Generate a judge contest

Read references/contest-format.md before authoring any file.

1. Confirm the contest parameters: contest id and name, target course and lecture, divisions (each with open/close datetimes), which concepts the contest covers, the number of problems (4-6 is typical), and target programming language(s). The contest belongs at <courseId>/<lectureId>/<contestId>.contest.yaml, and its problems belong under the same course's problems/ directory. If any are missing, ask before generating when the user is available; otherwise choose reasonable values from the context and list them as open decisions in the final report.

2. Plan the problem set as drafts (id, name, overview, difficulty, score):

   • Use only concepts the learners have already studied in the covered materials.
   • Span difficulties from 初級 to harder levels so every participant can solve something and strong participants stay challenged; order problems from easy to hard.
   • Assign scores increasing with difficulty (e.g. 100, 200, 300, ...).

3. Generate each problem directory by following the generate-judge-problems skill workflow (model answer first, run it to produce .out files unless the custom judge does not compare stdout, cross-check with a second language, validate).

4. Write <contestId>.contest.yaml per references/contest-format.md, listing every division and every problem with its score. Place the file inside the lecture directory when the contest belongs to a course.

5. Validate with the exercode-problem CLI (it ships with @exercode/problem-utils, which must be a declared repository dependency — see setup-exercode-course-repository) and fix until clean; rerun after every fix until both commands report success:

   bunx exercode-problem validate-problem <problemDir>...
   bunx exercode-problem validate-contest <contestYamlPath> --problems-dir <problemsDir>
   

   Address warnings too unless they are intentional; if you keep a warning, tell the user why in the final report.