feat(harness): workspace.py — Workspace lifecycle + VersionedSkillStore/PromptStore

- ResolvedPaths frozen dataclass: store_dir, videos_dir, questions_dir,
  skills_dir, prompts_dir, workspace_dir, db_path, analyses_dir, runs_dir
- init_workspace: create ws + copy seed weights from store
- init_workspace_from_seed: create from seed with fail-fast questions check
- load_manifest / resolve_paths: manifest I/O + path resolution
  (skills/prompts resolve to workspace, videos/questions to store)
- update_manifest: key whitelist validation
- record_run: idempotent history append + per-video wiki dirs
- read_best / update_best: best pointer independent of current
- list_video_ids: videos with tree.json
- archive_workspace: move to .archive/<name>-<ts>
- VersionedSkillStore: implements core/evolution/protocols.py::SkillStore
- VersionedPromptStore: implements core/evolution/protocols.py::PromptStore
- 21 tests all passing (incl. Protocol compliance checks)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-07-07 12:41:59 -04:00
parent b052c1f3ee
commit d349fe1148
2 changed files with 997 additions and 0 deletions
+519
View File
@@ -0,0 +1,519 @@
"""app/harness/workspace 单元测试。
覆盖 Workspace 生命周期管理、manifest 读写、
VersionedSkillStore / VersionedPromptStore Protocol 合规。
"""
from __future__ import annotations
import json
import shutil
from pathlib import Path
import pytest
from app.harness.workspace import (
ResolvedPaths,
VersionedPromptStore,
VersionedSkillStore,
archive_workspace,
init_workspace,
init_workspace_from_seed,
list_video_ids,
load_manifest,
read_best,
record_run,
resolve_paths,
update_best,
update_manifest,
)
from core.evolution.protocols import PromptStore, SkillStore
# ---------------------------------------------------------------------------
# Fixtures
# ---------------------------------------------------------------------------
@pytest.fixture()
def store_dir(tmp_path: Path) -> Path:
"""构建一个最小 Store 目录结构供测试使用。"""
sd = tmp_path / "store"
sd.mkdir()
# videos:两个含 tree.json 的视频目录
for vid in ("vid_A", "vid_B"):
vdir = sd / "videos" / vid
vdir.mkdir(parents=True)
(vdir / "tree.json").write_text("{}")
# skills/v1 + prompts/v1
s1 = sd / "skills" / "v1"
s1.mkdir(parents=True)
(s1 / "temporal-reasoning.md").write_text("skill content A")
(s1 / "spatial-analysis.md").write_text("skill content B")
p1 = sd / "prompts" / "v1"
p1.mkdir(parents=True)
(p1 / "system.md").write_text("system prompt v1")
(p1 / "extract.md").write_text("extract prompt v1")
# questions/benchmarks/Video-MME
q = sd / "questions" / "benchmarks" / "Video-MME"
q.mkdir(parents=True)
(q / "q1.json").write_text('{"id": "q1"}')
# seed "initial"
seed_dir = sd / "seeds" / "initial"
seed_dir.mkdir(parents=True)
shutil.copytree(s1, seed_dir / "skills")
shutil.copytree(p1, seed_dir / "prompts")
baseline_db = seed_dir / "baseline.db"
baseline_db.write_bytes(b"")
(seed_dir / "seed.json").write_text(
json.dumps({"baseline_run_id": "baseline-001", "parent": None})
)
return sd
@pytest.fixture()
def workspace_dir(tmp_path: Path) -> Path:
"""返回一个不存在的 workspace 路径。"""
return tmp_path / "ws_test"
# ---------------------------------------------------------------------------
# ResolvedPaths
# ---------------------------------------------------------------------------
def test_resolved_paths_frozen() -> None:
"""ResolvedPaths 实例应不可变。"""
rp = ResolvedPaths(
store_dir=Path("/s"),
videos_dir=Path("/v"),
questions_dir=Path("/q"),
skills_dir=Path("/sk"),
prompts_dir=Path("/p"),
workspace_dir=Path("/w"),
db_path=Path("/d"),
analyses_dir=Path("/a"),
runs_dir=Path("/r"),
)
with pytest.raises(AttributeError):
rp.store_dir = Path("/other") # type: ignore[misc]
# ---------------------------------------------------------------------------
# init_workspace
# ---------------------------------------------------------------------------
def test_init_workspace(store_dir: Path, workspace_dir: Path) -> None:
"""init_workspace 应创建 manifest、拷贝权重。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
# manifest 存在
manifest = json.loads((workspace_dir / "manifest.json").read_text())
assert manifest["current"]["skills"] == "skills/v1"
assert manifest["current"]["prompts"] == "prompts/v1"
assert manifest["current"]["questions"] == "questions/benchmarks/Video-MME"
# 权重已拷入 workspace
assert (workspace_dir / "skills" / "v1" / "temporal-reasoning.md").exists()
assert (workspace_dir / "prompts" / "v1" / "system.md").exists()
# analyses, runs 目录已创建
assert (workspace_dir / "analyses").is_dir()
assert (workspace_dir / "runs").is_dir()
# 重复创建应报错
with pytest.raises(FileExistsError):
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
# ---------------------------------------------------------------------------
# init_workspace_from_seed
# ---------------------------------------------------------------------------
def test_init_workspace_from_seed(store_dir: Path, workspace_dir: Path) -> None:
"""从种子初始化应拷权重 + baseline.db + 返回 baseline_run_id。"""
run_id = init_workspace_from_seed(
workspace_dir,
store_dir,
seed_name="initial",
questions="benchmarks/Video-MME",
)
assert run_id == "baseline-001"
# 权重在 workspace
assert (workspace_dir / "skills" / "v1" / "temporal-reasoning.md").exists()
assert (workspace_dir / "prompts" / "v1" / "system.md").exists()
# baseline.db -> harness.db
assert (workspace_dir / "harness.db").exists()
# manifest 正确
manifest = json.loads((workspace_dir / "manifest.json").read_text())
assert manifest["current"]["skills"] == "skills/v1"
def test_init_workspace_from_seed_missing_questions(store_dir: Path, workspace_dir: Path) -> None:
"""questions ref 不存在应 fail-fastFileNotFoundError),不创建 workspace。"""
with pytest.raises(FileNotFoundError, match="questions ref"):
init_workspace_from_seed(
workspace_dir,
store_dir,
seed_name="initial",
questions="benchmarks/NONEXISTENT",
)
# workspace 不应被创建
assert not workspace_dir.exists()
# ---------------------------------------------------------------------------
# load_manifest
# ---------------------------------------------------------------------------
def test_load_manifest(store_dir: Path, workspace_dir: Path) -> None:
"""正常加载 manifest。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
manifest = load_manifest(workspace_dir)
assert "current" in manifest
assert "history" in manifest
assert manifest["current"]["skills"] == "skills/v1"
def test_load_manifest_missing(workspace_dir: Path) -> None:
"""manifest 不存在应 FileNotFoundError。"""
workspace_dir.mkdir(parents=True)
with pytest.raises(FileNotFoundError):
load_manifest(workspace_dir)
# ---------------------------------------------------------------------------
# update_manifest
# ---------------------------------------------------------------------------
def test_update_manifest_invalid_key(store_dir: Path, workspace_dir: Path) -> None:
"""非法 key 应 KeyError。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
with pytest.raises(KeyError, match="无效"):
update_manifest(workspace_dir, bad_key="skills/v2")
def test_update_manifest_valid(store_dir: Path, workspace_dir: Path) -> None:
"""合法 key 应更新 manifest current。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
update_manifest(workspace_dir, skills="skills/v2")
manifest = load_manifest(workspace_dir)
assert manifest["current"]["skills"] == "skills/v2"
# ---------------------------------------------------------------------------
# record_run
# ---------------------------------------------------------------------------
def test_record_run_idempotent(store_dir: Path, workspace_dir: Path) -> None:
"""同 run_id 调用两次不应重复追加 history。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
run_dir1 = record_run(workspace_dir, "run_001")
run_dir2 = record_run(workspace_dir, "run_001")
# 返回路径一致
assert run_dir1 == run_dir2
# history 只有一条
manifest = load_manifest(workspace_dir)
matched = [h for h in manifest["history"] if h["run_id"] == "run_001"]
assert len(matched) == 1
# run 目录存在
assert run_dir1.is_dir()
# per-video wiki 目录存在
assert (run_dir1 / "vid_A" / "wiki").is_dir()
assert (run_dir1 / "vid_B" / "wiki").is_dir()
# ---------------------------------------------------------------------------
# update_best / read_best
# ---------------------------------------------------------------------------
def test_update_best_independent_of_current(store_dir: Path, workspace_dir: Path) -> None:
"""best 应独立于 current——更新 best 不影响 current。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
# 初始无 best
assert read_best(workspace_dir) is None
# 设置 best
update_best(
workspace_dir,
skills="skills/v3",
prompts="prompts/v3",
val_acc=0.85,
run_id="run_005",
epoch=3,
)
best = read_best(workspace_dir)
assert best is not None
assert best["skills"] == "skills/v3"
assert best["val_acc"] == 0.85
assert best["epoch"] == 3
# current 不受影响
manifest = load_manifest(workspace_dir)
assert manifest["current"]["skills"] == "skills/v1"
# ---------------------------------------------------------------------------
# archive_workspace
# ---------------------------------------------------------------------------
def test_archive_workspace(store_dir: Path, workspace_dir: Path) -> None:
"""归档应把 workspace 移到 .archive/ 下。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
archived = archive_workspace(workspace_dir)
# 原目录不存在
assert not workspace_dir.exists()
# 归档目录存在并包含 manifest
assert archived.is_dir()
assert (archived / "manifest.json").exists()
# 归档路径在 .archive 下
assert archived.parent.name == ".archive"
def test_archive_workspace_missing(workspace_dir: Path) -> None:
"""归档不存在的 workspace 应 FileNotFoundError。"""
with pytest.raises(FileNotFoundError):
archive_workspace(workspace_dir)
# ---------------------------------------------------------------------------
# list_video_ids
# ---------------------------------------------------------------------------
def test_list_video_ids(store_dir: Path, workspace_dir: Path) -> None:
"""列出 workspace 引用的所有视频 ID。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
ids = list_video_ids(workspace_dir)
assert ids == ["vid_A", "vid_B"]
# ---------------------------------------------------------------------------
# resolve_paths
# ---------------------------------------------------------------------------
def test_resolve_paths(store_dir: Path, workspace_dir: Path) -> None:
"""resolve_paths 应正确解析绝对路径。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
rp = resolve_paths(workspace_dir)
# skills_dir/prompts_dir 解析到 workspace(非 store
ws_abs = workspace_dir.resolve()
assert rp.skills_dir == ws_abs / "skills" / "v1"
assert rp.prompts_dir == ws_abs / "prompts" / "v1"
# videos/questions 解析到 store
store_abs = store_dir.resolve()
assert rp.videos_dir == store_abs / "videos"
assert rp.questions_dir == store_abs / "questions" / "benchmarks" / "Video-MME"
# db_path, analyses, runs
assert rp.db_path == ws_abs / "harness.db"
assert rp.analyses_dir == ws_abs / "analyses"
assert rp.runs_dir == ws_abs / "runs"
# ---------------------------------------------------------------------------
# VersionedSkillStore
# ---------------------------------------------------------------------------
def test_versioned_skill_store_read(store_dir: Path, workspace_dir: Path) -> None:
"""read_skill 应返回文件全文。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
skills_dir = workspace_dir / "skills" / "v1"
store = VersionedSkillStore(skills_dir)
content = store.read_skill("temporal-reasoning.md")
assert content == "skill content A"
def test_versioned_skill_store_read_missing(store_dir: Path, workspace_dir: Path) -> None:
"""读取不存在的 skill 文件应 FileNotFoundError。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
skills_dir = workspace_dir / "skills" / "v1"
store = VersionedSkillStore(skills_dir)
with pytest.raises(FileNotFoundError):
store.read_skill("nonexistent.md")
def test_versioned_skill_store_list(store_dir: Path, workspace_dir: Path) -> None:
"""list_skill_files 应列出所有 skill 文件名(排序)。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
skills_dir = workspace_dir / "skills" / "v1"
store = VersionedSkillStore(skills_dir)
files = store.list_skill_files()
assert sorted(files) == ["spatial-analysis.md", "temporal-reasoning.md"]
# ---------------------------------------------------------------------------
# VersionedPromptStore
# ---------------------------------------------------------------------------
def test_versioned_prompt_store(store_dir: Path, workspace_dir: Path) -> None:
"""VersionedPromptStore 读写功能验证。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
prompts_dir = workspace_dir / "prompts" / "v1"
store = VersionedPromptStore(prompts_dir)
content = store.read_prompt("system.md")
assert content == "system prompt v1"
files = store.list_prompt_files()
assert sorted(files) == ["extract.md", "system.md"]
def test_versioned_prompt_store_read_missing(store_dir: Path, workspace_dir: Path) -> None:
"""读取不存在的 prompt 文件应 FileNotFoundError。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
prompts_dir = workspace_dir / "prompts" / "v1"
store = VersionedPromptStore(prompts_dir)
with pytest.raises(FileNotFoundError):
store.read_prompt("nonexistent.md")
# ---------------------------------------------------------------------------
# Protocol 合规
# ---------------------------------------------------------------------------
def test_skill_store_protocol_compliance(store_dir: Path, workspace_dir: Path) -> None:
"""VersionedSkillStore 应满足 core/evolution/protocols.py::SkillStore Protocol。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
skills_dir = workspace_dir / "skills" / "v1"
store = VersionedSkillStore(skills_dir)
assert isinstance(store, SkillStore)
def test_prompt_store_protocol_compliance(store_dir: Path, workspace_dir: Path) -> None:
"""VersionedPromptStore 应满足 core/evolution/protocols.py::PromptStore Protocol。"""
init_workspace(
workspace_dir,
store_dir,
questions="benchmarks/Video-MME",
skills_version="v1",
prompts_version="v1",
)
prompts_dir = workspace_dir / "prompts" / "v1"
store = VersionedPromptStore(prompts_dir)
assert isinstance(store, PromptStore)