A Hacker News mention inbox your team actually works through

Build me a Hacker News mention inbox that our founder, growth lead and product manager open every day instead of scrolling the site hoping to catch something. I keep a watch list of keywords in the app: our product name, our main competitors, and the category terms we care about. The app pulls every matching story and comment into a single triage queue that a person opens and works through. There is no schedule and no trigger; a Refresh mentions button in the app runs the fetch.

Fetching the mentions. Use the Hacker News Search Latest Posts (Algolia, by date) operation for the freshest hits since the last refresh, and Search Posts (Algolia) for the ranked backfill when I add a new keyword or set the inbox up for the first time. For each hit, use Get Item to load the parent story and the surrounding thread so I can read the mention in its context, and Get User to pull the commenter's karma and account age so I can judge who is talking. Match keywords case insensitively against titles and comment text, and key everything on the Hacker News item id so refreshing never creates duplicate rows.

The inbox view. Each row shows the keyword it matched, whether the hit is a story or a comment, the parent story title with its points and comment count, an excerpt with the keyword shown in context, the author with karma and account age, how long ago it was posted, and a permalink to the thread on Hacker News. Rank rows by thread points and comment velocity so a front page mention outranks a dead comment in an old thread. Collapse repeat hits on the same thread into one row with a count I can expand, so one popular story does not flood the queue.

Triage state is the whole point. Every mention carries a state I set: new, needs reply, handled or ignored. Persist the mentions and the states in the General Input Database, with state stored per user so each person can work the queue without overwriting anyone else. Default the view to new, and give me filter tabs by state and by keyword. Let me mute a keyword that turns out to be noisy so its future matches stop appearing, without deleting the history I already collected.

Add a saved quotes view. From any mention I can highlight a line, save it with a note and a tag, and it lands in a quotes archive I can browse, filter and copy from. This becomes our voice of customer record, so keep the author, the thread link and the date on every saved quote.

Every row has a Read the thread and draft a response button that starts a background agent inside the app. The agent walks the full comment tree with Get Item and checks the commenter with Get User, then writes its output back into the app: a summary of what the objection or question actually is, a note on whether anyone has already answered it and how that answer landed, a suggested plain spoken reply I can post myself, and a short context brief on the thread and who is in it. Store the agent output against the mention in the General Input Database and show it in a detail panel with a copy button and a regenerate option, with a visible running state while the agent works.

Two actions close the loop. Share to Slack sends the mention, its excerpt, the permalink and the agent brief to a channel using the Slack Bot Send a Message operation, with a channel picker built from List Channels. File as issue creates a Linear issue for genuine bug reports or feature requests using Create Issue, putting the Hacker News permalink, the excerpt and the agent brief in the description, with a team picker built from List Teams. Record the resulting Linear issue URL and the Slack message link on the row so nothing gets filed or shared twice.

One hard constraint: Hacker News is read only and unauthenticated here, so the app never posts, votes or replies. The agent drafts and a human posts from their own account. Keep this Hacker News only rather than a multi source listening tool.