McRitchie Studio
Say Hi 👋
Agents Builders
← Back to Docs

agents/alex/sops/full-cycle.md

Full Cycle

Status: Active

This is Alex's full-cycle SOP. It runs review, QA release, and production
deploy with explicit ship authority.

Scope

full-cycle composes existing release atoms:

  1. Avi-style review-only PR review.
  2. Steffon's self-healing qa-release.
  3. Avi's production-deploy.

This SOP crosses the production gate. Use it only when Mr. McRitchie launched
full-cycle or otherwise granted production ship authority in this session.

Entry

Run this SOP from the McRitchie Studio primary checkout:

cd /Users/alex/projects/mcritchie-studio

Use the production board by default. Do not add --local.

Preconditions

There is work to move:

  • submitted PRs to review
  • reviewed work to prepare
  • an assembled release to ship

If the queue is empty and release == main, report "nothing to run" and stop.

Procedure

Refresh intake:

bin/qa-intake --refresh --apps mcritchie-studio,turf-monster,rolio

Run the three atoms in sequence:

# pr-review atom: review submitted PRs to reviewed or blocked
bin/release prepare --yes
bin/release ship --yes

Summon posture per atom (interactive tree visibility). When you drive
full-cycle interactively (a terminal is attached), summon each atom with the
posture its own SOP prescribes, so the crew shows up as a live Claude Code
sub-agent tree:

  1. Review — summon an Avi subagent via the Agent tool (subagent_type: avi) that SUPERVISES the pr-review procedure. Avi never reviews the code himself; he spawns the PRIMARY and LIGHT reviewers in parallel as sibling children, so the two experts NEST under the Avi node as a deeper branch.
  2. QA release — summon a Steffon subagent (subagent_type: steffon) that executes bin/release prepare --yes.
  3. Ship — DIRECT-DRIVE bin/release ship --yes from THIS orchestrating session. Do NOT wrap the ship in a subagent: it is the one irreversible gate, and a wrapper agent that dies mid-ship would orphan the release → main state with no terminal to recover it.
  • Caveat. bin/release is a script, so a summoned subagent is a thin driver around it — the tree node is real, but the script's subprocess internals live inside that agent. The tree is ephemeral (it dies with the session) and the autonomous heartbeat has no terminal, so it renders no tree. The durable, full-visibility surface is the Activities timeline, not the sub-agent tree.

Keep the same guards each atom owns:

  • Review in waves of five or fewer reviewers.
  • Use Steffon's qa-release sweep for merge plus QA.
  • Ship only after QA-green.
  • Do not deploy production unless ship authority is explicit.

Review and preparation take time; new PRs may land in submitted meanwhile.
After the sweep, re-check bin/task list --stage submitted and run one
straggler round — review anything new, then re-run bin/release prepare --yes
so the same candidate picks up the round-2 reviewed tasks and refreshes QA —
before shipping.

Cold start note: launched bare, you are the conductor (Deploy lane), not a
feature agent — do not follow the build gate's task/worktree flow. You
orchestrate review, sweep, and ship on work that is already built; you do not
review diffs yourself, create tasks, or write feature code.

For a single-task expedite, run Avi's
deploy-with-task SOP instead of pushing
one task past pending release work.

Exit Seam

The release is shipped, or the run cleanly reports no work. Report:

  • reviewed and blocked tasks
  • QA URL
  • production SHA
  • release slug
  • smoke result when a ship happens

Related