Multi-model code review skill for Claude Code — invoke DeepSeek, Doubao, Qwen, OpenAI & more in parallel to review your git diff. 多模型并行代码审查 skill,同时调用多个 AI 模型交叉验证。
Self-evolving, eval-gated harness skill for Claude Code agents: learn from traces, detect CLAUDE.md drift, and update rules/skills only when a held-out eval gate proves it. Stdlib Python, drop into any repo.
Fable Workflow — a portable Claude Code / Cursor agent skill that makes any model (Opus, Sonnet, Haiku, Fable) find the unknowns before building, not after.
A production-grade autonomous coding skill for Claude Code, Codex, OpenClaw, and Hermes. It routes tasks by complexity and risk, researches current language/framework anti-patterns, plans before coding, verifies with evidence, uses specialist reviews, pivots failed strategies, and performs bounded repair loops.
Cut PDF token costs in Claude Code by 65-99%: a skill that reads PDFs as clean local text instead of native page images, with search, page navigation, and a reader agent for whole-document work.