Start a bounded AI-agent development loop

A bounded loop gives an agent room to make progress without turning an open prompt into an unreviewable process. The boundary limits execution; tests and review establish correctness.

Define one outcome

Create one task with a concrete validation gate. Record dependencies as blockers rather than burying them in prose.

./.juno_task/scripts/kanban.sh create --body-file task.md --status todo
./.juno_task/scripts/kanban.sh ready --sort asc

Initialize and run once

npm install -g @yylo/cli@next
yy init --task "Implement the selected ready task and run its exact gate" --subagent claude
yy start -b shell -s claude -i 1 -v

Without -p, YYLO uses the initialized project prompt. Start with one iteration, not a hopeful large number.

Review before continuing

• Read the kanban response and check that it names the exact validation evidence.
• Inspect the commit and ensure unrelated changes were preserved.
• Run the gate yourself when the risk warrants it.
• Continue the captured session only when continuity is useful.

For a queue that is safe to advance repeatedly, Run Until Completion adds stale detection. It does not remove the need for bounded per-iteration work.