prompt evaluation

A practical guide for Bash users to make LLMs reliable, parseable CLI components: set up curl/jq and an ai() helper; write prompts with explicit roles, goals, and constraints; demand strict JSON; use delimiters and few-shot examples; add a test harness to diff outputs; control temperature/tokens and chunk large inputs. Includes copy‑paste snippets and real‑world patterns for logs, configs, and automation.