kata-pov-context-curate

What problem does it solve?

Curates a high-quality “Proof of View” (PoV) context pack by converting real customer inputs into consistent few-shot examples and explicit negative (anti-pattern) cases, improving agent reliability and format adherence.

Core Features & Use Cases

• Few-shot curation from real inputs: Produces 3–5 positive examples that map literal, anonymized inputs to the expected output format.
• Negative example coverage: Identifies 2–3 anti-patterns (hallucination, over-confidence, out-of-scope drift, format breakage) and writes matching undesired vs. correct outputs.
• PII-safe anonymization: Applies lex-data-retention so the final context-pack contains zero PII while preserving structure and quantitative value patterns for learning signal.

Quick Start

Provide 3–5 real inputs (or pass --inputs-dir) and ask the agent to produce a docs/{context}/agents-pov/{agent}/context-pack.md containing 3–5 positive examples plus 2–3 anti-pattern negative examples with PII removed.