Since #4509, Claude Code skills run their start step as they load, so
the paid /plan-ceo-review auto-decide check could miss a run that really
did auto-decide. It now accepts the start output in the loaded skill
text.
Why: one of the five passing runs captured for that check was credited
only through the session journal, and that path needed a start command
in the tool calls.
Nothing to do: this changes test code and the journal reader the tests
use; no skill text changes.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>