Assembly graphs contain valuable yet frequently overlooked information that can enhance assembly completeness and accuracy by revealing contig connectivity. Here, we demonstrate how leveraging these information-rich structures enables the discovery of hidden diatom diversity in environmental DNA shotgun datasets. While GetOrganelle has previously been used for organellar genome assembly from isolated tissues, we present its first application to water metagenomic data. We tested the efficiency of this tool on three eDNA samples with varying diatom abundances, finding that GetOrganelle alone often results in fragmented scaffolds due to the complexity of mixed-species samples. By implementing additional manual disentanglement of assembly graphs, we successfully recovered fully complete plastome structures. From high-abundance bloom samples, we recovered complete plastomes of Stephanodiscus hantzschii (129,551 bp and 129,553 bp) with 99.9% pairwise identity, despite originating from distinct geographical locations (USA and Czech Republic). From a lower-abundance non-bloom sample, we reconstructed a potentially novel Cyclotella species plastome (133,867 bp) showing only 92.4% similarity to its closest available reference Cyclotella atomus. Our assembly quality assessment confirmed effective manual disentanglement of target sequences from complex assembly graphs, even when abundance was low. By integrating sequence similarity, gene order conservation, and phylogenetic analysis, we achieved robust species-level resolution and resolved taxonomic uncertainties previously identified. Our findings demonstrate that mining metagenomic data with GetOrganelle reveals previously hidden diversity and provides higher taxonomic resolution than traditional methods. This approach proves especially valuable for diatoms and other microeukaryotes, where reference organellar genomes remain severely underrepresented in existing databases.