In directed and signed graphs the presence of both edge directions and signs makes the community detection problem very challenging. Most existing algorithms rely on modularity maximization, a global metric that often overlooks local graph properties within individual communities. In many real-world scenarios, such as who-trusts-who networks, meaningful communities are defined by local constraints –for example, groups of individuals commonly trusted or distrusted by the same set of people. We propose here a novel methodology for discovering locally consistent and closed communities in directed and signed graphs. Our approach first transforms a signed directed graph into an undirected tripartite graph, where each node is represented in its three roles: source, target of positive edge, and target of negative edge. We then embed nodes of this graph into a lower-dimensional space, identify dense regions of source nodes as seed communities, and grow the seeds through closure operations to include all related and consistent target and source nodes. Experiments on two public datasets demonstrate that our method uncovers semantically meaningful and structurally consistent communities that are not discovered by other existing algorithms.