subagent-driven-development-herdrlisted

# Subagent-Driven Development (herdr overlay) **Precondition:** `HERDR_ENV=1`. If unset, use stock `superpowers:subagent-driven-development`. Follow `superpowers:subagent-driven-development` exactly, with ONE change: dispatch each **implementer** as a live herdr pane via `shepherdr` instead of the native Agent tool, so you can watch and audit it. **Do not add a codex or second-opinion review here** — the review and fix loop is unchanged from stock SDD (second opinions belong only to the brainstorm and plan gates). ## Setup (once) Create a workers tab (or reuse the one the `worktree.created` hook made): ```bash herdr tab create --workspace <workspace-id> --label workers ``` Record the workers tab's root pane id (from the `tab create` JSON, `result.root_pane`) as your `--from-pane`. ## Per task 1. **Implement (as a visible pane).** Write the task's isolated context (task text + interfaces + global constraints) to `<scratch>/task-<n>.context.md`, then: ```bash shepherdr dispatch --role implementer \ --context-file <scratch>/task-<n>.context.md \ --from-pane <workers-root-pane> --task-id task-<n>-impl --scratch <scratch> ``` The implementer runs headless in the pane, streaming its work (tool calls + reasoning) so you can audit it live and afterward; the pane stays open. Read its result from the printed `result_file` (or `result_text`). 2. **Review + fix per stock SDD.** Run the task review and any fix loop exactly as `superpowers:subagent-driven-development` prescrib