v4.2.0

Minor Changes

• 4865aed: Tighten what the server tells a connecting agent. The instructions now say
  that for work done with these tools the journey playbook is the procedure,
  whether it comes from the plugin's runpod-mcp skill or from
  read-guide; the plugin's other skills stay the lane for the CLIs, image
  builds and golden paths. A resource the user names by id is in scope for the
  change they asked for and is deleted only when they ask for the delete, and
  create-cluster waits for the user's go on the quoted shape and total hourly
  price. Jobs are submitted with run-endpoint and waited on with
  get-job-status; runsync-endpoint is for a warm endpoint and a fast job.

  The 402 hint points at list-billing, and the 403 hint says a permission error
  on a resource the agent did not create is final.

  Tool descriptions: create-cluster prices the whole shape (pods times GPUs per
  pod) and waits for the go; create-endpoint describes the CPU choice next to
  the GPU one; update-endpoint puts idleTimeout under workers and states
  that gpu.pools and gpu.excludedTypes are one selection; delete-endpoint,
  delete-pod, delete-cluster and delete-secret allow the one resource the
  user asked to delete by id.

• 6ffb28e: The journey playbooks now come only from @runpod/plugin-knowledge, bumped to
  1.7.1. The server no longer keeps its own copy under runpod://skills/: an
  agent loads the plugin's runpod-mcp skill, or read-guide runpod-mcp without
  the plugin, and that routes it to the journey guides (discovery,
  lifecycle-crud, serverless-deploy, pod-deploy, pod-doctor,
  endpoint-ops, cost-audit). resources/list serves the plugin guides as
  runpod://guides/<id>. The instructions and the ask_question text point at
  read-guide instead of runpod://skills/.

  The instructions also tell an agent whose instructions look out of date to
  reconnect the server (or ask the user to), since reconnecting re-runs
  initialize and delivers the current instructions and tool list.