kelly-agent-evallisted

# Agent Eval & Regression Board ## Overview Use this skill as a generic quality gate for teams shipping multiple LLM-agent workflows who need to catch regressions before a release. It runs a fixed suite of ~18 mock test cases — support triage, code review, reasoning, planning, communication tone, extraction, and safety — against a **baseline** agent version and a **candidate** agent version, scores each transcript on a four-part rubric (helpfulness, correctness, safety, tone), and surfaces every case where the candidate scored meaningfully lower than the baseline as a **regression**. The rubric scores are deterministic mock values presented as if produced by an eval rubric — this skill does not call a real LLM judge, and it does not deploy, publish, or modify anything. It only reads/writes local handoff files. Default interaction mode: App UI. Unless the user explicitly asks for chat-only handling, check onboarding/config, generate or load the local eval run, start or reuse the local app with `app/start.sh`, and give the actual local URL. Use chat-only mode only when the user says "chat only", "no UI", or similar. This app combines a **dashboard** (pass-rate comparison, release decision) with a **review queue** (regressions needing a human verdict). ## App UI Screenshots <table> <tr> <td width="50%"><img src="assets/screenshots/overview.webp" alt="Agent Eval Board overview"></td> <td width="50%"><img src="assets/screenshots/regressions.webp" alt="Agent Eval B