Reporting marginal/borderline/almost not significant, yet significant effects I have a figure as follows. Confidence intervals are only slightly bounded away from each other at the edges. Thus, $x$ seems to have an association with $y$.
How would you describe such an effect? Like every other significant effect or should I be more cautious while interpreting the relationship. For example, there was a borderline effect of $x$ on $y$?

 A: If I understand you correctly, the intervals overlap in between the range of your data, but not at the edges. As a general rule, if your 95% confidence intervals at the extremes are bounded away from each other (i.e., no overlap), then your estimate of the difference will be deemed statistically significant by conventional standards. However, the converse is not likely to always be true. Thus, at the edges you could also have some marginal degree of overlap and still find a "significant" effect.
A downward trend is apparent, but the effect of $x$ on $y$ is marginal. As a visual heuristic, if a straight, horizontal line can fit within the error ribbon, then the effect is not significant. As indicated in your post, your effect is, in fact, significant. I would recommend being transparent about the relationship. For instance, try reporting your $p$-value as an equality (e.g., $p = 0.499$) rather than an inequality (e.g., $p < 0.05)$. Doing so will make the "borderline effect" appear less pronounced (see, e.g., Wasserstein et al. 2019).
In sum, yes, it is safe to report this as a marginally significant, downward sloping trend, but I would be transparent about the degree of overlap, if any, you are observing.
