The authors propose WEDGE, a weakness‑guided environment generation framework that tailors training data to the specific failure modes of a pretrained navigation policy. Given a policy, WEDGE leverages an LLM to extract executable weakness features by contrasting successful and failed rollouts, then validates these features on a diagnostic set and selects the most impactful weakness based on its success‑rate gap. The validated weaknesses are mapped to underlying environmental factors, which WEDGE modulates to create scenarios that lie in the difficulty regimes where the policy currently underperforms, thereby delivering targeted adaptation experience. Experiments across three navigation agents—each employing distinct adaptation mechanisms—reveal that policies exhibit unique weakness profiles and benefit from correspondingly customized environments. Weakness‑conditioned adaptation consistently surpasses baseline random and hard‑example generation on the ProcTHOR domain, and the performance gains transfer to Uni‑NaVid and Gemini, extending further to HM3D after ProcTHOR adaptation. Iterative re‑diagnosis demonstrates that dominant weaknesses can either persist or evolve as the policy improves, providing updated cues for subsequent environment construction. These results illustrate that policy failures serve not only as diagnostic signals but also as actionable guidance for generating more effective training worlds.