Find the themes Help Scout tags miss and file them in Linear

Build me a recurring-issue explorer that our product manager opens every week to see what support is actually drowning in. Help Scout tags are hard to aggregate and its native reports will not correlate topics over time, so this app has to do both the counting and the theme finding itself.

Main view. A date range picker (default: the last 7 days) drives a handler that calls Help Scout Get Conversations Report for the selected start and end. Help Scout report endpoints require ISO 8601 timestamps and accept previousStart and previousEnd, so always pass the immediately preceding window of the same length and let the API return the period-over-period deltas directly. Render a chart of conversation volume across the range, and under it a table of the top tags showing this period's count, the previous period's count, the absolute change and the percent change, sorted so the fastest climbing themes sit at the top with a clear up or down indicator.

Drill-in. Clicking any tag row opens the conversations behind it via Help Scout List Conversations, filtered to that tag and the same date range, showing subject, customer, date, status and current tags, with pagination. Clicking a conversation opens the real customer thread via List Threads, rendered as customer and agent messages in order so the PM can read what people actually wrote.

Find themes. A button at the top of the themes panel kicks off a background agent because tag data alone is unreliable. The agent reads the last 30 days of conversations (List Conversations across the window, then List Threads on each to get the actual text) and clusters them into plain-language themes independent of whatever tags people happened to apply. Each theme carries a short name, a one-paragraph summary, the count of conversations in it, three or four verbatim customer quotes with the conversation each came from, and the list of source conversation ids. Write the result back into app storage so the themes panel renders next to the tag trend table, with the last run timestamp and a running state while the agent works.

File to Linear. Every theme row gets a File to Linear button. First show a team picker populated by Linear List Teams and let the user confirm which team the issue lands in (and edit the proposed title) before anything is created. Then a background agent calls Linear Create Issue in the chosen team with the theme name as the title and a description containing the theme summary, three or four verbatim customer quotes, links back to each source Help Scout conversation, and the volume figure plus the trend as the justification for working on it.

After the issue is created, the same agent tags the source conversations with "reported" using Help Scout Update Conversation Tags. That endpoint replaces the full tag set wholesale via PUT, so read each conversation's existing tags first and send the existing tags plus "reported". Never overwrite what is already there.

Persist the filed issue's identifier and URL against the theme in app storage and show it on the theme row, refreshing status with Linear Get Issue, so the PM can see at a glance what has already been escalated. Themes whose conversations already carry the reported tag should be visibly marked as filed with their File button disabled, so the same complaint never gets filed twice.

Design notes. This is a weekly ritual for a single product manager, so favor one dense screen (volume chart, tag trend table, themes panel) with drill-in as a detail pane rather than a multi-step wizard. Help Scout allows 400 requests per minute per account and writes count double, so throttle bulk thread reads and handle 429 responses with the Retry-After header. Be resilient to conversations with no quotable customer text, and keep the tag trend table useful even before the first theme run has completed.