[ahegyes]

Summary. The two review commits of #66, which merged before they were pushed:

• Migration runner: the scheduled hook {option}/migration/run is longer than the lock {option}_lock through which the constructor validated the option, so an option of 178 to 186 bytes passed construction and threw only when a batched step first scheduled its continuation, leaving the migration stuck. The runner builds its ScheduledHook in the constructor, which refuses a hook name over 191 bytes there, and schedules that instance.
• Wording:
  • AGENTS.md says that a hook built from an id or option name the consumer passes takes it whole as its first level, which covers {option}/migration/run;
  • the updater's hook docblock loses a semicolon;
  • the consumer request's record key initialized becomes boot_succeeded, after the hook it counts.

Production code +18 −8 (2 files), tests +13 −5 (6 files), docs +1 −1 (AGENTS.md).

Testing.

• The tree equals the one Name every hook by the slash grammar and move to GPL-3.0-or-later #66's final gate ran on: composer quality-check with Unit 192 and Integration 1,346, composer test:integration:multisite (41) and composer test:integration:php74 (2).
• Before the change: the runner's new row, a 177-byte option accepted and a 178-byte one refused, failed until the constructor built the hook.