shepherd

How to expose a long-running local dev/test server from a Shepherd worktree so the operator's live preview finds it, by writing the port to a `.shepherd-preview` file. Load when starting a dev server, test server, or any other long-running local preview in this session.

How a Shepherd session opens a pull request - exactly one PR per session, what to do when the work is too large for one (promote to an epic, or ship a slice plus a follow-up issue), and how to declare manual operator steps so they survive merge. Load before running `gh pr create`, or as soon as a task looks like it needs more than one PR.

Onboard a repo into Shepherd's work-execution model — turn a PRD or stated goals into a drainable backlog of issues/epics (tracer-bullet vertical slices, epic-dag dependencies) and seed repo-specific agent instructions, then point at the first task to drain. Use when onboarding an existing codebase to Shepherd, bootstrapping a greenfield project from a PRD, or asked to turn intent into Shepherd-shaped issues/epics.

Review all open PRs for plausibility, validity and product-fit, propose a merge order, then merge on approval. Use when asked to "work through the open PRs", "review and merge the PR queue", "run the merge train", or triage what should land next.