Prompts and model use
Understand which FEHM workflows are deterministic and where optional external commands may use an LLM.
Ordinary indexing, context retrieval, skill routing and static reports run without an LLM. FEHM estimates the size of generated context; model tokens are consumed when an assistant or model reads that context.
Inspect prompts
fehm prompts
fehm prompts --jsonThe /prompts workspace inventories prompts and tool contracts, shows evidence, compares candidates, and retains revision/evaluation records. It does not invent missing roles, permissions or requirements from weak evidence.
Evaluate honestly
The implemented evaluation workflow uses supplied cases and recorded outputs. Consult fehm help and the technical prompt specification for input schemas. Recorded model outputs are not proof of a new live model run. This checkout does not expose the older fehm ai-evaluations command.
Optional external commands
Semantic summaries can invoke a configured command provider. Skill routing supports an explicitly configured reranker command. Those commands may contact a model service and incur costs; FEHM cannot guarantee their data handling. Inspect inputs and command configuration before enabling them.
Prompt application uses its own explicit --apply --approve flow. Do not treat a proposed candidate as a tested fix.