DeepEval4Claude

DeepEval4Claude grades your Claude agent's responses against evaluation criteria modeled on the quality standards used by top-tier consulting firms. It targets two failure modes most off-the-shelf evals ignore: sycophancy — the agent agreeing with a flawed premise instead of pushing back — and silent ambiguity, where confident-sounding answers quietly sidestep the actual question. One command installs it, with no API keys, SDKs, or account required. It's MIT licensed and free. DeepEval4Claude is aimed at developers and teams running Claude-based workflows who need actionable quality signals that go beyond basic pass/fail test suites and catch the subtle ways an agent's output can be wrong.

Used by developers and teams running Claude-based agents to score outputs against consulting-grade rubrics and catch sycophancy and silent ambiguity beyond basic pass/fail tests.