c9f69d9d6d
范围（不含 design-story-foundation、docs/、humanization/README.md 等进行中改动）:
1. 新增 harness/ 控制平面
- skill_harness.py 静态审计：32 个运行时 Skill 的 frontmatter/manifest/文档污染，当前 0 问题
- run_selected.py 选择性执行器：manifest 与磁盘一一对账、依赖阻断、
空跑与 skip-only 失败关闭、AST 测试形状门
- manifests/skills.json：32 个 Skill 的合同责任方与协作领域登记
- manifests/test-inventory.json：81 个测试资产登记
- specs/skill-testing.md 与 README.md：测试分层、证据边界与 harness 职责
2. 实现测试从 .claude/skills/*/scripts/ 迁至 tests/skills/<skill>/
- 71 个测试文件迁移并修复项目根与临时目录运行导入
- 数据库触发器测试宽泛异常收窄为 psycopg.errors.RaiseException
- 抽取离线大测试拆出真实 PG smoke（默认阻断，不计入离线通过）
- 抽取 presence 去重边界拆出独立测试：493 + 78 = 571 项检查不变
3. 运行时文档清理
- 13 个 SKILL.md 移除自测/离线验证段落、测试命令与测试文件事实源表述，
只保留运行时合同；业务运行合同、额度、授权与离线模式均保留
4. SoT 同步
- AGENTS.md：新增 Skill 领域索引（7 个合同责任方分组，覆盖 32 个运行时 Skill）
- 领域 07：测试入口改由 harness/manifests/ 登记，SKILL.md 不承载测试命令
- humanization 覆盖矩阵：活动测试路径同步迁移
验证证据: harness 自测 15 项 + runner 自测 13 项通过；静态审计 32 Skill / 0 问题；
73 个非数据库测试通过；8 个集成条目中 6 个 PostgreSQL 项被依赖门明确阻断；
py_compile 与 git diff --check 通过。未连接 PostgreSQL、网络、真实模型或额度。
已知边界: 真正 skill_behavior_eval 仍为 0，尚未验证任何 Skill 自然语言行为；
evaluate-frozen-replay 的 raw 存储边界冲突留待单独治理。
99 lines
3.6 KiB
Python
99 lines
3.6 KiB
Python
#!/usr/bin/env python3
|
||
"""parse_outline 窗覆盖门离线测试：不连库、不调用模型。"""
|
||
|
||
import pathlib
|
||
import sys
|
||
|
||
import click
|
||
|
||
PROJECT_ROOT = pathlib.Path(__file__).resolve().parents[3]
|
||
SCRIPT_DIR = PROJECT_ROOT / ".claude" / "skills" / "deconstruct-book" / "scripts"
|
||
sys.path.insert(0, str(SCRIPT_DIR))
|
||
|
||
import parse_outline as po # noqa: E402
|
||
|
||
|
||
def check(name, condition):
|
||
"""输出与现有 parse-book 离线测试一致的简明结果。"""
|
||
|
||
if not condition:
|
||
raise AssertionError(name)
|
||
print(f"[PASS] {name}")
|
||
|
||
|
||
def main():
|
||
"""覆盖连续、头部、中部、尾部和重叠窗五类边界。"""
|
||
|
||
check("coverage-连续无缝", po.outline_coverage_gaps([(1, 33), (34, 65)], 1, 65) == [])
|
||
check(
|
||
"coverage-识别中部缝隙",
|
||
po.outline_coverage_gaps([(951, 983), (1015, 1047)], 951, 1047) == [(984, 1014)],
|
||
)
|
||
check("coverage-识别头部缝隙", po.outline_coverage_gaps([(5, 10)], 1, 10) == [(1, 4)])
|
||
check("coverage-识别尾部缝隙", po.outline_coverage_gaps([(1, 6)], 1, 10) == [(7, 10)])
|
||
check(
|
||
"coverage-重叠窗不误报",
|
||
po.outline_coverage_gaps([(1, 20), (15, 30), (31, 40)], 1, 40) == [],
|
||
)
|
||
po.ensure_outline_coverage([(1, 33), (34, 65)], 1, 65, context="测试")
|
||
try:
|
||
po.ensure_outline_coverage([(951, 983), (1015, 1047)], 951, 1047, context="测试")
|
||
except click.ClickException as error:
|
||
check("coverage-缝隙非零阻断", "984-1014" in str(error))
|
||
else:
|
||
raise AssertionError("coverage-缝隙非零阻断")
|
||
|
||
original_chat = po.chat_degrade
|
||
try:
|
||
replies = iter([("{}", {}), ('{"outline":"补窗成功"}', {"prompt_tokens": 1})])
|
||
po.chat_degrade = lambda _prompt, _model: next(replies)
|
||
outline, usage = po.generate_outline_with_retry("prompt", "model", max_attempts=2)
|
||
check("window-空输出重试后成功", outline == "补窗成功" and usage["prompt_tokens"] == 1)
|
||
|
||
calls = {"count": 0}
|
||
|
||
def always_empty(_prompt, _model):
|
||
calls["count"] += 1
|
||
return "{}", {}
|
||
|
||
po.chat_degrade = always_empty
|
||
try:
|
||
po.generate_outline_with_retry("prompt", "model", max_attempts=2)
|
||
except click.ClickException as error:
|
||
check("window-有限重试耗尽即停书", calls["count"] == 2 and "2 次" in str(error))
|
||
else:
|
||
raise AssertionError("window-有限重试耗尽即停书")
|
||
|
||
def exhausted_chain(_prompt, _model):
|
||
raise po.SensitiveError("全部渠道耗尽")
|
||
|
||
po.chat_degrade = exhausted_chain
|
||
try:
|
||
po.generate_outline_with_retry("prompt", "model", max_attempts=2)
|
||
except click.ClickException as error:
|
||
check("window-治理链耗尽即停书", "治理链耗尽" in str(error))
|
||
else:
|
||
raise AssertionError("window-治理链耗尽即停书")
|
||
finally:
|
||
po.chat_degrade = original_chat
|
||
|
||
unordered = [
|
||
(3, 83, 984, 1014, "补插窗", "revised"),
|
||
(1, 1, 1, 33, "首窗", "revised"),
|
||
(4, 52, 1015, 1047, "后续窗", "revised"),
|
||
(2, 82, 2226, 2250, "末窗", "revised"),
|
||
]
|
||
ordered = po.order_outline_windows(unordered)
|
||
check("check-按绝对章号排序", [row[2] for row in ordered] == [1, 984, 1015, 2226])
|
||
rendered = po.render_coherence_outlines(ordered)
|
||
check(
|
||
"check-阶段号按章序重编",
|
||
"【阶段2（第984–1014章）】" in rendered and "【阶段83" not in rendered,
|
||
)
|
||
|
||
print("\n全部离线自测通过：11 项（未连库、未发任何 LLM 调用）")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|