Agentica — интеграция с прокси для многоагентных систем

integration

agentica-claude-proxy is a Claude Code skill that provides a technical guide for integrating the Agentica SDK with the Claude Code CLI proxy when building or debugging multi-agent orchestration systems. It documents the full request chain — Agentica Agent → S_M_BASE_URL (localhost:2345) → Claude Proxy (localhost:8080) → claude -p → Claude CLI — along with three critical implementation requirements: the --allowedTools flag (Read, Write, Edit, Bash) to prevent permission errors, SSE streaming format (data: {...}\n\n) to avoid APIConnectionErrors, and a REPL response format using Python return blocks. The skill includes ready-to-use anti-hallucination prompt engineering to stop agents from falsely claiming success without invoking tools, path sandboxing rules (avoid /tmp/, use project-relative workspace/ paths), server startup commands, curl-based proxy tests, and a reference error table covering the most common failure modes. Targeted at developers building Agentica-powered agent swarms on top of Claude Code.