{
  "version": 1,
  "paths": {
    "update.md": "#2994 (epic #1671 Phase 6.3, FINAL slice): fragmentizes update.md onto the marker grammar, admitting `state:next-channel` (channel-banner section) and a new dedicated `init.update` entry point (cmdInitUpdate). update.md previously carried NO `gsd_run query init.*` call at all; the new INIT_UPDATE fetch (reusing the already-resolved $GSD_TOOLS from get_installed_version, dual .cjs/PATH-shim invocation style, matching the pre-existing update-context call) plus its surrounding comment outweighs the single `<!-- gsd:section -->` marker-pair stub that replaces the extracted channel-banner prose (now living in gsd-core/workflows/update/steps/channel-banner.md). Net SOURCE growth is +703 bytes (26,024 -> 26,727). The pre-existing TAG=\"next\"/TAG=\"latest\" case-statement in parse_update_channel is deliberately left byte-identical — issue #815's regression test (tests/issue-815-update-next-channel.test.cjs) asserts that literal text stays in the workflow, since the npm dist-tag selection must run in the workflow's own shell before any gsd_run round-trip. The EMITTED artifact composeWorkflow produces at install time still includes the extracted prose verbatim (markers strip, gap+section bodies re-join byte-for-byte) when the atom is unresolved (section_manifest null -> read-everything fallback), so installed behavior is unchanged; only the SOURCE file's on-disk byte count moves."
  }
}