V0 now includes 20 Review Mode challenges. They follow the same quality bar as the first challenge: real source material, traceable links, an adapted AI diff, expected findings, and a structured rubric.
Task Mode is still a later phase. It needs a runner, starter repositories, visible tests, and hidden tests before those tasks should be presented as live practice.
Challenges are versioned repository assets, not database-only rows.
challenges/
review/
001-sympy-point2d-ai-patch/
metadata.json
README.zh.md
README.en.md
ai-pr.diff
expected-findings.json
rubric.md
Each challenge should make these questions reviewable:
- Is the source real and traceable?
- Is the adapted AI patch reproducible?
- Are the expected findings and rubric inspectable?
- What engineering judgment is being trained?
- 001 Can this AI fix be merged? Reviewing a SymPy Point2D regression
- 002 Do not trust internal middleware headers
- 003 Server Actions relative redirect SSRF review
- 004 Axios baseURL absolute URL bypass review
- 005 Express route pattern ReDoS review
- 006 CookieJar prototype pollution blacklist review
- 007 JWT verification without algorithm pinning review
- 008 Django username newline anchor review
- 009 Django password reset token email binding review
- 010 Django admin save_as_new permission review
- 011 Django readonly JSONField display review
- 012 Requests redirect method chain review
- 013 Requests urllib3 exception boundary review
- 014 pytest skipif string-condition cache review
- 015 Sphinx autodoc empty all review
- 016 Astropy nested CompoundModel separability review
- 017 Astropy NDData mask propagation None-branch review
- 018 xarray update preserves dask chunks review
- 019 scikit-learn SVM empty support vectors review
- 020 SymPy partitions dictionary reuse review
Task Mode will add:
starter/tests/visible/tests/hidden/validator.shsolution.patch
Until the runner exists, the live V0 set is the 20 Review Mode challenges.