/push and /release — publish, then ship, as two deliberate steps
/push publishes the current branch and nothing else. /release
is the moment pending branches get reviewed and a version goes out. Neither
one does the other's job.
/push publishes the current branch and nothing else. /release
is the moment pending branches get reviewed and a version goes out. Neither
one does the other's job.
Until now, only a /review survey could land a branch on dev. That rule is
amended: a /work session may merge its own finished work onto integration — when
the operator answers its end-of-run question in-session, and the branch passes a
mechanical scoped-merge gate.
/tag answers one question — what's the next version, and can I mark it —
without doing anything a release actually requires. It never pushes, and it
is deliberately not a release.
/deploy publishes from a project's deploy branch (usually main). If dev
has commits not yet promoted there, deploying silently shipped whatever
main happened to have — even if newer, already-reviewed work was sitting
one merge away. /deploy now checks for that gap and asks first.
Every project accumulates the same small gaps: no robots.txt, no llms.txt,
a single stray favicon.ico instead of a real icon set, no dependency-update
bot. /optimize scans for the standards that actually apply to this
project's type, and turns the gaps into .plans you pick from — not a wall of
text you have to triage by hand.
Plans that span multiple sessions — especially the ones a human is
assigned to complete — had no glanceable way to show step-level progress
between sessions. The plan template now carries a ## Progress
checklist that /draft seeds and /work keeps in sync.
Every project already has a way to go live — a workflow file, a vercel.json, a
deploy.php, or just a production remote someone pushes to. /deploy
finds that path and runs it. It only proposes tooling when a project genuinely
has none.
Security holes die in chat transcripts. /audit turns an exhaustive code-
and-dependency pass into prioritized .plans bug plans so the fleet can
close them like any other ready work.
The best feature in your toolchain is worthless if nobody knows the command for it. A new doctrine rule closes that gap: before acting on a request, an Anchor agent checks whether a skill or slash-command already available in the session would do the job faster or more correctly — and if one clearly fits, it says so in a single line and offers to use it, then gets on with the work.
Agent completion is no longer a self-certify archive. When Done when holds,
/work always moves the plan to review-needed/. Humans archive with
/review.