resumelisted

Resume an atomic run in this project. The workflow runtime's own resume works only within the same session; this skill performs the cross-session logical re-entry from persisted artifacts. Steps: 1. Run `"${CLAUDE_PLUGIN_ROOT}"/bin/run-state.sh status` (a run id in $ARGUMENTS overrides `active_run`). If the status is `complete`, report that there is nothing to resume and stop. 2. Reconstruct where the run stopped from `.atomic-cc/runs/<run_id>/`: - the highest-numbered `turn-*/` or `iter-*/` directory and its receipt/report = last completed implementation stage; - the latest `review-round-latest.json` = the last review round, its per-reviewer verdicts, and its `consolidated_findings`; - `decision.json` = the terminal decision, including unresolved blocking findings, unproven requirements, and any `escalation` or `failure_reason`. 3. Report the reconstruction to the user in 3–5 lines, and say explicitly which of the three terminal shapes this is, because they resume differently: - **escalation** — the worker stopped for a product/architecture/scope decision. Do NOT re-enter the workflow: put the question to the user, get the decision, and pass it as part of the objective on re-entry. - **blocked** — a repeated environment blocker (dependency or tool failure) tripped the threshold. Re-entering unchanged will reproduce it; fix or report the environment first. - **needs_human** — the budget ran out without the gate converging. Re-entry is meaningful. 4. Re-ent