YYLO

2026-07-18 · Updated 2026-07-19 · 1 min read

Run kanban tasks safely in parallel

Select dependency-ready work, cap concurrency, isolate prompts, and review structured aggregation before closing tasks.

By Juno AI INC · parallel-runner · yylo-ledger

Parallelism is safe only when tasks do not compete for an ordering boundary, a mutable file, or an external quota they cannot share.

Make dependencies explicit

sh
./.juno_task/scripts/kanban.sh deps add --id UI_TASK --blocked-by API_TASK
./.juno_task/scripts/kanban.sh ready --sort asc

ready means every declared blocker is satisfied. A related task is context, not a dependency edge.

Include the task placeholder

A custom prompt does not inject the task automatically. Include {{task_id}} and keep the file reviewable.

sh
./.juno_task/scripts/parallel_runner.sh \
  --kanban-filter "ready --tag frontend --limit 6" \
  --prompt-file .juno_task/prompts/implement-task.md \
  --parallel 2

Keep the filter quoted. Start below provider and machine limits. Pi live tmux mode is only supported with --parallel 1.

Inspect, then retry narrowly

Read parallel_runner_status.json, each item JSON, and aggregation_*.json. A failed item should be retried by task ID after its failure is understood; do not relaunch every successful task. Use parallel_runner_wait.sh --timeout SECONDS for deterministic nonblocking waits and --stop-all only when broad session shutdown is intentional.