This collection of articles explores the complexities of managing skills for AI agents, moving beyond simple installation to address practical challenges. It covers techniques for optimizing skill loading to reduce token usage and latency, methods for determining if a skill actually improves task success rather than just being relevant, and the risks of silent overrides when similar skills are installed together. The pieces also delve into compiling skills into state machines for more reliable execution and the pitfalls of simplistic acceptance criteria for skill edits. AI
IMPACT These methods address core challenges in making AI agents more reliable and efficient, impacting how developers build and deploy complex AI systems.
RANK_REASON The cluster aggregates research papers and methods related to AI agent skill management. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv:2609.30123
- arXiv:2609.32274
- arXiv:2609.35692
- arXiv:2609.36043
- Claude Code
- .codex
- Copilot CLI
- Gemini CLI
- HEXIS
- Qwen3.8-27B
- Qwen3_8B
- SAGE
- SkillDelta
- Workday AI Research
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →