What moved where (7.0)¶

7.0 cut the skills a user can type to seven front doors - /steer:setup, /steer:spec, /steer:work, /steer:audit, /steer:status, /steer:next, /steer:build - plus /steer:standards, the fallback for surfaces where no hook injects the rules.

Nothing was removed. Every skill still exists and still does the same work; it is now entered as a mode of the door that owns its area. The point is that you no longer have to know which of twenty-odd names to reach for - and in practice you rarely type any of them, because the model routes your plain-language ask to the right door itself.

The renames¶

Arguments carry through unchanged - /steer:adr accept 3 becomes /steer:spec adr accept 3, /steer:sync --check becomes /steer:setup sync --check.

What did not change¶

/steer:reference, /steer:report and /steer:loop also stopped being typable, but their invocation strings are the same. No front door absorbed them: the model reaches them from an always-on rule, so a mention in your CLAUDE.md or README is still correct as written. Ask for what you want ("load the polyrepo reference", "report this steer bug") rather than typing the name.

/steer:loop additionally requires the automation opt-in in policy/ before it will scaffold anything.

Migrating a repo¶

Run /steer:setup sync. Its ledger step rewrites the old invocations everywhere they were materialized - CLAUDE.md, README.md, the PR template, spec/tracker.md, feature intent.md files, mise.toml, policy/*.yml - and shows you the diff before writing. It leaves append-only records alone (spec/history/, ADRs, audit reports), where a past /steer:adopt is a true account of what was run.

The same sync re-copies the generated Copilot and .agents/skills/ surfaces, so a Copilot teammate picks up the regenerated .agents/skills/ tree in the same pass.