worklisted

Work from the active plan, not from memory alone. ## Steps 1. **Resolve the target plan path** (must run before any branch or plan-file operations): - `/work` operates on single-file plans (`docs/plans/active/<date>-<slug>.md`) only. - Enumerate candidates: `.md` files directly under `docs/plans/active/` (excluding `.gitkeep`). **Ignore directories** — if a directory is the only entry, stop and report: this looks like a Ralph Loop directory plan (`_manifest.md` + `slice-*.md`); Ralph Loop's autonomous execution system was retired (see git history for the removed orchestration scripts, and `internal/org` for the current autonomous execution surface, `/org`). Convert the plan to a single-file plan via `/plan` and re-run `/work`, or use `/org` for autonomous multi-seat execution. - If exactly one candidate file exists, use it. - If multiple candidate files exist, ask via AskUserQuestion which plan this `/work` run targets, and use the selected path. - If none exist, stop and ask the user to run `/plan` first. - Downstream steps in this skill — and downstream skills (`/cross-review`, `/pr`) — MUST use this resolved path instead of rescanning `docs/plans/active/`. 2. **Resolve or resume the task worktree**, based on the plan resolved in Step 1: a. Read the resolved plan to extract metadata (type, issue number, slug). b. Determine branch name by running `./scripts/branch-name.sh from-plan <resolved-plan-path>`. c. Branch names must validate with `./scri