script-build-and-framework-tmp/script_build_host/tests/test_prompt_quality.py

from __future__ import annotations

from agent import get_preset

from script_build_host.agents.presets import register_script_presets
from script_build_host.agents.prompts import (
    COMPARISON_WORKER_PROMPT,
    COMPOSE_WORKER_PROMPT,
    DIRECTION_WORKER_PROMPT,
    ELEMENT_SET_WORKER_PROMPT,
    EVIDENCE_RETRIEVAL_WORKER_PROMPT,
    PARAGRAPH_WORKER_PROMPT,
    PLANNER_PROMPT,
    ROOT_VALIDATOR_PROMPT,
    SCRIPT_CANDIDATE_VALIDATOR_PROMPT,
    STRUCTURE_WORKER_PROMPT,
    script_build_prompt_manifest,
)


def test_logical_retrieval_prompt_keeps_provider_choice_host_owned() -> None:
    manifest = {str(item["preset"]): item for item in script_build_prompt_manifest()}
    frozen = manifest["script_evidence_retrieval_worker"]
    assert frozen["source_id"] == "evidence_retrieval_worker.md"
    assert frozen["content"] == EVIDENCE_RETRIEVAL_WORKER_PROMPT
    assert "RetrievalPipeline" in EVIDENCE_RETRIEVAL_WORKER_PROMPT
    assert "不由模型选择" in EVIDENCE_RETRIEVAL_WORKER_PROMPT

    register_script_presets()
    assert (
        get_preset("script_evidence_retrieval_worker").system_prompt
        == EVIDENCE_RETRIEVAL_WORKER_PROMPT
    )


def test_creative_prompts_preserve_quality_and_convergence_contracts() -> None:
    assert "前期可探索不同的真实路径" in PLANNER_PROMPT
    assert "只处理最主要的缺陷簇" in PLANNER_PROMPT
    assert "受众可感知结果" in DIRECTION_WORKER_PROMPT
    assert "不预写 Phase 2" in DIRECTION_WORKER_PROMPT
    assert "段落方案" in DIRECTION_WORKER_PROMPT
    assert "不是完成的正文" in STRUCTURE_WORKER_PROMPT
    assert "不可随意互换" in STRUCTURE_WORKER_PROMPT
    assert "受众最终能直接看到" in PARAGRAPH_WORKER_PROMPT
    assert "不是提纲、创作说明" in PARAGRAPH_WORKER_PROMPT
    assert "只是抽象评价" in ELEMENT_SET_WORKER_PROMPT
    assert "先按完全相同的 criteria 独立评价" in COMPARISON_WORKER_PROMPT
    assert "不在 Compose 阶段补写或润色" in COMPOSE_WORKER_PROMPT


def test_validator_prompts_require_audience_visible_semantic_evidence() -> None:
    dimensions = (
        "D1 Goal 实现",
        "D2 具体可执行",
        "D3 结构叙事",
        "D4 元素一致",
        "D5 主题与 persona",
        "D6 信息与事实",
    )
    for dimension in dimensions:
        assert dimension in SCRIPT_CANDIDATE_VALIDATOR_PROMPT
    assert "标签、自述和 Artifact URI 本身都不能证明内容质量" in (
        SCRIPT_CANDIDATE_VALIDATOR_PROMPT
    )
    assert "真实受众的阅读顺序" in ROOT_VALIDATOR_PROMPT
    assert "空 evidence、空正文、零独立 Element 或零 Link" in ROOT_VALIDATOR_PROMPT