robinlidberg-dot

robinlidberg-dot/cap-shield-mcp

Skill Claude CodeCodex

Measure and reduce what an agent sends into its context. Use when an agent's context is large, when retrieval quality matters, when the user asks about token cost or context bloat, or when agent answers are wrong in ways that look like missing information rather than reasoning failure. Also use to measure compression…

robinlidberg-dot/cap-shield-mcp

MCP server Claude CodeCodexCursor

Context selection and compression for AI agents — measured, not estimated. Runs locally from the cap-shield Python package. Needs 1 environment variable to run.