Turning an engineering request into an executable PDE simulation requires consistent geometry, physics, boundary conditions and numerical settings. Recent agent‑assisted systems can automate the setup, yet successful execution alone does not guarantee that the specification preserves the original intent or correctly handles missing information. We investigate whether agents can reliably produce complete executable JSON specifications covering geometry creation, meshing, solving and result export.\n\nThe workflow combines model extraction, deterministic scientific and capability checks, explicit clarification, and validated local patches that keep unrelated accepted fields intact. Specification correctness, execution completion and numerical verification are evaluated separately. On 120 template‑related requests the full workflow achieves a 98.3% correct first‑turn specification rate, compared with 92.5% for schema‑assisted single‑shot and 82.5% for model‑only, gains of 5.8 and 15.8 percentage points respectively. No false‑ready decisions occur among the 42 requests that require clarification or rejection.\n\nComponent ablations reveal the contribution of scientific checks and conditional repair. Downstream studies cover 1D, 2D and 3D steady and unsteady, linear and nonlinear problems, including flow and elasticity with complex three‑dimensional wing geometries. A 30‑case gallery demonstrates automated execution from retained JSON specifications; in 12 live user‑to‑result tests, 9 achieve joint specification, execution, export and numerical‑reference success (75.0%). These results support traceable, engineer‑reviewed simulation assistance within the tested scope, while fresh‑response regressions highlight consistency as a priority for further improvement.\n\nReview: The paper shows that a systematic combination of scientific validation and conditional patching markedly improves agent reliability for complex PDE specifications, offering a verifiable path toward automated engineering simulation, yet consistency across diverse requests remains an open challenge.