migrate

> **Warning: this skill rewrites files in `meta/` and `.claude/accelerator*.md`.** Recovery is via VCS revert. Before running, ensure your repo is committed and you understand what each pending migration does. The safety guards (clean-tree check, preview) exist to give you a moment to stop — they are not a substitute for understanding the changes. ## When to invoke Run `/accelerator:migrate` after upgrading the Accelerator plugin to a version that bundles new migrations. The SessionStart hook will tell you when this is needed: ``` [accelerator] .accelerator/state/migrations-applied is behind the plugin (highest applied: 0001-rename-tickets-to-work; highest available: 0002-...). Run /accelerator:migrate to bring it up to date. ``` You can also run it proactively — if no migrations are pending, it prints `No pending migrations.` and exits cleanly. **Upgrade sequence.** After pulling a new plugin version, run `/accelerator:migrate` before invoking any skill that reads or writes paths affected by pending migrations. Skills do not gate themselves on pending migrations; the SessionStart hook only warns when migrations are pending. Running write-side skills (e.g., `/accelerator:research-codebase`) between the plugin upgrade and the migration may produce results written to or read from new-default paths that do not yet exist on disk. ## How it works `accelerator migrate` (the `accelerator-migrate` sub-binary) runs the migration lifecycle in-process — there is no forked child p...