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。
89 lines
4.3 KiB
Python
89 lines
4.3 KiB
Python
#!/usr/bin/env python3
|
|
"""技能 2 前置预防离线测试：负约束只取 active 规则、声音账投影、串作品拒绝。"""
|
|
import pathlib
|
|
import sys
|
|
import unittest
|
|
from unittest.mock import patch
|
|
|
|
PROJECT_ROOT = pathlib.Path(__file__).resolve().parents[3]
|
|
SCRIPT_DIR = PROJECT_ROOT / ".claude" / "skills" / "prevent-ai-flavor" / "scripts"
|
|
sys.path.insert(0, str(SCRIPT_DIR))
|
|
|
|
import prevent_ai_flavor as prev # noqa: E402
|
|
|
|
LEDGER = {
|
|
"schema_version": "voice-baseline-v1",
|
|
"work_ref": "synthetic:demo",
|
|
"narrator": {"sentence_habits": ["短句收束"], "punctuation_habits": ["少用感叹号"]},
|
|
"characters": {"老周": {"verbal_tics": ["我说小子"], "sample_lines": ["我说小子，茶要凉了。"]}},
|
|
"untouchable_verbal_tics": {"老周": ["我说小子"]},
|
|
"protected_spans": ["青云城的雨说来就来"],
|
|
"blacklist": ["值得注意的是"],
|
|
}
|
|
|
|
|
|
class PreventionContractTest(unittest.TestCase):
|
|
def test_contract_without_ledger_still_valid(self):
|
|
contract = prev.build_prevention_contract("synthetic:demo")
|
|
self.assertEqual(contract["schema_version"], "ai-flavor-prevention-v2")
|
|
self.assertGreaterEqual(len(contract["negative_constraints"]), 10)
|
|
self.assertEqual(contract["positive_samples"], [])
|
|
self.assertIsNone(contract["built_from"]["voice_ledger_sha256"])
|
|
for item in contract["negative_constraints"]:
|
|
self.assertTrue(item["avoid_examples"], item["rule_id"])
|
|
self.assertIn("keep_examples", item)
|
|
self.assertIn("boundary_examples", item)
|
|
self.assertIn("regression_traps", item)
|
|
|
|
def test_ledger_projects_positive_samples_and_blacklist(self):
|
|
contract = prev.build_prevention_contract("synthetic:demo", voice_ledger=LEDGER)
|
|
kinds = {(p["kind"], p["text"]) for p in contract["positive_samples"]}
|
|
self.assertIn(("verbal_tic", "我说小子"), kinds)
|
|
self.assertIn(("voice_sample", "我说小子，茶要凉了。"), kinds)
|
|
self.assertIn(("narrator_habit", "短句收束"), kinds)
|
|
self.assertEqual(contract["blacklist"], ["值得注意的是"])
|
|
self.assertEqual(contract["protected_spans"], ["青云城的雨说来就来"])
|
|
self.assertTrue(any("黑名单表达式" in item for item in prev.render_writer_constraints(contract)))
|
|
self.assertIsNotNone(contract["built_from"]["voice_ledger_sha256"])
|
|
|
|
def test_work_ref_mismatch_is_rejected(self):
|
|
with self.assertRaisesRegex(prev.PreventionContractError, "不一致"):
|
|
prev.build_prevention_contract("synthetic:other", voice_ledger=LEDGER)
|
|
|
|
def test_candidate_ledger_is_rejected_even_in_offline_projection(self):
|
|
bad = dict(LEDGER, status="candidate")
|
|
with self.assertRaisesRegex(prev.PreventionContractError, "canonical"):
|
|
prev.build_prevention_contract("synthetic:demo", voice_ledger=bad)
|
|
|
|
def test_bad_schema_version_is_rejected(self):
|
|
bad = dict(LEDGER, schema_version="nope")
|
|
with self.assertRaisesRegex(prev.PreventionContractError, "schema_version"):
|
|
prev.build_prevention_contract("synthetic:demo", voice_ledger=bad)
|
|
|
|
def test_cli_default_persists_run(self):
|
|
import argparse # noqa: F401 (确认 CLI 依赖可导入)
|
|
import tempfile
|
|
import json
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
output = pathlib.Path(tmp) / "contract.json"
|
|
with patch.object(prev, "persist_prevention", return_value={"run_id": "prev-x"}) as persist, \
|
|
patch.object(prev, "_load_db_ledger", return_value=None):
|
|
code = prev.main(["--work-ref", "synthetic:demo", "--output", str(output)])
|
|
self.assertEqual(code, 0)
|
|
persist.assert_called_once()
|
|
contract = json.loads(output.read_text(encoding="utf-8"))
|
|
self.assertEqual(contract["work_ref"], "synthetic:demo")
|
|
|
|
def test_cli_offline_skips_db(self):
|
|
import tempfile
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
output = pathlib.Path(tmp) / "contract.json"
|
|
with patch.object(prev, "persist_prevention") as persist:
|
|
code = prev.main(["--work-ref", "synthetic:demo", "--output", str(output), "--offline"])
|
|
self.assertEqual(code, 0)
|
|
persist.assert_not_called()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|