src/tests/test_ctx_init.py

"""Tests for ctx_init — bootstrap ~/.claude/ scaffolding."""

from __future__ import annotations

import builtins
import io
import json
import sys
import tarfile
from pathlib import Path
from types import SimpleNamespace

import networkx as nx
import ctx_init as ci


def test_ensure_directories_creates_standard_tree(tmp_path: Path) -> None:
    created = ci.ensure_directories(tmp_path)
    # First call should create every standard subdir.
    assert len(created) == len(ci._STANDARD_SUBDIRS)
    for sub in ci._STANDARD_SUBDIRS:
        assert (tmp_path / sub).is_dir(), f"missing {sub}"


def test_ensure_directories_is_idempotent(tmp_path: Path) -> None:
    first = ci.ensure_directories(tmp_path)
    assert len(first) > 0
    second = ci.ensure_directories(tmp_path)
    assert second == [], "second call should not recreate anything"


def test_seed_user_config_writes_once(tmp_path: Path) -> None:
    tmp_path.mkdir(exist_ok=True)
    first = ci.seed_user_config(tmp_path)
    assert first is not None
    assert first.exists()
    body = first.read_text(encoding="utf-8")
    assert "skill-system-config.json" in body

    # Second call returns None (file already exists, force=False).
    second = ci.seed_user_config(tmp_path)
    assert second is None


def test_seed_user_config_respects_force(tmp_path: Path) -> None:
    target = tmp_path / "skill-system-config.json"
    target.write_text("user-custom-content", encoding="utf-8")
    # Without force → don't touch.
    assert ci.seed_user_config(tmp_path, force=False) is None
    assert target.read_text() == "user-custom-content"
    # With force → overwrite.
    result = ci.seed_user_config(tmp_path, force=True)
    assert result == target
    assert "skill-system-config.json" in target.read_text()


def test_main_creates_everything_in_dry_mode(tmp_path: Path, monkeypatch,
                                              capsys) -> None:
    """End-to-end: ``ctx-init`` (no flags) creates dirs + config + toolboxes
    without touching hooks or graph."""
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)

    # Short-circuit subprocess.run to avoid spawning a real toolbox/graph CLI
    # in tests. Verify that main() doesn't call install_hooks or build_graph
    # when those flags are absent.
    calls: list[list[str]] = []

    class _FakeResult:
        returncode = 0
        stdout = ""
        stderr = ""

    def fake_run(cmd, **kwargs):
        calls.append(list(cmd))
        return _FakeResult()

    monkeypatch.setattr(ci.subprocess, "run", fake_run)

    rc = ci.main([])
    assert rc == 0
    # toolbox init should have been invoked
    toolbox_calls = [c for c in calls if "toolbox" in " ".join(c)]
    assert toolbox_calls, "toolbox init not invoked"
    # inject_hooks / wiki_graphify must NOT be invoked without flags
    for c in calls:
        assert "inject_hooks" not in " ".join(c)
        assert "wiki_graphify" not in " ".join(c)

    out = capsys.readouterr().out
    assert "[ok]" in out
    assert "[skip] hook injection" in out
    assert "[skip] graph install" in out


def test_main_treats_existing_toolboxes_as_idempotent_skip(
    tmp_path: Path,
    monkeypatch,
    capsys,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)

    class _FakeResult:
        returncode = 1
        stdout = ""
        stderr = (
            "Global config already has 5 toolbox(es). "
            "Use --force to overwrite."
        )

    monkeypatch.setattr(ci.subprocess, "run", lambda *_args, **_kwargs: _FakeResult())

    rc = ci.main(["--model-mode", "skip"])
    captured = capsys.readouterr()

    assert rc == 0
    assert "starter toolboxes already present" in captured.out
    assert "toolbox init returned" not in captured.err
    assert "Global config already has" not in captured.err


def test_main_auto_wizard_in_terminal_configures_custom_model(
    tmp_path: Path,
    monkeypatch,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "_stdio_is_interactive", lambda: True)
    monkeypatch.setattr(
        ci,
        "recommend_harnesses",
        lambda goal, top_k=5, model_provider=None, model=None: [],
    )

    answers = iter([
        "y",                  # hooks
        "enriched",           # knowledge mode
        "n",                  # graph
        "custom",             # model mode
        "openai/gpt-5.5",     # model
        "",                   # provider default: openai
        "",                   # api key env default: OPENAI_API_KEY
        "",                   # base URL
        "build CAD artifacts",
        "windows python",      # runtime / OS
        "supervised",          # autonomy
        "filesystem shell",    # allowed tools
        "pytest ruff",         # verification
        "private repo",        # privacy / network
        "mcp",                 # attach mode
        "n",                  # validate model
    ])
    monkeypatch.setattr(builtins, "input", lambda _prompt: next(answers))
    calls: list[list[str]] = []

    class _FakeResult:
        returncode = 0
        stdout = ""
        stderr = ""

    def _fake_run(cmd: list[str], **_kwargs: object) -> _FakeResult:
        calls.append(list(cmd))
        return _FakeResult()

    monkeypatch.setattr(ci.subprocess, "run", _fake_run)

    rc = ci.main([])

    assert rc == 0
    assert any("ctx.adapters.claude_code.inject_hooks" in c for c in calls)
    assert not any("ctx.core.wiki.wiki_graphify" in c for c in calls)
    profile = json.loads((tmp_path / "ctx-model-profile.json").read_text())
    assert profile["mode"] == "custom"
    assert profile["provider"] == "openai"
    assert profile["model"] == "openai/gpt-5.5"
    assert profile["api_key_env"] == "OPENAI_API_KEY"
    assert profile["goal"] == "build CAD artifacts"
    assert profile["knowledge_mode"] == "enriched"
    assert profile["harness_requirements"] == {
        "runtime": "windows python",
        "autonomy": "supervised",
        "tools": "filesystem shell",
        "verification": "pytest ruff",
        "privacy": "private repo",
        "attach_mode": "mcp",
    }
    user_config = json.loads((tmp_path / "skill-system-config.json").read_text())
    assert user_config["knowledge"]["mode"] == "enriched"


def test_wizard_flag_prompts_without_tty(tmp_path: Path, monkeypatch) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "_stdio_is_interactive", lambda: False)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    monkeypatch.setattr(
        ci,
        "recommend_harnesses",
        lambda goal, top_k=5, model_provider=None, model=None: [],
    )

    answers = iter([
        "n",                  # hooks
        "local",              # knowledge mode
        "claude-code",        # model mode
        "maintain FastAPI services",
    ])
    monkeypatch.setattr(builtins, "input", lambda _prompt: next(answers))

    rc = ci.main(["--wizard"])

    assert rc == 0
    profile = json.loads((tmp_path / "ctx-model-profile.json").read_text())
    assert profile["mode"] == "claude-code"
    assert profile["goal"] == "maintain FastAPI services"
    assert profile["knowledge_mode"] == "local"
    user_config = json.loads((tmp_path / "skill-system-config.json").read_text())
    assert user_config["knowledge"]["mode"] == "local"


def test_explicit_args_do_not_auto_wizard_in_terminal(
    tmp_path: Path,
    monkeypatch,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "_stdio_is_interactive", lambda: True)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    monkeypatch.setattr(
        builtins,
        "input",
        lambda _prompt: (_ for _ in ()).throw(AssertionError("unexpected prompt")),
    )

    assert ci.main(["--model-mode", "skip", "--knowledge-mode", "local"]) == 0
    user_config = json.loads((tmp_path / "skill-system-config.json").read_text())
    assert user_config["knowledge"]["mode"] == "local"


def test_main_with_hooks_flag_invokes_inject(tmp_path: Path, monkeypatch) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    calls: list[list[str]] = []

    class _FakeResult:
        returncode = 0
        stdout = ""
        stderr = ""

    def _fake_run(cmd: list[str], **_kwargs: object) -> _FakeResult:
        calls.append(list(cmd))
        return _FakeResult()

    monkeypatch.setattr(ci.subprocess, "run", _fake_run)
    rc = ci.main(["--hooks"])
    assert rc == 0
    assert any("ctx.adapters.claude_code.inject_hooks" in c for c in calls)
    assert not any(c == "inject_hooks" for call in calls for c in call)


def _write_graph_archive(tmp_path: Path) -> Path:
    source = tmp_path / "archive-source"
    graph_out = source / "graphify-out"
    graph_out.mkdir(parents=True)
    (graph_out / "graph.json").write_text(
        json.dumps({"graph": {"export_id": "test-export"}, "nodes": [], "links": []}),
        encoding="utf-8",
    )
    (graph_out / "graph-delta.json").write_text(
        json.dumps({"export_id": "test-export", "nodes": [], "edges": []}),
        encoding="utf-8",
    )
    (graph_out / "communities.json").write_text(
        json.dumps({"export_id": "test-export", "total_communities": 0}),
        encoding="utf-8",
    )
    (graph_out / "graph-report.md").write_text(
        "# Graph Report\n\n> Export ID: test-export\n",
        encoding="utf-8",
    )
    (graph_out / "graph-export-manifest.json").write_text(
        json.dumps({
            "version": 1,
            "export_id": "test-export",
            "artifacts": {
                "graph": "graph.json",
                "delta": "graph-delta.json",
                "communities": "communities.json",
                "report": "graph-report.md",
            },
        }),
        encoding="utf-8",
    )
    external = source / "external-catalogs" / "skills-sh"
    external.mkdir(parents=True)
    (external / "catalog.json").write_text("{}", encoding="utf-8")
    entities = source / "entities" / "skills"
    entities.mkdir(parents=True)
    (entities / "current.md").write_text("# Current\n", encoding="utf-8")
    (source / "index.md").write_text("# Wiki\n", encoding="utf-8")
    archive = tmp_path / "wiki-graph.tar.gz"
    with tarfile.open(archive, "w:gz") as tf:
        for path in sorted(source.rglob("*")):
            if path.is_file():
                tf.add(path, arcname=path.relative_to(source).as_posix())
    return archive


def test_main_with_graph_flag_installs_prebuilt_graph(
    tmp_path: Path,
    monkeypatch,
) -> None:
    claude = tmp_path / "home"
    archive = _write_graph_archive(tmp_path)
    monkeypatch.setattr(ci, "_claude_dir", lambda: claude)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    monkeypatch.setattr(
        ci,
        "_find_local_graph_archive",
        lambda _install_mode="runtime": archive,
        raising=False,
    )
    monkeypatch.setattr(
        ci,
        "_download_graph_archive",
        lambda _dest, **_kwargs: (_ for _ in ()).throw(
            AssertionError("unexpected release download")
        ),
        raising=False,
    )
    calls: list[list[str]] = []

    class _FakeResult:
        returncode = 0
        stdout = ""
        stderr = ""

    def _fake_run(cmd: list[str], **_kwargs: object) -> _FakeResult:
        calls.append(list(cmd))
        return _FakeResult()

    monkeypatch.setattr(ci.subprocess, "run", _fake_run)
    rc = ci.main(["--graph", "--model-mode", "skip"])
    assert rc == 0
    graph_json = claude / "skill-wiki" / "graphify-out" / "graph.json"
    graph_payload = json.loads(graph_json.read_text(encoding="utf-8"))
    assert graph_payload["graph"]["export_id"] == "test-export"
    assert not (
        claude / "skill-wiki" / "entities" / "skills" / "current.md"
    ).exists()
    assert not any("ctx.core.wiki.wiki_graphify" in c for c in calls)
    assert not any(c == "wiki_graphify" for call in calls for c in call)


def test_graph_install_rejects_incomplete_archive(
    tmp_path: Path,
    monkeypatch,
) -> None:
    source = tmp_path / "incomplete-source"
    graph_out = source / "graphify-out"
    graph_out.mkdir(parents=True)
    (graph_out / "graph.json").write_text(
        json.dumps({"graph": {"export_id": "partial"}, "nodes": []}),
        encoding="utf-8",
    )
    archive = tmp_path / "incomplete-wiki-graph.tar.gz"
    with tarfile.open(archive, "w:gz") as tf:
        tf.add(graph_out / "graph.json", arcname="graphify-out/graph.json")

    claude = tmp_path / "home"
    monkeypatch.setattr(
        ci,
        "_find_local_graph_archive",
        lambda _install_mode="runtime": archive,
    )

    assert ci.build_graph(claude) == 1
    assert not (claude / "skill-wiki" / "graphify-out" / "graph.json").exists()


def test_graph_install_validation_does_not_parse_full_graph_json(
    tmp_path: Path,
    monkeypatch,
) -> None:
    wiki = tmp_path / "wiki"
    graph_out = wiki / "graphify-out"
    graph_out.mkdir(parents=True)
    (wiki / "index.md").write_text("# Wiki\n", encoding="utf-8")
    (graph_out / "graph.json").write_text(
        json.dumps({"graph": {"export_id": "test-export"}, "nodes": [], "links": []}),
        encoding="utf-8",
    )
    (graph_out / "graph-delta.json").write_text(
        json.dumps({"export_id": "test-export", "nodes": [], "edges": []}),
        encoding="utf-8",
    )
    (graph_out / "communities.json").write_text(
        json.dumps({"export_id": "test-export", "total_communities": 0}),
        encoding="utf-8",
    )
    (graph_out / "graph-report.md").write_text(
        "# Graph Report\n\n> Export ID: test-export\n",
        encoding="utf-8",
    )
    (graph_out / "graph-export-manifest.json").write_text(
        json.dumps({
            "version": 1,
            "export_id": "test-export",
            "artifacts": {
                "graph": "graph.json",
                "delta": "graph-delta.json",
                "communities": "communities.json",
                "report": "graph-report.md",
            },
        }),
        encoding="utf-8",
    )
    external = wiki / "external-catalogs" / "skills-sh"
    external.mkdir(parents=True)
    (external / "catalog.json").write_text("{}", encoding="utf-8")

    def guarded_read(path: Path) -> object:
        if path.name == "graph.json":
            raise AssertionError("install validation must not parse full graph.json")
        return json.loads(path.read_text(encoding="utf-8"))

    monkeypatch.setattr(ci, "_read_json_file", guarded_read)

    ci._validate_graph_install_tree(wiki)


def test_graph_install_force_prunes_stale_generated_files(
    tmp_path: Path,
    monkeypatch,
) -> None:
    archive = _write_graph_archive(tmp_path)
    claude = tmp_path / "home"
    stale = claude / "skill-wiki" / "entities" / "skills" / "stale.md"
    stale.parent.mkdir(parents=True)
    stale.write_text("# Stale\n", encoding="utf-8")
    monkeypatch.setattr(ci, "_claude_dir", lambda: claude)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    monkeypatch.setattr(
        ci,
        "_find_local_graph_archive",
        lambda _install_mode="runtime": archive,
    )

    assert ci.main([
        "--graph",
        "--graph-install-mode", "full",
        "--force",
        "--model-mode", "skip",
    ]) == 0
    assert not stale.exists()
    assert (claude / "skill-wiki" / "entities" / "skills" / "current.md").is_file()


def test_graph_install_rejects_path_traversal_archive(
    tmp_path: Path,
    monkeypatch,
) -> None:
    archive = tmp_path / "malicious-wiki-graph.tar.gz"
    payload = b"owned"
    with tarfile.open(archive, "w:gz") as tf:
        info = tarfile.TarInfo("../evil.txt")
        info.size = len(payload)
        tf.addfile(info, io.BytesIO(payload))

    claude = tmp_path / "home"
    monkeypatch.setattr(
        ci,
        "_find_local_graph_archive",
        lambda _install_mode="runtime": archive,
    )

    assert ci.build_graph(claude) == 1
    assert not (tmp_path / "evil.txt").exists()
    assert not (claude / "evil.txt").exists()


def test_main_with_requested_hook_failure_exits_nonzero(
    tmp_path: Path, monkeypatch
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)

    class _FakeResult:
        def __init__(self, returncode: int) -> None:
            self.returncode = returncode
            self.stdout = ""
            self.stderr = ""

    def fake_run(cmd, **kwargs):
        if "ctx.adapters.claude_code.inject_hooks" in cmd:
            return _FakeResult(7)
        return _FakeResult(0)

    monkeypatch.setattr(ci.subprocess, "run", fake_run)

    assert ci.main(["--hooks"]) == 7


def test_main_custom_model_writes_profile_and_recommends_harness(
    tmp_path: Path,
    monkeypatch,
    capsys,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)

    recommendation_calls: list[dict[str, object]] = []

    def fake_recommend(
        goal: str,
        top_k: int = 5,
        model_provider: str | None = None,
        model: str | None = None,
    ) -> list[dict[str, object]]:
        recommendation_calls.append({
            "goal": goal,
            "top_k": top_k,
            "model_provider": model_provider,
            "model": model,
        })
        return [{"name": "text-to-cad", "type": "harness", "score": 0.8}]

    monkeypatch.setattr(
        ci,
        "recommend_harnesses",
        fake_recommend,
    )

    rc = ci.main([
        "--model-mode", "custom",
        "--model", "openai/gpt-5.5",
        "--goal", "turn text prompts into CAD",
    ])

    assert rc == 0
    profile = json.loads((tmp_path / "ctx-model-profile.json").read_text())
    assert profile["mode"] == "custom"
    assert profile["provider"] == "openai"
    assert profile["model"] == "openai/gpt-5.5"
    assert profile["api_key_env"] == "OPENAI_API_KEY"
    assert recommendation_calls[0]["model_provider"] == "openai"
    assert recommendation_calls[0]["model"] == "openai/gpt-5.5"
    assert "text-to-cad" in capsys.readouterr().out


def test_main_custom_model_records_structured_harness_requirements(
    tmp_path: Path,
    monkeypatch,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    recommendation_calls: list[dict[str, object]] = []

    def fake_recommend(
        goal: str,
        top_k: int = 5,
        model_provider: str | None = None,
        model: str | None = None,
    ) -> list[dict[str, object]]:
        recommendation_calls.append({
            "goal": goal,
            "top_k": top_k,
            "model_provider": model_provider,
            "model": model,
        })
        return []

    monkeypatch.setattr(ci, "recommend_harnesses", fake_recommend)

    rc = ci.main([
        "--model-mode", "custom",
        "--model", "openai/gpt-5.5",
        "--goal", "build a code agent",
        "--harness-runtime", "windows python",
        "--harness-autonomy", "supervised",
        "--harness-tools", "filesystem shell browser",
        "--harness-verify", "pytest ruff",
        "--harness-privacy", "private repo no secrets",
        "--harness-attach-mode", "mcp",
    ])

    assert rc == 0
    profile = json.loads((tmp_path / "ctx-model-profile.json").read_text())
    assert profile["harness_requirements"] == {
        "runtime": "windows python",
        "autonomy": "supervised",
        "tools": "filesystem shell browser",
        "verification": "pytest ruff",
        "privacy": "private repo no secrets",
        "attach_mode": "mcp",
    }
    query = str(recommendation_calls[0]["goal"])
    assert "windows python" in query
    assert "filesystem shell browser" in query
    assert "pytest ruff" in query
    assert "private repo no secrets" in query
    assert "mcp" in query


def test_main_custom_model_no_fit_points_to_harness_plan(
    tmp_path: Path,
    monkeypatch,
    capsys,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    monkeypatch.setattr(ci, "recommend_harnesses", lambda *args, **kwargs: [])

    rc = ci.main([
        "--model-mode", "custom",
        "--model", "ollama/llama3.1",
        "--model-provider", "ollama",
        "--goal", "private local CAD workflow",
        "--harness-runtime", "linux server",
        "--harness-tools", "filesystem shell",
        "--harness-verify", "pytest",
        "--harness-privacy", "offline source code",
        "--harness-attach-mode", "mcp",
    ])

    assert rc == 0
    output = capsys.readouterr().out
    assert "no harness recommendations matched yet" in output
    assert "ctx-harness-install --recommend" in output
    assert "--model-provider \"ollama\"" in output
    assert "--harness-runtime \"linux server\"" in output
    assert "--harness-tools \"filesystem shell\"" in output
    assert "--harness-verify \"pytest\"" in output
    assert "--harness-privacy \"offline source code\"" in output
    assert "--harness-attach-mode \"mcp\"" in output
    assert "--plan-on-no-fit" in output


def test_recommend_harnesses_uses_wiki_frontmatter_for_fit(
    tmp_path: Path,
    monkeypatch,
) -> None:
    wiki = tmp_path / "wiki"
    page = wiki / "entities" / "harnesses" / "text-to-cad.md"
    page.parent.mkdir(parents=True)
    page.write_text(
        """---
title: Text to CAD
type: harness
tags:
  - cad
runtimes:
  - python
model_providers:
  - openai
capabilities:
  - Generate CAD artifacts from natural language prompts
    with OpenSCAD and mesh validation
repo_url: https://github.com/earthtojake/text-to-cad
---
# Text to CAD
""",
        encoding="utf-8",
    )
    graph = nx.Graph()
    graph.add_node(
        "harness:text-to-cad",
        label="text-to-cad",
        type="harness",
        tags=["cad"],
    )
    monkeypatch.setattr(ci, "_load_recommendation_graph", lambda: graph)
    import ctx_config

    monkeypatch.setattr(
        ctx_config,
        "cfg",
        SimpleNamespace(
            wiki_dir=wiki,
            claude_dir=tmp_path / ".claude",
            recommendation_top_k=5,
            harness_recommendation_min_fit_score=0.85,
        ),
    )

    results = ci.recommend_harnesses(
        "turn text prompts into CAD openscad openai gpt-5 harness",
        model_provider="openai",
        model="openai/gpt-5.5",
    )

    assert results
    assert results[0]["name"] == "text-to-cad"
    assert results[0]["fit_score"] >= 0.85
    assert "openai" in results[0]["fit_signals"]
    assert "gpt-5" not in results[0]["fit_signals"]
    assert "gpt-5" not in results[0]["missing_signals"]
    assert "openscad" in results[0]["fit_signals"]


def test_load_recommendation_graph_uses_configured_wiki_dir(
    tmp_path: Path,
    monkeypatch,
) -> None:
    wiki = tmp_path / "custom-wiki"
    out = wiki / "graphify-out"
    out.mkdir(parents=True)
    graph = nx.Graph()
    graph.add_node("harness:custom", label="custom", type="harness")
    data = nx.node_link_data(graph)
    (out / "graph.json").write_text(json.dumps(data), encoding="utf-8")

    import ctx_config

    monkeypatch.setattr(ctx_config, "cfg", SimpleNamespace(wiki_dir=wiki))

    loaded = ci._load_recommendation_graph()

    assert "harness:custom" in loaded


def test_recommend_harnesses_surfaces_reliability_rubric(
    tmp_path: Path,
    monkeypatch,
) -> None:
    wiki = tmp_path / "wiki"
    harness_dir = wiki / "entities" / "harnesses"
    harness_dir.mkdir(parents=True)
    (harness_dir / "reliable-agent.md").write_text(
        """---
title: Reliable Agent
type: harness
tags:
  - agents
model_providers:
  - openai
capabilities:
  - Persistent project context and task state
  - Permission limits, sandbox rules, and policy checks
  - Automated tests, evals, retry loops, and validation gates
verify_commands:
  - pytest
repo_url: https://example.test/reliable-agent
---
# Reliable Agent
""",
        encoding="utf-8",
    )
    graph = nx.Graph()
    graph.add_node(
        "harness:reliable-agent",
        label="reliable-agent",
        type="harness",
        tags=["agents"],
    )
    monkeypatch.setattr(ci, "_load_recommendation_graph", lambda: graph)
    import ctx_config

    monkeypatch.setattr(
        ctx_config,
        "cfg",
        SimpleNamespace(
            wiki_dir=wiki,
            claude_dir=tmp_path / ".claude",
            recommendation_top_k=5,
            harness_recommendation_min_fit_score=0.20,
            harness_reliability_weights={
                "context": 0.34,
                "constraints": 0.33,
                "convergence": 0.33,
            },
        ),
    )

    results = ci.recommend_harnesses(
        "openai agent workflow with tests and sandbox",
        model_provider="openai",
        model="openai/gpt-5.5",
    )

    assert results
    recommendation = results[0]
    assert recommendation["name"] == "reliable-agent"
    assert recommendation["reliability_score"] >= 0.90
    assert set(recommendation["reliability_dimensions"]) == {
        "context",
        "constraints",
        "convergence",
    }
    assert recommendation["reliability_dimensions"]["context"]["matched_terms"]
    assert recommendation["reliability_dimensions"]["constraints"]["matched_terms"]
    assert recommendation["reliability_dimensions"]["convergence"]["matched_terms"]
    assert "context" in recommendation["reliability_reason"]
    assert "constraints" in recommendation["reliability_reason"]
    assert "convergence" in recommendation["reliability_reason"]


def test_recommend_harnesses_prefers_reliable_harness_when_fit_ties(
    tmp_path: Path,
    monkeypatch,
) -> None:
    wiki = tmp_path / "wiki"
    harness_dir = wiki / "entities" / "harnesses"
    harness_dir.mkdir(parents=True)
    (harness_dir / "thin-agent.md").write_text(
        """---
title: Thin Agent
type: harness
tags:
  - agents
model_providers:
  - openai
capabilities:
  - Agent workflow orchestration
repo_url: https://example.test/thin-agent
---
# Thin Agent
""",
        encoding="utf-8",
    )
    (harness_dir / "reliable-agent.md").write_text(
        """---
title: Reliable Agent
type: harness
tags:
  - agents
model_providers:
  - openai
capabilities:
  - Agent workflow orchestration
  - Persistent context state and durable task documents
  - Permission limits, sandbox boundaries, and approval policies
  - Automated tests, evals, validation gates, and retry loops
verify_commands:
  - pytest
repo_url: https://example.test/reliable-agent
---
# Reliable Agent
""",
        encoding="utf-8",
    )
    graph = nx.Graph()
    for slug in ("thin-agent", "reliable-agent"):
        graph.add_node(
            f"harness:{slug}",
            label=slug,
            type="harness",
            tags=["agents"],
        )
    monkeypatch.setattr(ci, "_load_recommendation_graph", lambda: graph)
    import ctx_config

    monkeypatch.setattr(
        ctx_config,
        "cfg",
        SimpleNamespace(
            wiki_dir=wiki,
            claude_dir=tmp_path / ".claude",
            recommendation_top_k=5,
            harness_recommendation_min_fit_score=0.20,
            harness_reliability_weights={
                "context": 0.34,
                "constraints": 0.33,
                "convergence": 0.33,
            },
        ),
    )

    results = ci.recommend_harnesses(
        "openai agent workflow",
        model_provider="openai",
        model="openai/gpt-5.5",
    )

    assert [row["name"] for row in results[:2]] == [
        "reliable-agent",
        "thin-agent",
    ]
    assert results[0]["fit_score"] == results[1]["fit_score"]
    assert results[0]["reliability_score"] > results[1]["reliability_score"]


def test_recommend_harnesses_avoids_semantic_model_load_by_default(
    tmp_path: Path,
    monkeypatch,
) -> None:
    graph = nx.Graph()
    graph.add_node("harness:langgraph", label="langgraph", type="harness")
    monkeypatch.setattr(ci, "_load_recommendation_graph", lambda: graph)
    monkeypatch.setattr(ci, "_harness_supports_provider", lambda *args, **kwargs: True)
    monkeypatch.setattr(ci, "_installed_harness_slugs", lambda _path: set())
    monkeypatch.setattr(
        ci,
        "_annotate_harness_fit",
        lambda *_args, **_kwargs: {"fit_score": 0.99, "fit_signals": ["agent"]},
    )
    import ctx_config

    monkeypatch.setattr(
        ctx_config,
        "cfg",
        SimpleNamespace(
            claude_dir=tmp_path / ".claude",
            recommendation_top_k=5,
            harness_recommendation_min_fit_score=0.85,
        ),
    )
    calls: dict[str, object] = {}

    def fake_recommend_by_tags(*_args, **kwargs):
        calls.update(kwargs)
        return [{"name": "langgraph", "type": "harness", "score": 1.0}]

    monkeypatch.setitem(
        sys.modules,
        "ctx.core.resolve.recommendations",
        type(
            "FakeRecommendModule",
            (),
            {
                "query_to_tags": staticmethod(lambda _query: ["agent"]),
                "recommend_by_tags": staticmethod(fake_recommend_by_tags),
            },
        ),
    )

    results = ci.recommend_harnesses(
        "build an agent workflow",
        model_provider="openai",
        model="openai/gpt-5.5",
    )

    assert results[0]["name"] == "langgraph"
    assert calls["query"] == "build an agent workflow"
    assert calls["entity_types"] == ("harness",)
    assert calls["use_semantic_query"] is False


def test_main_custom_model_requires_model(tmp_path: Path, monkeypatch) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)

    assert ci.main(["--model-mode", "custom"]) == 1


def test_validate_model_flag_invokes_connection_check(
    tmp_path: Path,
    monkeypatch,
) -> None:
    monkeypatch.setattr(ci, "_claude_dir", lambda: tmp_path)
    monkeypatch.setattr(ci, "seed_toolboxes", lambda force=False: 0)
    monkeypatch.setattr(
        ci,
        "recommend_harnesses",
        lambda goal, top_k=5, model_provider=None, model=None: [],
    )
    calls: list[dict] = []

    def fake_validate(**kwargs):
        calls.append(kwargs)
        return 0

    monkeypatch.setattr(ci, "validate_model_connection", fake_validate)

    rc = ci.main([
        "--model-mode", "custom",
        "--model", "ollama/llama3.1",
        "--validate-model",
    ])

    assert rc == 0
    assert calls == [{
        "model": "ollama/llama3.1",
        "api_key_env": None,
        "base_url": None,
    }]