A Skill page must open its work

One Skill page in a private first-run notebook had a useful finding about a published draft and a scheduling mismatch. Its citations were clickable, but the reader stopped each retained run part after 4,096 characters. Several dialogs ended inside serialized diagnostics. The links to the draft and its cover appeared on the page as dead text. A reviewer could see the conclusion without opening the work behind it.

The page also described the Skill's intended fact checking as though the observed drafts had passed it. The instruction file established a rule, not evidence that a particular run followed it. The two problems reinforced each other: strong wording made a weakly inspectable result sound settled.

We kept the full bounded run part available in the owner-only source dialog and made absolute local artifact paths open as local files. The writer now links an inspected output beside a concrete quality finding. When no artifact survives, it attributes the statement to the run report and leaves the artifact's contents unknown. The Skill lead calls a documented outcome an aim until an openable result supports stronger language. We also preserve the mapped invocation title because changing it can detach later run history.

An independent role-based AI review reopened one accepted Skill page in Chrome at desktop and phone widths. It followed all 14 displayed citations, opened the linked local drafts and image, and checked the first-fold wording, Full memory, and privacy state. No P1 or P2 remained in that sample. The raw run dialogs are still long on a phone, and local files are mutable or may be missing on another machine. Immutable artifact snapshots and claim-focused navigation remain the next test of whether a source link stays useful over time.

The 1.9.0a32 preview notes give the exact install pin and trial limits. The first-run guide shows how to start.