Lectures

Tommy gets hung up on the fact that he has no log: there was a good version, but where it was — he didn’t write it down. And then he gets an idea: let requests be written automatically, and let the drudgery be handled by an invisible crew of “it runs itself.”

“Just make it happen by itself” is not a spec. Sam chops the dream into boring pieces, while Alex gives the hook spec, HOOK_SPEC, as three fields: event, action, and where to write. Tommy puts together a decision log for one project.

Alex breaks down event-driven mechanics in plain terms: event → action, handler, three controls (subscription, switch, filter). Code that does not run in order, but waits for events. A shift from “I do it by hand” to “I configure the system.”

The agent’s event mechanism is called hook. Alex shows how to attach a hook in words: the requests.log journal writes itself, and the computer says “Done” out loud. And there is a hard boundary: a hook is an external trace, not the agent’s memory.

Tommy assembles his hook from scratch and on his own — from a blank setup to a live log. He catches the main point: you can forget — and nothing should get lost. And in the chat with “Tom2036” a warning drops in about the price of “set it and forget it” — but Tommy only snatches the pleasant half.

Two level articles in one list: how to read a request log without mistaking a submitted prompt for a recorded decision, and what to do when a hook stays silent, fires twice, or gets in the way.