Turn LiveChat transcripts into weekly Linear product issues

Every Monday at 9am, mine the previous week of LiveChat conversations for recurring product problems, file the real ones as engineering work in Linear, and post a ranked summary in Slack.

Trigger: cron, weekly, Mondays at 9am in the org's local timezone. LiveChat has no polling trigger on this platform, so the date range is handled inside the workflow rather than by the trigger.

Step 1. Pull the transcripts. Use LiveChat's List Archives to search archived chats from the previous seven days (last Monday at 00:00 up to this Monday at 00:00). List Archives pages, so read next_page_id from each response and pass it back as page_id on the next call, and keep going until there is no next_page_id. Do not read only the first page, a partial week silently under-counts every theme. Where the archive result does not carry the full message history, call Get Chat to fetch the chat and its thread, and use List Threads for any chat that spans more than one thread. Read what the customer actually wrote and ignore agent replies, canned greetings, and system events.

Step 2. Group the customer messages into themes by the underlying problem being described, not by shared keywords. Count the distinct chats supporting each theme, so one chat that mentions the same problem five times still counts once. Drop every theme supported by fewer than 3 separate chats. That threshold is the guard against noise, and I would rather see three solid themes than fifteen speculative ones.

Step 3. For each qualifying theme, run Linear's Search Issues with the plainest wording of the problem before creating anything. Try a couple of phrasings rather than one, since an existing issue is unlikely to be titled exactly the way I would describe the theme, and consider both open and recently closed issues. If a matching issue already exists, do not file a duplicate: keep the existing issue and carry it into the Slack summary as already tracked. If nothing covers the theme, call Linear's List Teams to resolve the id of our product team, then use Create Issue against that team.

Each new issue should carry: a title naming the problem in plain language a product manager would recognize; a description with a short paragraph describing the problem as customers experience it; the number of separate chats affected; and two or three verbatim customer quotes with the date of each chat. Copy the quotes exactly as written, typos and all. Do not paraphrase them and do not tidy them up, they are the evidence.

Step 4. Work out which themes are growing. Run List Archives a second time over the seven days before the reporting week, paging the same way, and count how many chats in that earlier window match each qualifying theme. A theme with a materially higher count this week than last is growing and should be flagged.

Step 5. Post the summary to Slack with Send a Message in our #product channel. Rank the themes by number of affected chats, highest first. For each theme give the theme name, the chat count, a one-line description of the problem, and either a link to the Linear issue that was just created or a note that an existing issue already covers it. Call out every growing theme explicitly with both numbers, for example 12 chats this week versus 4 last week. Keep it factual and skimmable rather than editorial.

Edge cases: if no theme reaches 3 chats, skip Linear entirely and post a short note to Slack saying nothing cleared the bar this week, with the total number of chats reviewed, so the team still knows the job ran. If LiveChat returns a 429, back off and retry rather than dropping pages. If the product team cannot be found via List Teams, post the summary to Slack with the themes written out in full and say the issues were not filed, rather than filing them into the wrong team.