Module assess 

Classify a target repository before anything lands.

The assessment is read-only evidence plus one classification computed from it by an explicit rule: greenfield when the target carries no release mechanism and no release history, brownfield when a release mechanism is already in place — a tool’s configuration, a landed destination, a landed block — and needs-decision when the target shows release activity that no recognized mechanism explains: tags with no tool behind them, or a second long-lived branch. The rule lives here so a routing skill reads a verdict it can cite instead of judging “some release setup” by feel. The gathering spawns git and reads the disk; the rule itself is pure and unit-tested.

Structs

Evidence

The evidence the classification is computed from.

Facts

The repository’s facts alone, with no record read: what the planner gathers beside its own read of the record.

Landing

The landing record’s presence, the one fact rk status owns that the routing needs before it reads the full report.

Enums

Classification

What the target is, for routing: the corpus verdict, computed from the repository’s evidence alone.

Constants

LONG_LIVED_BRANCHES

Branch names that conventionally outlive a topic.

RELEASE_MARKERS

Files that mark a release mechanism, whichever tool owns it.

Functions

classify

Compute the verdict from the evidence. Pure, so the rule is testable without a repository.

gather

Gather the evidence at target, reading and never writing.

gather_facts

Gather the repository’s facts at target, the record aside.

long_lived_among

The long-lived branch names among refs, given as full ref names.