lightsagehq/lightsage

Diagnose Lightsage eval runs and completed results, then turn failure evidence and execution diagnostics into prioritized fixes and validation plans. Use for failed or regressed evals, recurring eval reviews, trace or tool-call investigation, and deciding what to change after an eval. Do not use to design configurations or analyze prompt visibility runs.

Analyze stored Lightsage prompt runs and turn visibility evidence into prioritized, verifiable actions. Use when a user asks what to do about brand mentions, recommendation position, sentiment, citations, answer-engine or coding-agent responses, or a recurring prompt-run review. Do not use for eval-result diagnosis or to start prompt runs.

Operate and troubleshoot Lightsage from a terminal, shell script, or CI workflow. Use only when the user explicitly asks for the lightsage CLI, command-line execution, scripting, or CI/CD. Do not use for outcome analysis that can be completed through a connected Lightsage MCP capability.