Intake

The formal path — brainstorm, spec, epic, tasks — is the preferred on-ramp, but most work does not arrive that way. It arrives as a dog-walk epiphany, a "we should probably…", an "oh, don't forget." Intake is the lossless net: it turns raw input into a tracked issue with near-zero friction, so nothing is lost before it can be planned.

Capturing the seed of an idea when it is fresh is more valuable than planning it out on the spot. Intake optimizes for that.

Three queues

Intake has three shapes, each a label, and — like everything strategic — all of them live in the org's .github, so the whole org-wide queue is one filtered view rather than something scattered across repositories:

Kind	What it captures	Graduates into
triage	A problem or bug not yet understood — needs diagnosis	an epic, or a task
idea	A spark — "what if we did this?"	a feature/epic
research	An investigation meant to produce a reproducible result	an epic with tooling + a report

The three are genuinely different intake shapes, not severity levels. In particular, research is not ad-hoc work: a result worth having is a result worth reproducing, so a research thread spawns automated tooling and pull requests and becomes a proper finite epic — never something done by hand and thrown away.

The lifecycle

1. Capture. The moment something is worth remembering, it becomes an intake issue in .github with the right --kind — a single, fast, correctly labelled record. No planning, no scoping; that is deliberately deferred.
2. Groom. On a periodic pass, each intake item is reviewed and dispositioned: promoted into a finite epic, filed as a task under an existing epic, dropped onto a repository's ad-hoc epic, or closed as obsolete.
3. Promote. An item that graduates leaves the queue and enters the formal model; the queue is meant to stay drained.

Intake is intentionally exempt from the every-task-needs-an-epic rule until it is groomed — that exemption is the whole point of a lossless net. Capture first; impose structure later.