v2.4.0

Added

• effectiveEffort in the delegate result: the current value of each option recognized as
  effort, keyed by option id, as the agent reported it when the prompt went out. It is
  present even when effort is omitted, because the model then runs at the level Cursor
  saved for it, and absent when no reply confirmed a value. Read from replies the bridge
  already receives, so it adds no ACP requests.
• doctor with deep: true reports each currentModelOptions entry's name, ACP
  category and currentValue when the agent sends them as strings. The id and values
  alone could not show that grok-4.7's reasoning_effort is its effort setting, or which
  level Cursor had saved for it.

Fixed

• effort reaches a model that advertises it under an unfamiliar id. An option the agent
  marks with ACP's thought_level category counts as effort whatever its id or name, and
  the fallback for agents that send no category also knows reasoning_effort and names
  containing "effort". grok-4.7 declares reasoning_effort ("Effort"), which was missed, so
  effort: "high" failed with invalid-effort claiming the model had no effort setting.
• invalid-effort no longer says Accepted: none when no option is recognized as effort.
  Missing an option is not proof the model has none, so the error lists the options the
  model advertised, with their names, categories and values.

Security

• Lockfile bump of hono to 4.13.7, clearing a moderate audit advisory.

Internal

• Lockfile bumps of @modelcontextprotocol/sdk to 1.30.1 and the dev dependency
  @types/node to 26.6.3.