Keep evaluation-ai independently usable without Loom. Define stable capability/protocol boundaries, document direct use from Claude Code, Cursor, Crush, OpenCode and Loom, and add an optional MCP adapter for evaluation/verification operations where useful. Keep transport/agent concerns out of core. Related: engineering-standards#37.
Keep evaluation-ai independently usable without Loom. Define stable capability/protocol boundaries, document direct use from Claude Code, Cursor, Crush, OpenCode and Loom, and add an optional MCP adapter for evaluation/verification operations where useful. Keep transport/agent concerns out of core. Related: engineering-standards#37.