First Lead-run QA pass: image review

The current public lmctl model is teamfile + members + lmctl chat. A repeatable workflow is the prompt you give the Lead, plus any durable project facts committed under durable-memory/.

Create and seed a team

Use a .lmctl teamfile with a Lead, a QA member, and optionally a Reviewer:

_MEMBER_ alias=Lead     provider=codex
_MEMBER_ alias=QA       provider=claude
_MEMBER_ alias=Reviewer provider=agy

Then lint and seed it:

lmctl lint ./team.lmctl
lmctl seed ./team.lmctl

Each team member has an alias and a provider. seed starts each provider CLI once so lmctl can capture a native session id and snapshot the member prompt.

Put the target image somewhere stable

Place the image in a path every member can read. For example:

mkdir -p /tmp/lmctl-image-qa
cp /path/to/local-image.png /tmp/lmctl-image-qa/sample.png

Ask the Lead to run the pattern

Give the Lead one concrete instruction:

lmctl chat ./team.lmctl Lead "Run an image QA pass on /tmp/lmctl-image-qa/sample.png. Ask QA to describe the image, ask Reviewer to check the answer for missed details or hallucinations, then report the final verdict."

lmctl chat blocks for the Lead's turn and prints the reply. The Lead can delegate to members with the same command. With the default queue setting, a busy target returns a busy error and the Lead should inspect or retry later. If the opt-in mailbox queue is enabled, busy member chat queues in that (sender, receiver) lane. Base queued delivery is the next chat from that same sender to that same receiver once the receiver is free. A chat from another sender does not flush that lane. With lmctl serve start running in normal daemon mode, mailbox relay can drain queued lanes proactively; without that accelerator, if the original sender goes idle waiting for the queued reply, delivery can deadlock rather than just take longer.

Check team/self status

lmctl status

Use status for the current caller/team view. It speaks teamfile, member, delegation, and mailbox vocabulary.