← ClaudeAtlas

resumelisted

Reconstruct swarm state after a pause, a crash, or a handoff to a different machine or person. Use when starting a session that should continue prior swarm work instead of starting fresh.
rmzi/portable-dev-system · ★ 24 · AI & Automation · score 53
Install: claude install-skill rmzi/portable-dev-system
# /resume — Swarm State Reconstruction Rebuilds `.claude/swarm/` state (phase, tier, tasks) and native `TaskCreate` entries so work can continue where it left off. Tries the cheapest, most complete source first and falls back only as needed — most resumes are same-machine and need nothing beyond a local file read. ## Invocation ``` /pds:resume # auto-discover: local state, then ticket, then label search /pds:resume 42 # explicit ticket number — skips discovery, goes straight to reconstruction ``` ## Why This Exists Claude Code's native Task-tool state (`~/.claude/tasks/`) and team config (`~/.claude/teams/`) are local to the machine that created them — no cross-device sync, and per Claude Code's own docs, `/resume`/`/rewind` do not restore in-process teammates. `.claude/swarm/*` is gitignored — local-only. The one thing that *is* durable and cross-machine is the GitHub ticket `/pds:ticket` maintains. This skill treats the ticket (and, failing that, the archived swarm report) as the recovery source of truth, in priority order from cheapest/most-precise to most expensive/coarsest. **Platform constraint, not a bug**: a previous session's in-process workers cannot be reattached, on any path below. This skill reconstructs *task and phase state*, not live agent processes. Phase 3's normal pull-model redispatch picks the work back up. ## Protocol ### 1. Same-machine fast path Check first, before anything else: ```bash ls .claude/swarm/checkpoint.json