Super Loop

The durable front door to a "super loop" — launching HEADLESS work sessions in BULK, one detached `/goal` worker per bounded account session slot, each resolving a top-ranked ready leaf and closing it by witnessed ancestry. Wraps the proven launchers (`tools/issue_dispatch.py --wave`, `fak dispatch wave`, `tools/launch_wave_detached.ps1`) with the discipline the raw scripts assume: PLAN by default, price the fan-out for tree-collisions AND account capacity before a single worker spawns, re-check the no-DoS preflight cap per spawn, and never confuse a launch with a ship. Regime-aware, not a fixed script: it reads the host first and adapts. Its FIRST move is a one-read gate that short-circuits when a standing dispatcher (`FleetIssueDispatch`) is already gardening — so it never marches the whole launch procedure only to discover the loop was already running. Otherwise it adapts — RECLAIM stale workers/residue before growing, RAMP a cold host canary-first (promote on witnessed commits), and run 12h+ MARATHONS as a cadence of waves with budget and stop signals. Use when the operator says "launch a wave", "run a super loop", "spin up N workers on the top issues", "fan out headless sessions", "drain the backlog in parallel", "start the overnight fleet", "clean up the old workers first", "ramp up slowly", or "keep it running for 12 hours".

mkdir -p ~/.claude/skills/super-loop
curl -L https://claudskills.com/skills/super-loop/SKILL.md \
  -o ~/.claude/skills/super-loop/SKILL.md

@misc{super-loop-2026,
  author    = {anthony-chaudhary},
  title     = {Super Loop [Claude Code skill]},
  year      = {2026},
  publisher = {ClaudSkills},
  url       = {https://claudskills.com/skills/super-loop/}
}