Classify a bigname request or diff before coding. Use whenever work may affect public API semantics, coverage or exhaustiveness, shared IDs or enums, manifests, migrations, work ownership, `crates/domain`, replay behavior, upstream citations, or consumer replacement/parity.
User-invoked only bigname verification loop. Use only when the user explicitly invokes `$verify-loop` or asks to run the verify loop after Codex has implemented work; it spawns `verification_reviewer`, triages findings, confirms real issues with failing tests or checks, fixes them, and repeats until the reviewer is clean or all findings are rejected.
Scope or implement one end-to-end bigname consumer capability. Use when work maps a first-party or public capability to routes, projections, execution behavior, manifests, tests, rollout criteria, or any parity/replacement claim.
Plan or review bigname manifest authority, source-family admission, discovery edges, capability flags, proxy tracking, start-block provenance, watch-plan effects, or manifest-driven invalidation behavior.
Review bigname raw facts, normalized events, canonicality, projection rebuilds, reorg repair, invalidation, execution artifacts, migrations, and storage-family write ownership.
Gather pinned `.refs/` evidence for ENSv1, ENSv2, Basenames, admitted app metadata, reference-indexer comparisons, or reference execution-client comparisons. Use when docs, manifests, tests, comments, or task output make upstream behavior claims or when a change may introduce a divergence.