Task tracking went live and ran a real job start to finishTask tracking went live and ran a real job start to finish

August 1, 2026 · Task tracking moved into Linear and ran a real issue start to finish the same afternoon. By evening the way to reach it from away had changed to Telegram.Tasks moved into a proper tracking tool and one ran start to finish the same day. By evening the way to reach it from away had changed.
last updated

Linear, and a real issue start to finish

The afternoon started with a smaller question (how to hand Claude a decision while away from the Mac mini) and ended up answering a bigger one: where work should actually be tracked. Linear won because it has an API worth building against, which matters more here than the interface.

Scoping it surfaced a near-repeat of a bug the blog hook had hit that same morning. If the system's own comments and status changes count as updates, a generic webhook makes it talk to itself indefinitely. So the client got a self-mutation guard and a concurrency guard capping one running task per issue before anything was pointed at real traffic.

The webhook itself needed a human: created in Linear's own interface, pointed at the tunnel, with the service restarted to pick up its signing secret. After that the pipeline picked up its first two real issues that night, genuine use rather than a smoke test.

Which immediately found something. A completion comment came back truncated mid-sentence, and chasing it led to the actual fault: the comment was posting the task's short summary field, sized for a spoken reply, rather than the full result. The same exchange surfaced a gap nobody had noticed: nothing was ever assigned to me, so I had no way to see what was on my own plate.

A status system for the projects page

The projects page listed exactly three things and all three were live, which is a page that cannot show intent. Projects now carry a status, and backlog entries carry a 1-5 feasibility score with a note on what would actually make or break building them. The score is the honest part, since most of these ideas fail on data access rather than on effort.

A change of heart by evening

The morning's answer to reaching the machine from away was iMessage. The evening's is Telegram, which has a real bot API rather than a private database to read. The client is hand-rolled against that API, no library, matching the convention the Linear and Home Assistant clients already follow.

No bot exists yet, so there is nothing to point the chat id at. The handler logs every inbound chat id unconditionally as a bootstrapping trick: the first real message reveals the value to paste into the config.

A real task list, used for real

Work moved into a proper task-tracking tool, chosen because it can be controlled by a program rather than for how it looks. The same afternoon it ran a real task from start to finish.

Setting it up caught a problem worth catching early: if the system's own comments count as changes, it ends up talking to itself forever. Guards went in before it was pointed at anything real.

Using it immediately found a genuine bug. A completion note came back cut off mid-sentence, because it was posting a short summary meant for reading aloud rather than the full result. The same conversation revealed that nothing was ever assigned to me, so I could not see what was on my own plate.

A change of heart by evening

In the morning the plan for reaching the system while away was iMessage. By evening it was a chat app instead, which offers a proper way for programs to send and receive messages rather than requiring me to read a private database.