Paste a spec, see the topology DAG and the simulated ledger.
Spec format: swarm-specs/<name>.yaml. Docs in
SKILL.md.
275-session insights audit showed 25 sessions burned coordinating cross-repo PR cascades manually
(M164 deploy-migration, M17 yg-mcp-core extraction, @v1 reusable workflow across 14 repos). The shape was always
the same: pick a repo, branch, transform, push, watch CI, repeat. /ork:swarm-migrate turns the
coordinator into a YAML file and the workers into parallel subagents.
Use the example or your own. One transform per spec; topology via upstream edges.
What the live .swarm-state.json would look like at the moment wave 1 starts.
merged
CI pending
blocked-on-upstream
novel-failure-paused
| repo | wave | status | PR | CI | last action |
|---|
gh pr merge --auto) merge.repos[]. Even if the transform would help elsewhere.