An agent evaluates an R active binding in a selected environment

This precomputed case exercises the envir and enclos controls declared by the endpoint’s eval method. Returned scalars travel as Arrow IPC and are decoded by ducknng inside the fresh DuckDB client.

pi --model gpt-5.4 -e ./extensions/pi-ducknng/index.ts -p \
  "Using the package tools, start the R adapter and follow " \
  "its ducknng manifest. Evaluate code that creates " \
  "`analysis <- new.env(parent = baseenv())`, installs a " \
  "read-only active binding `mean_mpg` whose getter returns " \
  "`mean(datasets::mtcars$mpg)`, and returns " \
  "`identical(parent.env(analysis), baseenv())`. In a fresh " \
  "call, evaluate `mean_mpg` with `envir = "analysis"`. " \
  "Close the endpoint. Report the parent check, endpoint " \
  "process identity, and decoded mean, beginning with " \
  "`AGENT_VERIFIED_ACTIVE_BINDING` only when all calls " \
  "succeed."

AGENT_VERIFIED_ACTIVE_BINDING parent_check=true endpoint_process=64342 mean_mpg=20.090625