tests/test_skill.py

import json
import os
import re
import subprocess
import sys
import tempfile
import unittest
from pathlib import Path


REPO = Path(__file__).resolve().parents[1]
SKILL = REPO / "skills" / "speak-human"
CHECK = SKILL / "scripts" / "check.py"
RULES = SKILL / "references" / "core-rules.json"
EVALS = REPO / "evals" / "trigger_eval.jsonl"
PLUGIN = REPO / ".claude-plugin" / "plugin.json"
VERSION = "2.1.1"
PACKAGE_FILES = {
    "LICENSE.txt",
    "SKILL.md",
    "SKILL.zh-CN.md",
    "agents/openai.yaml",
    "assets/review-output-template.md",
    "references/core-method.md",
    "references/communication-modes.md",
    "references/language-signals.md",
    "references/social-media.md",
    "references/core-rules.json",
    "references/voice-profile.schema.json",
    "scripts/check.py",
}
MODES = {
    "general",
    "status_update",
    "technical_explanation",
    "decision_recommendation",
    "professional_document",
    "readme_release",
    "resume_portfolio",
    "conversation_reply",
}
PRIVATE_PLACEHOLDERS = re.compile(
    r"\[(?:INTERNAL_EMPLOYEE|PRIVATE_CORPUS|LOCAL_USERNAME)\]"
)
SOCIAL_CONTEXT = re.compile(
    r"社媒|小红书|微博|抖音|social media|X post|Reddit|Instagram|TikTok",
    re.IGNORECASE,
)
POSITIVE_ACTION = re.compile(
    r"\$speak-human|speak-human|改写|重写|润色|改得|改成|压短|整理成|帮我回复|"
    r"写得.{0,10}(?:清楚|自然|像人)|解释.{0,20}(?:给|让)|"
    r"rewrite|review|make .{0,50}(?:clearer|easier)|turn .{0,50}into",
    re.IGNORECASE,
)
HUMAN_WRITING = re.compile(
    r"进度|状态更新|汇报|解释|OAuth|技术|decision memo|recommendation|"
    r"leadership|客户|实施方案|email|vendor|README|release note|"
    r"简历|portfolio|同事|消息|公告|小红书|项目复盘|报告|文档",
    re.IGNORECASE,
)
EXPLICIT_NON_TRIGGER = re.compile(
    r"不要改写|不需要润色|逐字翻译|do not draft|do not edit|"
    r"只报告|只给我|source-code|off-by-one|CSV|JSON|柱状图|头像|"
    r"重复 PDF|发布时间|不做风格改写",
    re.IGNORECASE,
)


def package_files() -> set[str]:
    return {
        path.relative_to(SKILL).as_posix()
        for path in SKILL.rglob("*")
        if path.is_file() and "__pycache__" not in path.parts
    }


def parse_frontmatter(path: Path) -> dict[str, str]:
    text = path.read_text(encoding="utf-8")
    match = re.match(r"(?s)^---\n(.*?)\n---\n", text)
    if not match:
        raise AssertionError(f"missing frontmatter: {path}")
    fields = {}
    for line in match.group(1).splitlines():
        if ":" in line:
            key, value = line.split(":", 1)
            fields[key.strip()] = value.strip()
    return fields


def run_check(*args: str) -> subprocess.CompletedProcess[str]:
    return subprocess.run(
        [sys.executable, str(CHECK), *args],
        text=True,
        encoding="utf-8",
        env={**os.environ, "PYTHONIOENCODING": "utf-8"},
        capture_output=True,
        check=False,
    )


def finding_kinds(result: subprocess.CompletedProcess[str]) -> set[str]:
    payload = json.loads(result.stdout)
    return {row["kind"] for row in payload["findings"]}


def reference_trigger(prompt: str) -> str:
    if "$speak-human" in prompt or "speak-human" in prompt:
        return "TRIGGER"
    if EXPLICIT_NON_TRIGGER.search(prompt):
        return "NO_TRIGGER"
    return (
        "TRIGGER"
        if POSITIVE_ACTION.search(prompt) and HUMAN_WRITING.search(prompt)
        else "NO_TRIGGER"
    )


class SpeakHumanContractTests(unittest.TestCase):
    def test_package_layout_and_frontmatter(self):
        self.assertEqual(package_files(), PACKAGE_FILES)
        for filename in ("SKILL.md", "SKILL.zh-CN.md"):
            fields = parse_frontmatter(SKILL / filename)
            self.assertEqual(set(fields), {"name", "description"})
            self.assertEqual(fields["name"], "speak-human")

    def test_version_and_core_mode_contract(self):
        payload = json.loads(RULES.read_text(encoding="utf-8"))
        plugin = json.loads(PLUGIN.read_text(encoding="utf-8"))
        self.assertEqual(payload["standard_version"], VERSION)
        self.assertEqual(payload["package_version"], VERSION)
        self.assertEqual(plugin["version"], VERSION)
        self.assertEqual(plugin["name"], "speak-human")
        self.assertIn("human-facing communication", plugin["description"].lower())
        self.assertNotIn("public copy", plugin["description"].lower())
        self.assertEqual(set(payload["modes"]), MODES)
        self.assertNotIn("scenes", payload)

    def test_social_media_is_an_optional_adapter(self):
        payload = json.loads(RULES.read_text(encoding="utf-8"))
        self.assertTrue(
            all(rule["domain"] == "GENERAL_CORE" for rule in payload["rules"])
        )
        self.assertNotIn("social_media", payload["modes"])
        self.assertTrue((SKILL / "references" / "social-media.md").exists())
        helper = CHECK.read_text(encoding="utf-8")
        for token in ("X 已爆", "小红书", "information_gap_diagnostic"):
            self.assertNotIn(token, helper)

    def test_openai_metadata_describes_general_communication(self):
        metadata = (SKILL / "agents" / "openai.yaml").read_text(encoding="utf-8")
        lines = [line.rstrip() for line in metadata.splitlines() if line.strip()]
        self.assertEqual(lines[0], "interface:")
        self.assertEqual(
            {line.strip().split(":", 1)[0] for line in lines[1:]},
            {"display_name", "short_description", "brand_color", "default_prompt"},
        )
        self.assertTrue(all(line.startswith("  ") for line in lines[1:]))
        self.assertIn('  display_name: "Speak Human"', lines)
        self.assertIn("$speak-human", metadata)
        self.assertRegex(metadata.lower(), r"communication|沟通")

    def test_bilingual_skill_keeps_core_contract(self):
        english = (SKILL / "SKILL.md").read_text(encoding="utf-8")
        chinese = (SKILL / "SKILL.zh-CN.md").read_text(encoding="utf-8")
        for token in (
            "status updates",
            "explanations",
            "recommendations",
            "emails",
            "README",
            "resumes",
            "replies",
            "Social media is an adapter, not the default.",
            "private | internal | external | public",
            "scripts/check.py",
            VERSION,
            "silent obligations",
            "Do not manufacture a lesson",
        ):
            self.assertIn(token, english)
        for token in (
            "进度汇报",
            "解释说明",
            "建议与决策",
            "邮件",
            "README",
            "简历",
            "回复",
            "社媒只是适配层，不是默认场景。",
            "private | internal | external | public",
            "scripts/check.py",
            VERSION,
            "内部要遵守的条件",
            "不要再强行补教训",
        ):
            self.assertIn(token, chinese)

    def test_all_public_version_surfaces_match(self):
        surfaces = {
            ".claude-plugin/plugin.json": PLUGIN.read_text(encoding="utf-8"),
            "CHANGELOG.md": (REPO / "CHANGELOG.md").read_text(encoding="utf-8"),
            "README.md": (REPO / "README.md").read_text(encoding="utf-8"),
            "README.zh-CN.md": (REPO / "README.zh-CN.md").read_text(
                encoding="utf-8"
            ),
            "skills/speak-human/SKILL.md": (SKILL / "SKILL.md").read_text(
                encoding="utf-8"
            ),
            "skills/speak-human/SKILL.zh-CN.md": (
                SKILL / "SKILL.zh-CN.md"
            ).read_text(encoding="utf-8"),
            "skills/speak-human/references/core-rules.json": RULES.read_text(
                encoding="utf-8"
            ),
        }
        for name, text in surfaces.items():
            self.assertIn(VERSION, text, name)

    def test_readme_relative_links_resolve(self):
        for name in ("README.md", "README.zh-CN.md", "中文用户看这里.md"):
            text = (REPO / name).read_text(encoding="utf-8")
            for target in re.findall(r"\]\(([^)]+)\)", text):
                if target.startswith(("http://", "https://", "#")):
                    continue
                path_only = target.split("#", 1)[0]
                self.assertTrue((REPO / path_only).exists(), f"{name}: {target}")

    def test_repository_contains_no_private_material(self):
        old_names = ("human" + "-voice-zh", "Human" + " Voice ZH")
        hits = []
        for path in REPO.rglob("*"):
            if (
                not path.is_file()
                or ".git" in path.parts
                or "__pycache__" in path.parts
                or "tests" in path.parts
                or path.name in {"LICENSE", "LICENSE.txt"}
            ):
                continue
            text = path.read_text(encoding="utf-8")
            for old_name in old_names:
                if old_name in text:
                    hits.append((path.relative_to(REPO).as_posix(), "old_name"))
            if PRIVATE_PLACEHOLDERS.search(text):
                hits.append(
                    (path.relative_to(REPO).as_posix(), "private_placeholder")
                )
        self.assertEqual(hits, [])

    def test_check_local_paths_follow_exposure(self):
        text = r"Build log: C:\Users\someone\project\output.txt"
        for exposure in ("external", "public"):
            result = run_check("--exposure", exposure, "--text", text)
            self.assertEqual(result.returncode, 2, result.stderr)
            self.assertIn("local_path", finding_kinds(result))

        internal = run_check("--exposure", "internal", "--text", text)
        self.assertEqual(internal.returncode, 0, internal.stderr)
        payload = json.loads(internal.stdout)
        self.assertEqual(payload["exposure"], "internal")
        self.assertEqual(payload["findings"], [])

    def test_check_unc_paths_follow_exposure(self):
        text = r"Build log: \\corp-server\private\draft.md"
        for exposure in ("external", "public"):
            result = run_check("--exposure", exposure, "--text", text)
            self.assertEqual(result.returncode, 2, result.stderr)
            self.assertIn("unc_path", finding_kinds(result))
            self.assertTrue(
                any(
                    row["kind"] == "unc_path" and row["severity"] == "P0"
                    for row in json.loads(result.stdout)["findings"]
                )
            )

        for exposure in ("private", "internal"):
            result = run_check("--exposure", exposure, "--text", text)
            self.assertEqual(result.returncode, 0, result.stderr)
            self.assertEqual(json.loads(result.stdout)["findings"], [])

    def test_check_credentials_are_always_flagged(self):
        samples = (
            ('api_token="abcdefghijk"', "credential_assignment", "abcdefghijk"),
            ("api_token='zyxwvutsrqp'", "credential_assignment", "zyxwvutsrqp"),
            (
                "OPENAI_API_KEY='sk-proj_abcdefghijklmnopqrstuvwxyz123456'",
                "credential_assignment",
                "sk-proj_abcdefghijklmnopqrstuvwxyz123456",
            ),
            (
                "refresh_token=rt_abcdefghijklmnopqrstuvwxyz123456",
                "credential_assignment",
                "rt_abcdefghijklmnopqrstuvwxyz123456",
            ),
            (
                "password='Tr0ub4dor!42'",
                "credential_assignment",
                "Tr0ub4dor!42",
            ),
            (
                "Authorization: Bearer ghp_abcdefghijklmnopqrstuvwxyz123456",
                "credential_bearer",
                "ghp_abcdefghijklmnopqrstuvwxyz123456",
            ),
        )
        for text, expected_kind, secret in samples:
            for exposure in ("private", "internal", "external", "public"):
                result = run_check("--exposure", exposure, "--text", text)
                self.assertEqual(result.returncode, 2, result.stderr)
                self.assertIn(expected_kind, finding_kinds(result))
                self.assertNotIn(secret, result.stdout)
                self.assertTrue(
                    any(
                        row["kind"] == expected_kind
                        and row["severity"] == "P0"
                        and row["preview"].endswith("[REDACTED]")
                        for row in json.loads(result.stdout)["findings"]
                    )
                )

    def test_check_credential_discussion_is_not_flagged(self):
        samples = (
            "The API token field accepts at least eight characters.",
            "Set the Authorization header to use the Bearer scheme.",
            "Never paste a password or secret into a public post.",
            "token: authentication",
        )
        for text in samples:
            result = run_check("--exposure", "public", "--text", text)
            self.assertEqual(result.returncode, 0, result.stderr)
            self.assertEqual(json.loads(result.stdout)["findings"], [])

    def test_check_internal_markers_follow_exposure(self):
        text = "【待校正】发布完成"
        for exposure in ("external", "public"):
            result = run_check("--exposure", exposure, "--text", text)
            self.assertEqual(result.returncode, 2, result.stderr)
            self.assertIn("internal_marker", finding_kinds(result))

        internal = run_check("--exposure", "internal", "--text", text)
        self.assertEqual(internal.returncode, 0, internal.stderr)
        self.assertEqual(json.loads(internal.stdout)["findings"], [])

    def test_check_file_input_and_voice_profile(self):
        with tempfile.TemporaryDirectory() as temp_dir:
            root = Path(temp_dir)
            source = root / "draft.txt"
            source.write_text("No sensitive material here.", encoding="utf-8")
            profile = root / "profile.json"
            profile.write_text(
                json.dumps(
                    {
                        "profile_version": "1.0.0",
                        "languages": ["en"],
                        "global_prefer": ["lead with the conclusion"],
                    }
                ),
                encoding="utf-8",
            )
            result = run_check(
                "--exposure",
                "internal",
                "--file",
                str(source),
                "--voice-profile",
                str(profile),
            )
            self.assertEqual(result.returncode, 0, result.stderr)
            payload = json.loads(result.stdout)
            self.assertEqual(payload["exposure"], "internal")
            self.assertTrue(payload["voice_profile_loaded"])
            self.assertEqual(payload["findings"], [])

            invalid = root / "invalid-profile.json"
            invalid.write_text(
                json.dumps(
                    {
                        "profile_version": "1.0.0",
                        "languages": ["en"],
                        "mode_preferences": {
                            "status_update": {"prefer": "not-a-list"}
                        },
                    }
                ),
                encoding="utf-8",
            )
            rejected = run_check(
                "--exposure",
                "internal",
                "--text",
                "No sensitive material here.",
                "--voice-profile",
                str(invalid),
            )
            self.assertEqual(rejected.returncode, 1)
            self.assertIn("must be a string list", rejected.stderr)

            duplicate_languages = root / "duplicate-languages.json"
            duplicate_languages.write_text(
                json.dumps(
                    {
                        "profile_version": "1.0.0",
                        "languages": ["en", "en"],
                    }
                ),
                encoding="utf-8",
            )
            rejected_duplicate = run_check(
                "--exposure",
                "internal",
                "--text",
                "No sensitive material here.",
                "--voice-profile",
                str(duplicate_languages),
            )
            self.assertEqual(rejected_duplicate.returncode, 1)
            self.assertIn("must not contain duplicates", rejected_duplicate.stderr)

    def test_trigger_and_non_trigger_fixtures(self):
        rows = [
            json.loads(line)
            for line in EVALS.read_text(encoding="utf-8").splitlines()
            if line.strip()
        ]
        self.assertEqual(len(rows), 24)
        self.assertEqual(len({row["id"] for row in rows}), 24)
        self.assertEqual(
            sum(row["expected"] == "TRIGGER" for row in rows),
            12,
        )
        self.assertEqual(
            sum(row["expected"] == "NO_TRIGGER" for row in rows),
            12,
        )
        positive_rows = [row for row in rows if row["expected"] == "TRIGGER"]
        self.assertLessEqual(
            sum(bool(SOCIAL_CONTEXT.search(row["prompt"])) for row in positive_rows),
            3,
            "social media must remain an optional adapter, not the core trigger surface",
        )
        for row in rows:
            self.assertEqual(
                reference_trigger(row["prompt"]),
                row["expected"],
                row["id"],
            )

    def test_published_package_contains_no_private_placeholders(self):
        hits = []
        for path in SKILL.rglob("*"):
            if not path.is_file() or path.name == "LICENSE.txt":
                continue
            text = path.read_text(encoding="utf-8")
            if PRIVATE_PLACEHOLDERS.search(text):
                hits.append(path.relative_to(SKILL).as_posix())
        self.assertEqual(hits, [])


if __name__ == "__main__":
    unittest.main()