We present the strongest robust constraints on primordial non-Gaussianity (PNG) from currently available galaxy surveys, combining large-scale clustering measurements and their cross-correlations with the cosmic microwave background. We update the data sets used by Giannantonio et al. (2012), and broaden that analysis to include the full set of two-point correlation functions between all surveys. In order to obtain the most reliable constraints on PNG, we advocate the use of the cross-correlations between the catalogs as a robust estimator and we perform an extended analysis of the possible systematics to reduce their impact on the results. To minimize the impact of stellar contamination in our luminous red galaxy (LRG) sample, we use the recent Baryon Oscillations Spectroscopic Survey catalog of Ross et al. (2011). We also find evidence for a new systematic in the NVSS radio galaxy survey similar to, but smaller than, the known declination-dependent issue; this is difficult to remove without affecting the inferred PNG signal, and thus we do not include the NVSS auto-correlation function in our analyses. We find no evidence of primordial non-Gaussianity; for the local-type configuration we obtain for the skewness parameter $ -36 < f_{\mathrm{NL}} < 45 $ at 95 % c.l. ($5 \pm 21$ at $1\sigma$) when using the most conservative part of our data set, improving previous results; we also find no evidence for significant kurtosis, parameterized by $g_{\mathrm{NL}}$. In addition to PNG, we simultaneously constrain dark energy and find that it is required with a form consistent with a cosmological constant.