mirror of
https://github.com/Orchestra-Research/AI-Research-SKILLs.git
synced 2026-09-19 05:04:39 +08:00
Zechen Zhang
f8afc9774d
The documented skill counts had drifted out of sync across the repo (CLAUDE.md said 90, CONTRIBUTING.md said 86/22, README sync line said 87, WELCOME/npm package said 86/22) while the actual repo has 98 skills across 23 categories. This corrects every current-facing reference and adds tooling to prevent recurrence. Historical changelog entries are intentionally left untouched. Changes: - CLAUDE.md: header 90 -> 98; fix per-category listings (01 Megatron-Core -> TorchTitan; 10 +ML Training Recipes; 13 +SwanLab; 14 +A-Evolve; 18 +Cosmos Policy/OpenPI/OpenVLA-OFT; 20 +Systems Paper Writing/Academic Plotting/Presenting Conference Talks). Per-category corrections harvested from PR #52 by @xiaolai. - CONTRIBUTING.md: 86/22 -> 98 across 23 categories; refresh the directory tree (04-peft -> 04-mechanistic-interpretability; add categories 0, 20-22). - README.md / WELCOME.md / demos/README.md / npm package README: fix current-facing counts to 98/23. - scripts/check-inventory.sh + .github/workflows/check-inventory.yml: CI guard that fails when documented counts diverge from the SKILL.md count on disk. - sync-skills.yml: prune build artifacts (node_modules/__pycache__/*.pyc/ .ipynb_checkpoints) before zipping and fail loudly if a skill zip exceeds 190 files, instead of hitting the marketplace 200-file rejection (issue #57). Addresses the inventory findings in issue #54. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>