Newsjack v0.1.16

Fixed

• Claude.ai marketplace sync no longer treats the source-checkout CLI shim as
  a hosted plugin executable. The marketplace now packages the dedicated
  plugins/newsjack/ directory instead of the repository root, while keeping
  the canonical skills and Medialyst connector configuration shared with local
  installs. CI and the release workflow now reject any regression that would
  expose a top-level bin/ through the hosted plugin package or reintroduce
  unsupported manifest metadata that Claude.ai would strip with a warning.

Added

• build-ai-visibility-panel turns any public URL and description into an
  evidence-bound prompt panel for AI-answer tracking. Six composable atoms
  research ICPs and buyer jobs, design intent coverage across six prompt
  groups, generate blinded prompt variants, run contamination and duplicate QA,
  and produce a provisional measurement plan with explicit denominators,
  sources, gaps, and human approval gates. Reports name the groups the way a
  marketer would — Brand, Shortlist, Category, Problem, Goal, Market — and keep
  the B0–B5 codes in the machine artifacts only. Synthetic fixtures and a
  ten-case public-URL eval exercise the workflow without private client data.

What's Changed

• eval: 4-model angle-quality benchmark (Sonnet 5 & 4.6 vs Opus 4.8 & Fable 5) by @elvisun in #54
• feat(skills): add same-outlet-ranker by @prcarly in #56
• feat(skills): add evidence-bound AI visibility prompt panels by @elvisun in #57
• feat(skills): add evidence-aware AI visibility writing audit by @elvisun in #58
• docs(readme): add AI visibility writing skill by @elvisun in #59
• docs(readme): complete the skill catalogue and split out AI visibility by @elvisun in #60

New Contributors

• @prcarly made their first contribution in #56

Full Changelog: v0.1.15...v0.1.16