Лучшие практики проектирования AI-агентов
vibe-coding
agents-best-practices is a Claude Code skill that helps design, scaffold, audit, and refactor agentic harnesses for any domain. Provider-neutral by design, it covers OpenAI, Anthropic, and OpenAI-compatible APIs across agent loops, tool design, environment-adaptive and late-bound capabilities, permissions, system prompts, planning, context compaction, memory, MCP connectors, self-refining recursive harnesses, observability, evals, prompt caching, and safety. The built-in MVP Builder Mode produces a concrete domain-specific blueprint — including a tool registry, permission matrix, and launch path — for the smallest reliable harness that can do useful work. Engineers building agents for research, finance, legal, support, healthcare, data analysis, or workflow automation will find it especially useful wherever a rigorous "model → action → observation" control loop is required.
- #agents-best-practices