"""Prompt/file contracts, not a substitute for trace-based execution validation."""

import json
from pathlib import Path

import pytest

from tests.test_diagnosis_orchestration_config import _catalog_rows


ROOT = Path(__file__).resolve().parents[2]
CONFIG = ROOT / "agentkit/config"
SKILLS = ROOT / "agentkit/skills"
CLINIC = SKILLS / "builtin/tidb-clinic-diagnosis"


def test_default_loading_does_not_require_general_manual():
    investigator = (CONFIG / "investibook/agents/investigator.md").read_text()
    analyst = (CONFIG / "investibook/agents/diagnosis-analyst.md").read_text()
    catalog = (CONFIG / "SKILL_CATALOG.md").read_text()
    entry = (CLINIC / "SKILL.md").read_text()
    assert "primary playbook" not in analyst
    assert "primary manual" not in analyst
    assert "default `core` + `tidb-clinic-diagnosis`" not in investigator
    assert "do not add the root `tidb-clinic-diagnosis` manual" in investigator
    assert "not a default required read" in catalog
    assert "General-method Fallback" in analyst
    assert "Low confidence, missing" in analyst
    assert "record the requested name/path and failure" in analyst
    assert len(entry.splitlines()) < 100
    assert "### Node C:" not in entry
    assert "references/general-diagnosis-fallback.md" in entry
    assert "references/evidence-collection.md" in entry
    assert (CLINIC / "references/general-diagnosis-fallback.md").is_file()
    assert (CLINIC / "references/evidence-collection.md").is_file()


@pytest.mark.parametrize(
    ("family", "specialist", "references"),
    [
        ("sql-plan-or-query", "tidb-query-tuning-autox",
         ("SUBSKILLS_INDEX.md", "references/case-contract.md", "workflow/diagnosis-classification/SUBSKILL.md")),
        ("transaction-or-lock-contention", "transaction-oncall-handbook", ()),
        ("cdc-or-replication-lag", "ticdc-health-inspection",
         ("references/dashboard-mapping.md", "references/inspection-rules.md")),
    ],
)
def test_required_specialists_resolve_to_readable_files_and_allowed_tools(family, specialist, references):
    catalog = {name: (group, path) for group, name, path in _catalog_rows()}
    family_text = (CLINIC / family / "SUBSKILL.md").read_text()
    assert "# Specialist Loading Contract" in family_text
    assert "task-selected route" in family_text
    assert specialist in family_text
    group, path = catalog[specialist]
    assert group == "diagnosis"
    assert (SKILLS / path / "SKILL.md").is_file()
    for reference in references:
        assert (SKILLS / path / reference).is_file()
    permissions = json.loads((CONFIG / "opencode.json.template").read_text())["agent"]["diagnosis-analyst"]["permission"]
    assert permissions["skill"][specialist] == "allow"
    assert permissions["bash"]["*"] == "deny"
    assert not any("collect_evidence" in command for command in permissions["bash"])


def test_optional_cdc_boundary_handbook_is_discoverable_but_not_loaded_for_lag_alone():
    catalog = {name: path for _, name, path in _catalog_rows()}
    assert (SKILLS / catalog["tikv-cdc"] / "SKILL.md").is_file()
    cdc = (CLINIC / "cdc-or-replication-lag/SUBSKILL.md").read_text()
    assert "Lag alone does not require it" in cdc
    assert "never deployed" in cdc
    assert "rather than declaring it healthy" in cdc
    assert "no direct data collection" in cdc
    assert "not a new\n30m/24h health sweep" in cdc
    assert "analyst-discovered primary class" in cdc


def test_nodegroup_comparability_is_in_mandatory_shared_rules():
    overview = (SKILLS / "builtin/diagnosis-overview/SKILL.md").read_text()
    assert "within the same nodegroup" in overview
    assert "capacity, quota, and workload/routing differences" in overview
    assert "do not infer membership from node" in overview
    assert "comparability is unknown" in overview


def test_shared_guardrails_and_multibranch_contract_survive_without_root_manual():
    investigator = (CONFIG / "investibook/agents/investigator.md").read_text()
    analyst = (CONFIG / "investibook/agents/diagnosis-analyst.md").read_text()
    description = json.loads((CONFIG / "opencode.json.template").read_text())["agent"]["diagnosis-analyst"]["description"]
    assert "hard diagnoses" in description
    assert "the investigator concludes those directly" in description
    assert "every exact selected family subskill path" in investigator
    assert "never state the expected conclusion" in investigator
    assert "Compare every selected branch explicitly" in analyst
    assert "stop synthesis, read the final family" in analyst
    assert "classification_override" in analyst
    assert "canonical window" in analyst
    assert "High TiDB CPU → also check Go GC metrics" in analyst
    assert "incorrect-conclusions.json" in analyst
    assert "write_analyst_report.py" in analyst
    assert "An expected specialist missing" in investigator


def test_fallback_does_not_restore_generic_vetoes_or_a_second_workflow():
    fallback = (CLINIC / "references/general-diagnosis-fallback.md").read_text()
    assert "not a second orchestration workflow" in fallback
    assert "Finish through the analyst's normal report writer" in fallback
    assert "are **NEVER** the primary root cause" not in fallback
    assert "If the suspected SQL is NOT in TopSQL top 20" not in fallback
    assert "not proof of absence" in fallback
    assert "SQL specialist candidate conditions" not in fallback
    assert "sql-plan-or-query/SUBSKILL.md" in fallback
    assert "## Data Access for Additional Queries" not in fallback


def test_unselected_sql_does_not_route_from_rank_or_missing_schema():
    sql = (CLINIC / "sql-plan-or-query/SUBSKILL.md").read_text()
    assert "For analyst-discovered candidates" in sql
    assert "For a task-selected SQL route, load AutoX first" in sql
    assert "Direct mechanism signal beyond rank" in sql
    assert "Claim-appropriate attribution" in sql
    assert "incomplete workflow" in sql
    assert "do not substitute an uncataloged skill" in sql
    assert "do not add an unselected SQL branch" in sql