Scaffolding tool and docs generate/describe the wrong eval scenario format #94

eval/scenarios/README.md is explicit: "All scenarios are JSON files only (no markdown counterparts)... Add new scenarios as <slug>.json." Every real file in eval/scenarios/ is .json. But three other places in the repo contradict this and would lead a contributor to produce the wrong format:

1. shared/scripts/scaffold-skill.mjs:60 — writes eval/scenarios/<skill-name>.md, a markdown file, not JSON.
2. docs/authoring-guide.md:52 — the "skill generation prompt template" tells an LLM to output "One scenario markdown file under eval/scenarios/".
3. CONTRIBUTING.md — describes eval scenarios as "simple markdown files" under "Add Evaluation Scenarios".

Why this matters

This is the exact tool/doc a new contributor is told to use (CONTRIBUTING.md → "Scaffold a new skill" → node shared/scripts/scaffold-skill.mjs). Following the documented path produces a scenario file in a format nothing else in the repo uses, that the harness doesn't expect, and that a reviewer will have to catch and ask to be redone by hand.

Reproduction

node shared/scripts/scaffold-skill.mjs test-skill "test description"
ls eval/scenarios/test-skill.md   # exists — wrong format
ls eval/scenarios/*.json | grep test-skill   # nothing — no JSON scenario was created