glassray

Drive Glassray Coach — the local AI-agent flow-improvement loop — from the CLI: derive a flow + assertion rules from the codebase, author glassray.yaml with a run recipe, pin inputs, then prove a change held with run/compare (pass rates + cost). Use when working on an AI agent that sends traces to a local Coach (glassray) server, when asked to set up glassray for a flow, or when asked about the agent's quality, failures, rules/evals, or a model swap.

Drive Glassray Coach — the local AI-agent trace debugger — from the CLI: inspect captured traces, define durable flows that scope your agent's behaviours, derive evals from the agent's own rules in code, and run the discover → fix → verify loop. Use when working on an AI agent that sends traces to a local Coach (glassray) server, or when asked about the agent's quality, failures, evals, or flows.