AgentHubv0.4

Deploy

/hub:deploy has four modes: check, staging, production, and rollback. All HEAVY. All gated. All logged.

Modes

  • /hub:deploy check — pre-flight only. Lints, tests, env audit. No deploy.
  • /hub:deploy staging — runs check then deploys to staging.
  • /hub:deploy production — staging gate must have passed in this session.
  • /hub:deploy rollback — reverts to the previous release tag.

When NOT to use it

  • No CI wired yet. Use your platform's native deploy first; add AgentHub once the pipeline is boring.
  • You're just testing locally. Use /hub:preview.

Invocation

/hub:deploy staging
/hub:deploy production
/hub:deploy rollback

What pre-flight actually checks

  • Git: clean tree, branch is ahead of main only by expected commits, no unresolved merge conflicts.
  • Tests: all green on latest commit; coverage hasn't regressed.
  • Env: required variables present in the target environment's config.
  • Dependencies: no known CVEs (via vulnerability-scanner skill).
  • Migrations: any new schema change has a down path.

Example session

Shipping to staging then prod
user
/hub:deploy staging
gate
HEAVY · hub:devops-engineer · ~48k tokens · preflight + deploy-staging · [y/n/tweak]
user
y
hub:devops-engineer
Preflight: 124 tests green, coverage stable, no new CVEs. All env vars present for staging. Deploying…
hub:devops-engineer
Staging up at https://staging.example.com. Smoke checks pass. Release tagged v1.4.0-rc.1.
user
looks good. /hub:deploy production
gate
HEAVY · hub:devops-engineer · ~35k tokens · staging gate passed this session · [y/n/tweak]
user
y
hub:devops-engineer
Deploying v1.4.0. Rolling update, 3 instances at a time. Post-deploy /healthz and error-rate watch for 5 min. Green. Release v1.4.0 live.

Rollback

/hub:deploy rollbackreverts to the previous release tag, pauses, and asks for confirmation before flipping traffic. Use when a post-deploy alert fires — it's faster than fixing forward on a hot path.

Pro tips

  • Always run check before staging if you haven't pushed in a while — it surfaces stale CI.
  • Use tags for rollback. If your deploys aren't tagged, rollback has nothing to target.
  • Don't ship from a dirty tree. The pre-flight refuses anyway, but fix the habit.

Next steps