webhook on the meeting_end event. The webhook payload includes the meeting id. As soon as we receive it, run the rest of the workflow.

Step 1. Fetch the meeting from Read AI. Use the Get Meeting operation with expand set to action_items, summary, and transcript. The action items each include the text and speaker context, and the transcript expansion gives us the surrounding quote we will paste into the Linear issue.

Step 2. For each action item, classify it as either engineering work or not engineering. Engineering work means bugs, feature requests, technical follow-ups, integration questions, infra or platform tasks, and anything that would reasonably live on the engineering backlog. Not engineering means sales follow-ups, scheduling, ops chores, customer success outreach, and similar non-technical tasks. Drop anything that is vague enough that you cannot tell what someone is supposed to do.

Step 3. For every action item you classified as engineering, call Linear Create Issue with the following shape. The title should be a short rewrite of the ask in imperative voice, for example 'Fix CSV export truncating on 10k rows' rather than the raw transcript line. The description should include a one or two sentence summary, then a markdown blockquote containing the relevant transcript excerpt (the speaker name and the lines around the action item), then a link back to the Read AI meeting so anyone can click through and listen. Pick a sensible team. If you can tell from the topic which team it belongs to, route it there; otherwise default to the main engineering team that I will configure on the workflow. Infer labels from the topic. Always apply a 'from-meeting' label so we can find them later, and add 'bug' if it sounds like a defect or 'customer-request' if the ask came from a customer on the call.

Step 4. After all the issues are created, post a single Slack message to our engineering channel using Slack Bot Send a Message. The message should name the meeting (and link to it on Read AI), then list each new Linear issue with its title and Linear URL. Keep it short. If zero engineering action items were found, do not post anything.

Make the Linear team, the Slack channel, and the default labels configurable so I can change them without editing the agent instructions. The point of this workflow is simple: no more 'we should fix that' lines getting lost between the call and the next standup.