Product · Use case
Scope the feature that names what it quietly breaks.
An agent has to scope a new feature. Most write a clean PRD — blind to the systems it touches, who owns them, and which of your last specs shipped clean versus slipped and why.
The usual way
A clean PRD, written in a vacuum.It breaks on contact
What it actually touches
The dependency graph the PRD always forgets.
The search indexing job is the one that slipped last time. Naming it now is the difference between a clean ship and a three-week tail.
Dependencies + owners from GitHubWhat's already proven
Your own track record says how this goes.
Shipped clean · CSV export
Scoped the auth dependency up front, named the owner, shipped on the day.
Slipped 3 weeks · Search filters
Ignored the indexing job — the same one this feature touches. It surfaced late and stalled the release.
Ship history from GitHub
12 tickets in this area, from ZendeskHow your team actually runs it
In the spec shape your team actually ships.
ADR format + decisions in NotionHugin already held all of it — the dependencies, the owners, what shipped clean, what slipped, and your format — connected.
Then the agent does the work
So the agent scopes it like your sharpest PM.
Hugin supplied the context — the dependencies, the owners, the track record, your format. The agent wrote the spec.
More tasks Hugin makes better