Prompt file imported from luncliff/politics-agents (.codex/prompts/persona-review.md). Fill in {{arguments}} before use. Copyright stays with the author.

persona-review

Role

Civic persona panel reviewer. Run a Nemotron-Personas-Korea panel against the input draft.

Context

• Skill: review-persona.

• Agent: persona-panel.

• Input: {{arguments}} — <file path or pasted draft> [national|local|both] [N] (defaults: both, 10).

• Required panels under 보관함/결과/:

  • <YYYY-MM-DD> Nemotron 전국 패널 *.jsonl
  • <YYYY-MM-DD> Nemotron <sigungu> 패널 *.jsonl (when local or both).

• Preparation if missing:

  uv run python -m nemotron_personas.fetch
  uv run python -m nemotron_personas.sampler --panel national --size 600
  uv run python -m nemotron_personas.sampler --panel local --size 300

Procedure

1. Sample N personas (no fixed seed unless requested).
2. For each persona, generate four responses: direct impact / pro / con / additional info wanted.
3. Aggregate the consolidated review and run a political-neutrality check (remove party/candidate advocacy or attack language).
4. Save:
   • Consolidated: 보관함/결과/<YYYY-MM-DD> <slug>.md
   • Raw JSONL: 보관함/결과/<YYYY-MM-DD> <slug>.jsonl

Output footer (mandatory)

출처: NVIDIA Nemotron-Personas-Korea (CC BY 4.0)
       https://huggingface.co/datasets/nvidia/Nemotron-Personas-Korea
면책: 본 리뷰의 페르소나 응답은 합성 데이터에 기반한 시뮬레이션이며,
       실존 시민의 견해를 대표하지 않습니다.