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