Learning from experience hinges on extracting reusable knowledge and deciding when it applies to new situations. OntoAct is a hypothesis‑driven neuro‑symbolic framework built on ontological abstraction; it converts individual experiences into axioms over entity types, relations, and conditions, making underlying patterns reusable across contexts.\ \ Within OntoAct, a large language model (LLM) leverages this world knowledge to propose hypotheses and actions, while a symbolic verifier checks the supporting conditions against current observations. Verification reveals missing evidence, and subsequent observations and intervention outcomes update hypothesis assessments and guide further actions.\ \ We also introduce the Tacit‑FLE benchmark, constructed on the Factorio Learning Environment (FLE) to diagnose and repair hidden faults in existing factories using inherited operational records. Its synthetic maintenance corpus distributes evidence across records, and a performed repair may leave the fault unresolved, forcing agents to reassess diagnoses through interaction.\ \ Across Tacit‑FLE, Mars, and ALFWorld, OntoAct achieves leading task performance with efficient interaction and inference costs.\ \ Review: By tightly coupling symbolic abstraction with LLM reasoning, OntoAct offers a structured, interpretable, and scalable approach to adaptive decision‑making in complex environments.