resumelisted

# work:resume — pick a project's work back up You sat back down at a project. Your job is to **re-establish what is in flight on it and get the not-progressing work moving again** — without a human reconstructing state by hand, and **without running the resumed work on the interactive/personal machine**. Scope: `$ARGUMENTS` (empty = auto-detect the project from the CWD). | This skill | Not this skill | |---|---| | One **project's** whole in-flight state, then resume it | `sessions:continue` — resume **one transcript** in this window | | Reconstruct across sessions + PRs + worktrees + tickets | `work:dispatch` — route **one** item to an executor | | **Offload** execution to workers | `work:loop` — drain the **whole board, every project** | | Put crashed work back to *work* on the fleet | crash recovery that only reopened terminal windows is gone — see `/continue recover` | ## The compute rule (load-bearing — read first) The session running `/work:resume` may be on the **personal laptop** (`role=personal`, marked *never auto-place* because it freezes under load). So this skill **never runs the resumed agent work "here."** The orchestrating session does only light, read-only reconstruction (steps 1–3). Every actual resume in step 4 **offloads to a `role=worker` device**. The placement engine already excludes the personal box: `agents devices pick` (what `agents run --device auto` uses) returns a worker-only pool. Do not hand-pick devices and do not fall back to local — a wo