master-skill 0.10.1 → 0.11.0

package/routing.json

@@ -0,0 +1,209 @@
+{
+  "version": 1,
+  "_comment": "Machine-readable routing table. Persona keywords are NOT duplicated here — they live in each prebuilt/<slug>/meta.json search_scope.keywords and are read at match time. This file holds only what has no other machine-readable home: teaching-mode short-circuit order, the topic->pairing fallback, and the scoring constants. scripts/validate-routing.py enforces that every keyword set is pairwise disjoint (no string appears twice, and no string is a substring of another row's string), so a match can never depend on iteration order.",
+  "weights": {
+    "keyword_hit": 3
+  },
+  "_min_keyword_length_comment": "Persona keywords shorter than this are ignored when scoring. Seven single-character doctrinal atoms (空 戒 定 慧 苦 禅 业) live in meta.json search_scope.keywords, and matching by containment made them fire on ordinary Chinese: 「有空吗」 scored Kumārajīva and Nāgārjuna, 「我戒烟了」 scored Buddhaghosa. Every one of them also appears inside a longer keyword (空性 / 持戒 / 戒定慧 / 毕竟空 …), so dropping them costs no real recall; queries that only carry the bare atom fall through to topic_pairings, which is what that layer is for.",
+  "min_keyword_length": 2,
+  "mode_rules": [
+    {
+      "mode": "master-curriculum",
+      "order": 1,
+      "note": "纵向 / 时序：请求学修路径，不是比较",
+      "keywords": [
+        "学习计划",
+        "学修次第",
+        "入门",
+        "先学什么",
+        "从哪开始",
+        "开始学",
+        "应该读",
+        "下一步读什么",
+        "路径推荐",
+        "按什么顺序",
+        "curriculum",
+        "roadmap"
+      ]
+    },
+    {
+      "mode": "master-debate",
+      "order": 2,
+      "note": "横向 / 对立：4 轮立论-反驳-回应-综合",
+      "keywords": [
+        "辩论",
+        "各执一词",
+        "谁更对",
+        "高下",
+        "之争",
+        "之辩",
+        "分判",
+        "debate"
+      ]
+    },
+    {
+      "mode": "compare-masters",
+      "order": 3,
+      "note": "横向 / 并列：单轮多祖师平铺对比",
+      "keywords": [
+        "对比",
+        "比较",
+        "不同",
+        "各派怎么看",
+        "各位祖师",
+        "多个角度",
+        "异同",
+        "compare"
+      ]
+    }
+  ],
+  "_situations_comment": "Vernacular felt-state routing, consulted after persona keyword scoring and before topic_pairings. search_scope.keywords are doctrinal retrieval terms; a beginner does not type 四念处, they type 坐不住. Empirically, 4 of the 11 rows in the README 「你的状况」 table resolved to default_pairing because no doctrinal keyword matched — those 4 are the rows below. The other 7 already route correctly off persona keywords and are deliberately NOT duplicated here: this layer holds only what the keyword layer cannot reach.",
+  "situations": [
+    {
+      "id": "scattered-mind",
+      "keywords": ["妄念", "杂念", "坐不住", "静不下来", "定不下来", "心乱"],
+      "masters": ["master-xuyun", "master-zhiyi", "master-ajahn-chah"],
+      "note": "参话头 / 止观 / 正念观察"
+    },
+    {
+      "id": "texts-opaque",
+      "keywords": ["看不懂", "读不懂", "理不清", "没有逻辑"],
+      "masters": ["master-xuanzang"],
+      "note": "唯识严密分析"
+    },
+    {
+      "id": "no-traction",
+      "keywords": ["无力感", "使不上力", "没有进步", "学佛很久", "提不起劲"],
+      "masters": ["master-yinguang"],
+      "note": "老实念佛"
+    },
+    {
+      "id": "plainest-practice",
+      "keywords": ["最朴素", "朴素", "最简单的修法"],
+      "masters": ["master-ajahn-chah"],
+      "note": "南传森林禅 · 出入息念"
+    }
+  ],
+  "topic_pairings": [
+    {
+      "id": "pureland",
+      "keywords": ["念佛", "往生", "净土"],
+      "masters": ["master-yinguang", "master-ouyi"],
+      "note": "净土专精 + 跨宗派"
+    },
+    {
+      "id": "chan-practice",
+      "keywords": ["参禅", "话头", "开悟"],
+      "masters": ["master-huineng", "master-xuyun"],
+      "note": "古今禅宗对比"
+    },
+    {
+      "id": "yogacara-vs-madhyamaka",
+      "keywords": ["唯识", "空有", "性相", "法相"],
+      "masters": ["master-xuanzang", "master-kumarajiva"],
+      "note": "唯识 vs 中观"
+    },
+    {
+      "id": "panjiao",
+      "keywords": ["判教", "圆融", "止观"],
+      "masters": ["master-zhiyi", "master-fazang"],
+      "note": "天台 vs 华严"
+    },
+    {
+      "id": "integrated-path",
+      "keywords": ["修行次第", "综合法门"],
+      "masters": ["master-ouyi", "master-yinguang"],
+      "note": "综合 vs 专修"
+    },
+    {
+      "id": "vinaya",
+      "keywords": ["戒律", "持戒", "律仪", "行持"],
+      "masters": ["master-xuyun", "master-atisha", "master-buddhaghosa"],
+      "note": "汉传禅门律 vs 印藏菩萨律 vs 上座部律藏注释（合并原「戒律/行持/日常」与「戒律/持戒/律仪」两行，去除键冲突）"
+    },
+    {
+      "id": "emptiness",
+      "keywords": ["般若", "空性", "中观", "缘起性空", "应成", "毕竟空"],
+      "masters": ["master-kumarajiva", "master-tsongkhapa", "master-huineng"],
+      "note": "早期中观译师 · 应成中观精确分判 · 禅宗直指（合并原「般若/空性」与「中观/缘起性空/应成」两行）"
+    },
+    {
+      "id": "lamrim",
+      "keywords": ["道次第", "三士道", "下士道", "中士道", "上士道", "lam rim"],
+      "masters": ["master-atisha", "master-tsongkhapa"],
+      "note": "印藏桥梁源头 vs 格鲁派系统化"
+    },
+    {
+      "id": "consciousness",
+      "keywords": ["心识", "阿赖耶", "心所", "末那"],
+      "masters": ["master-xuanzang", "master-buddhaghosa", "master-huineng"],
+      "note": "唯识 vs 上座部阿毗达摩 vs 禅宗直指"
+    },
+    {
+      "id": "ascetic",
+      "keywords": ["苦行", "闭关", "山中修行", "头陀"],
+      "masters": ["master-xuyun", "master-milarepa"],
+      "note": "汉传禅门头陀 vs 藏传瑜伽士传统"
+    },
+    {
+      "id": "mindfulness",
+      "keywords": ["正念", "观心", "觉知"],
+      "masters": ["master-huineng", "master-ajahn-chah", "master-mahasi-sayadaw"],
+      "note": "禅宗自性 vs 南传 sati 朴素 vs 缅甸标记法"
+    },
+    {
+      "id": "meditation-objects",
+      "keywords": ["禅修方法", "业处", "所缘"],
+      "masters": ["master-buddhaghosa", "master-mahasi-sayadaw", "master-ajahn-chah"],
+      "note": "论藏四十种业处 vs 标记法腹部起伏 vs 朴素观心"
+    },
+    {
+      "id": "insight-stages",
+      "keywords": ["七清净", "十六观智", "观智"],
+      "masters": ["master-buddhaghosa", "master-mahasi-sayadaw"],
+      "note": "《清净道论》原典 vs 现代缅甸应用（原行含「道次第」，已移除以避免与 lamrim 冲突）"
+    },
+    {
+      "id": "renunciation",
+      "keywords": ["出离心", "暇满", "无常"],
+      "masters": ["master-yinguang", "master-atisha", "master-ajahn-chah"],
+      "note": "净土 · 噶当 · 上座部跨传统出离观对比"
+    },
+    {
+      "id": "bodhicitta",
+      "keywords": ["菩提心", "慈悲", "自他相换"],
+      "masters": ["master-atisha", "master-ouyi"],
+      "note": "印藏自他相换 vs 跨宗派融通"
+    },
+    {
+      "id": "teacher",
+      "keywords": ["上师", "善知识", "依止"],
+      "masters": ["master-xuyun", "master-atisha", "master-tsongkhapa"],
+      "note": "汉传善知识 vs 噶当依止论 vs 格鲁视师如佛"
+    },
+    {
+      "id": "scholastic",
+      "keywords": ["论师风格", "经院严密", "因明"],
+      "masters": ["master-xuanzang", "master-tsongkhapa", "master-buddhaghosa"],
+      "note": "唯识 · 应成中观 · 上座部三大论师传统"
+    },
+    {
+      "id": "four-traditions",
+      "keywords": ["四大传统", "四方对照"],
+      "masters": [
+        "master-nagarjuna",
+        "master-huineng",
+        "master-tsongkhapa",
+        "master-buddhaghosa"
+      ],
+      "note": "印度中观 · 禅 · 应成中观 · 上座部论藏，四方系统对照"
+    },
+    {
+      "id": "cross-tradition-meditation",
+      "keywords": ["跨传统禅修", "大手印"],
+      "masters": ["master-huineng", "master-milarepa", "master-ajahn-chah"],
+      "note": "禅 · 大手印 · 森林禅，跨传统禅修法"
+    }
+  ],
+  "default_pairing": ["master-kumarajiva", "master-yinguang"]
+}

package/scripts/check-gate-liveness.py

@@ -0,0 +1,222 @@
+#!/usr/bin/env python3
+"""Assert that this repo's gates actually examined something.
+
+Three shipped defects were the same shape — a gate examined an empty set and
+reported success:
+
+  - `pytest.ini` listed `testpaths = tests` while CI passed `scripts/tests/`
+    explicitly, so neither suite ever ran the other's cases.
+  - `tests/test_voice_rules.py` globbed `prebuilt/<slug>/voice.md` when
+    voice.md lives under `references/`. The empty glob parametrized every case
+    over an empty set: nothing asserted, reported green.
+  - The fidelity smoke — a branch-protection-required check — writes
+    `{"skipped": true, "reason": "no_api_key"}` and exits 0 when the secret is
+    missing, which it always has been.
+
+None of those was a wrong assertion. Each was an assertion that never ran, and
+a passing check is indistinguishable from a check that did nothing unless
+something asserts otherwise. That is this script's whole job.
+
+Usage:
+    python3 scripts/check-gate-liveness.py            # check this repo
+    python3 scripts/check-gate-liveness.py --json     # machine-readable
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import subprocess
+import sys
+from pathlib import Path
+
+# A verdict — as opposed to a skip, an error, or a dry run.
+GRADED_STATUSES = {"PASS", "FAIL"}
+
+
+def check_every_test_file_collects(
+    test_files: list[str], collected_counts: dict[str, int]
+) -> list[str]:
+    """Every test file must contribute at least one collected test.
+
+    A file that imports cleanly and yields nothing is the voice_rules failure:
+    pytest reports success because there was nothing to fail.
+    """
+    problems = []
+    for path in sorted(test_files):
+        if collected_counts.get(path, 0) < 1:
+            problems.append(
+                f"{path} collected 0 tests — it asserts nothing but reports green "
+                "(empty glob or empty parametrize?)"
+            )
+    return problems
+
+
+def check_testpaths_cover_suites(
+    testpaths: list[str], test_dirs: list[str]
+) -> list[str]:
+    """Every directory holding tests must be reachable from a bare `pytest`."""
+    covered = set(testpaths)
+    return [
+        f"{d} holds tests but is not in pytest.ini testpaths — a bare `pytest` skips it"
+        for d in sorted(test_dirs)
+        if d not in covered
+    ]
+
+
+def check_graded_suites_graded_something(suites: list[dict]) -> list[str]:
+    """A graded fidelity suite that produced no verdict must not read as a pass.
+
+    Dry runs are exempt: grading nothing is what a dry run is for.
+    """
+    problems = []
+    for suite in suites:
+        if suite.get("mode") == "dry_run":
+            continue
+        verdicts = [
+            r for r in suite.get("results", [])
+            if str(r.get("status", "")).upper() in GRADED_STATUSES
+        ]
+        if not verdicts:
+            problems.append(
+                f"{suite.get('master', '?')}: graded suite produced 0 verdicts "
+                "— it graded nothing (missing API key, or every call errored)"
+            )
+    return problems
+
+
+def check_catalog_matches_filesystem(
+    catalog: dict, prebuilt_dir: Path, root: Path
+) -> list[str]:
+    """The catalog and `prebuilt/` must name the same set of skills."""
+    entries = catalog.get("skills", [])
+    if not entries:
+        problems = ["skill-catalog.json lists no skills — an empty catalog validates vacuously"]
+        return problems
+
+    problems = []
+    catalog_sources = set()
+    for entry in entries:
+        source = entry.get("source", "")
+        catalog_sources.add(source)
+        if source.startswith("prebuilt/") and not (root / source).is_dir():
+            problems.append(
+                f"{entry.get('name', '?')}: catalog points at {source}, which does not exist"
+            )
+
+    if prebuilt_dir.is_dir():
+        for d in sorted(p for p in prebuilt_dir.iterdir() if p.is_dir()):
+            rel = f"prebuilt/{d.name}"
+            if rel not in catalog_sources:
+                problems.append(
+                    f"{d.name}: directory exists under prebuilt/ but no catalog entry "
+                    "claims it — it ships to nobody and no gate examines it"
+                )
+    return problems
+
+
+def check_every_skill_has_fixtures(prebuilt_dir: Path) -> list[str]:
+    """Every prebuilt skill must carry at least one fidelity fixture."""
+    if not prebuilt_dir.is_dir():
+        return [f"{prebuilt_dir} does not exist — nothing to examine"]
+
+    problems = []
+    for d in sorted(p for p in prebuilt_dir.iterdir() if p.is_dir()):
+        fixtures = d / "tests" / "fidelity.jsonl"
+        if not fixtures.exists():
+            problems.append(f"{d.name}: no tests/fidelity.jsonl — nothing grades this skill")
+            continue
+        lines = [ln for ln in fixtures.read_text(encoding="utf-8").splitlines() if ln.strip()]
+        if not lines:
+            problems.append(f"{d.name}: tests/fidelity.jsonl is empty — it grades 0 cases")
+    return problems
+
+
+# ---------------------------------------------------------------------------
+# Repo-level wiring
+# ---------------------------------------------------------------------------
+
+
+def discover_test_files(root: Path) -> list[str]:
+    return sorted(
+        str(p.relative_to(root))
+        for d in ("tests", "scripts/tests")
+        for p in (root / d).glob("test_*.py")
+        if (root / d).is_dir()
+    )
+
+
+def discover_test_dirs(root: Path) -> list[str]:
+    return sorted(
+        d for d in ("tests", "scripts/tests")
+        if (root / d).is_dir() and any((root / d).glob("test_*.py"))
+    )
+
+
+def read_testpaths(root: Path) -> list[str]:
+    ini = root / "pytest.ini"
+    if not ini.exists():
+        return []
+    for line in ini.read_text(encoding="utf-8").splitlines():
+        if line.strip().startswith("testpaths"):
+            return line.split("=", 1)[1].split()
+    return []
+
+
+def collect_counts(root: Path) -> dict[str, int]:
+    """Ask pytest what it actually collects, per file."""
+    proc = subprocess.run(
+        [sys.executable, "-m", "pytest", "--collect-only", "-q"],
+        cwd=root, capture_output=True, text=True,
+    )
+    counts: dict[str, int] = {}
+    for line in proc.stdout.splitlines():
+        match = re.match(r"^([\w./-]+\.py)::", line.strip())
+        if match:
+            counts[match.group(1)] = counts.get(match.group(1), 0) + 1
+    return counts
+
+
+def run_all(root: Path) -> list[str]:
+    problems: list[str] = []
+
+    test_files = discover_test_files(root)
+    if not test_files:
+        return ["no test files found at all — this check would pass vacuously"]
+
+    problems += check_every_test_file_collects(test_files, collect_counts(root))
+    problems += check_testpaths_cover_suites(read_testpaths(root), discover_test_dirs(root))
+
+    catalog_path = root / "skill-catalog.json"
+    if catalog_path.exists():
+        catalog = json.loads(catalog_path.read_text(encoding="utf-8"))
+        problems += check_catalog_matches_filesystem(catalog, root / "prebuilt", root)
+
+    problems += check_every_skill_has_fixtures(root / "prebuilt")
+    return problems
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--root", type=Path, default=Path(__file__).resolve().parent.parent)
+    parser.add_argument("--json", action="store_true", help="machine-readable output")
+    args = parser.parse_args()
+
+    problems = run_all(args.root)
+
+    if args.json:
+        print(json.dumps({"problems": problems, "ok": not problems}, ensure_ascii=False, indent=2))
+    elif problems:
+        print(f"✗ {len(problems)} gate-liveness problem(s):\n")
+        for p in problems:
+            print(f"  - {p}")
+        print("\nA gate that examines nothing reports the same green as one that passes.")
+    else:
+        print("✓ gate liveness ok — every gate examined a non-empty set")
+
+    return 1 if problems else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())