OmniRoute — сжатие токенов для сессий Claude

★ 7.9 · ai-tooling

omni-compression is a Claude Code skill that manages token compression in OmniRoute using three engines: RTK (terminal, build, and git output), Caveman (prose and chat history), and Stacked mode (an rtk → caveman pipeline for mixed coding sessions), achieving savings of 46–95%. The skill calls the OmniRoute REST API to switch compression modes at runtime, preview compression on arbitrary text via POST /api/compression/preview — returning compressed text and savings_pct — and configure Caveman language packs (pt-BR, es, de, fr, ja) and accessibility-tree filtering for Playwright/MCP browser tools. Caveman offers four intensity levels: lite, standard, aggressive, and ultra. It requires OMNIROUTE_URL and OMNIROUTE_KEY, and applies transparently to all requests without code changes — making it practical for teams looking to reduce LLM token costs across long agent sessions.