3dcb87f941
问题：Do Worker 可能是廉价模型，字面理解任务。"清理旧产品实现"+src/ = 全删。 根因：Arc 用强模型规划，没考虑 Do Worker 推理能力弱。 修复： - SKILL.md: 新增 INV-16 — 禁止歧义词、否定约束显式化、文件范围精确化、完成标准可验证 - commands/arc.md: 新增"弱模型安全"章节，4 条硬规则 - arc_mode.py: _audit_task_safety() 代码级扫描危险词和目录级文件范围 - 危险词: 清理/清除/删除所有/重构整个/重写全部 等 - 目录级范围(src/)+无否定约束 → 警告 - 警告写入 execution-plan.json 的 safetyWarnings 字段 - parallel_review_mode 在写 plan 前自动执行 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
383 lines
16 KiB
Python
383 lines
16 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
AirArc mode — V2 架构规划器。
|
||
L1 代码级保障：allowed-tools 限制为只读。
|
||
产出 execution-plan.json（完整 DAG）+ plan-delta.json（增量重规划）。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import json
|
||
import sys
|
||
from pathlib import Path
|
||
|
||
from air_runtime.contracts import now_iso
|
||
from air_runtime.io import atomic_json_write, safe_json_load
|
||
from air_runtime.paths import airplan_root, todo_path as get_todo_path
|
||
from air_runtime.review import build_parallel_review, render_review_markdown
|
||
from air_runtime.task_graph import TaskGraph, TaskNode, Edge, PlanDelta
|
||
from air_runtime.events import EventLog
|
||
from air_runtime.paths import event_log_path
|
||
from air_runtime.utils import ordered_unique
|
||
|
||
|
||
# INV-16: 弱模型安全 — 危险词列表
|
||
DANGEROUS_TASK_WORDS = [
|
||
"清理", "清除", "删除所有", "重构整个", "重写全部",
|
||
"clear", "delete all", "remove all", "rewrite entire", "refactor whole",
|
||
]
|
||
DIR_LEVEL_FILE_SCOPE_PATTERNS = [
|
||
"src/", "lib/", "include/", "tests/", "modules/",
|
||
]
|
||
|
||
def _audit_task_safety(task_id: str, task_text: str, files_dirs: str, done_when: str) -> list[str]:
|
||
"""INV-16: 扫描单个任务的危险词和宽泛文件范围，返回警告列表。"""
|
||
warnings = []
|
||
text_lower = task_text.lower()
|
||
for word in DANGEROUS_TASK_WORDS:
|
||
if word.lower() in text_lower:
|
||
warnings.append(f"[{task_id}] 含危险词 '{word}' — 建议改为精确描述（如'修改 CMakeLists.txt 去掉 sipclient 依赖'而非'清理{word}'）")
|
||
for pattern in DIR_LEVEL_FILE_SCOPE_PATTERNS:
|
||
if pattern in files_dirs and not any(f.endswith(ext) for ext in [".cpp", ".hpp", ".h", ".py", ".ts", ".md", ".json", ".txt", ".cmake"] for f in files_dirs.split(",")):
|
||
if not done_when or "不" not in done_when:
|
||
warnings.append(f"[{task_id}] 文件范围含目录级 '{pattern}' 但 done_when 无否定约束 — Worker 可能误解文件范围")
|
||
break
|
||
return warnings
|
||
|
||
|
||
class ArcPhaseGate:
|
||
"""三阶段门控：discussing → proposing → confirmed。
|
||
execution-plan.json 仅在 phase=confirmed 时允许写入。
|
||
"""
|
||
PHASES = ["discussing", "proposing", "confirmed"]
|
||
|
||
def __init__(self, state_path: Path):
|
||
self._state_path = state_path
|
||
|
||
@property
|
||
def current_phase(self) -> str:
|
||
data = safe_json_load(self._state_path) or {}
|
||
return data.get("arcPhase", "discussing")
|
||
|
||
def advance_to(self, phase: str) -> None:
|
||
if phase not in self.PHASES:
|
||
raise ValueError(f"invalid phase: {phase!r}")
|
||
idx_current = self.PHASES.index(self.current_phase)
|
||
idx_target = self.PHASES.index(phase)
|
||
if idx_target <= idx_current:
|
||
return
|
||
data = safe_json_load(self._state_path) or {}
|
||
data["arcPhase"] = phase
|
||
data[f"arcPhase_{phase}At"] = now_iso()
|
||
atomic_json_write(self._state_path, data)
|
||
|
||
def can_write_plan(self) -> bool:
|
||
return self.current_phase == "confirmed"
|
||
|
||
def confirm_architecture(self, user_confirmation: str) -> bool:
|
||
"""检查用户确认文本中的关键词，确认后推进到 confirmed。"""
|
||
confirm_keywords = ["确认", "可以", "同意", "confirm", "yes", "ok", "好的", "没问题"]
|
||
if any(kw in user_confirmation.lower() for kw in confirm_keywords):
|
||
self.advance_to("confirmed")
|
||
return True
|
||
return False
|
||
|
||
|
||
def _paths(project_root: Path) -> dict[str, Path]:
|
||
root = airplan_root(project_root) / "state" / "airarc"
|
||
return {
|
||
"root": root,
|
||
"state": root / "state.json",
|
||
"reviews_dir": root / "reviews",
|
||
"execution_plan_json": root / "reviews" / "execution-plan.json",
|
||
"execution_plan_md": root / "reviews" / "execution-plan.md",
|
||
"plan_delta_json": root / "reviews" / "plan-delta.json",
|
||
"task_graph_json": root / "reviews" / "task-graph.json",
|
||
}
|
||
|
||
|
||
def _ensure_dirs(paths: dict[str, Path]) -> None:
|
||
paths["reviews_dir"].mkdir(parents=True, exist_ok=True)
|
||
|
||
|
||
def _export_task_graph_json(graph: TaskGraph, path: Path) -> None:
|
||
data = {
|
||
"dispatchFrozen": graph.dispatch_frozen, # P1-21
|
||
"nodes": {nid: {"id": n.id, "status": n.status, "task": n.task,
|
||
"filesDirs": n.files_dirs, "doneWhen": n.done_when,
|
||
"inDegree": n.in_degree, "outEdges": n.out_edges,
|
||
"writeSet": n.write_set, "testRequired": n.test_required,
|
||
"adrRefs": n.adr_refs} # P1-21
|
||
for nid, n in graph.nodes.items()},
|
||
"edges": [{"source": e.source, "target": e.target, "kind": e.kind} for e in graph.edges],
|
||
}
|
||
atomic_json_write(path, data)
|
||
|
||
|
||
def _build_graph_from_todo(todo_path: Path) -> tuple[TaskGraph, list[str]]:
|
||
"""从 todo.md 构建初始 DAG，返回图和未满足 Done When 条件的任务列表。"""
|
||
from air_runtime.todo_parser import parse_tasks
|
||
tasks = parse_tasks(todo_path)
|
||
graph = TaskGraph()
|
||
violations = [] # P1-19.1: 记录 Done When 不含"测试通过"的任务
|
||
|
||
for t in tasks:
|
||
# P1-21: 从 todo.md ADR 列提取 adr_refs
|
||
adr_refs = []
|
||
if hasattr(t, "adr") and t.adr:
|
||
adr_refs = [a.strip() for a in t.adr.split(",") if a.strip()]
|
||
node = TaskNode(
|
||
id=t.task_id, status=t.status, task=t.task,
|
||
files_dirs=t.files_dirs, done_when=t.done_when,
|
||
adr_refs=adr_refs,
|
||
)
|
||
graph.add_node(node)
|
||
|
||
# P1-19.1: Done When 必须包含"测试通过"
|
||
if t.done_when and "测试通过" not in t.done_when:
|
||
violations.append(t.task_id)
|
||
|
||
# P1-19.1: 注入边界测试任务
|
||
_inject_boundary_tests(graph, tasks)
|
||
|
||
return graph, violations
|
||
|
||
|
||
def _inject_boundary_tests(graph: TaskGraph, tasks: list) -> None:
|
||
"""P1-19.1: 为每个模块边界注入测试任务。
|
||
|
||
规则：
|
||
- 每个模块的公共接口必须有对应的接口测试任务
|
||
- 每个模块的核心逻辑必须有对应的单元测试任务
|
||
- 测试任务标记 test_required=True
|
||
- Done When 必须包含"测试通过"
|
||
"""
|
||
# 从 tasks 提取模块信息（通过 files_dirs 推断模块）
|
||
module_files: dict[str, set[str]] = {}
|
||
for t in tasks:
|
||
if t.files_dirs:
|
||
for fd in t.files_dirs.split(","):
|
||
fd = fd.strip()
|
||
if fd:
|
||
# 取第一级目录作为模块名
|
||
parts = fd.split("/")
|
||
if len(parts) > 1:
|
||
module = parts[0]
|
||
else:
|
||
module = fd.split(".")[0] if "." in fd else fd
|
||
module_files.setdefault(module, set()).add(fd)
|
||
|
||
# 为每个模块创建测试任务
|
||
test_counter = 0
|
||
for module, files in module_files.items():
|
||
if not files:
|
||
continue
|
||
|
||
# 接口测试任务（模块边界）
|
||
test_counter += 1
|
||
interface_test_id = f"T-TEST-{test_counter:03d}"
|
||
interface_test_node = TaskNode(
|
||
id=interface_test_id,
|
||
status="TODO",
|
||
task=f"[边界测试] {module} 模块接口测试",
|
||
files_dirs=",".join(sorted(files)),
|
||
done_when="测试通过",
|
||
test_required=True,
|
||
)
|
||
graph.add_node(interface_test_node)
|
||
|
||
# 单元测试任务（核心逻辑）
|
||
test_counter += 1
|
||
unit_test_id = f"T-TEST-{test_counter:03d}"
|
||
unit_test_node = TaskNode(
|
||
id=unit_test_id,
|
||
status="TODO",
|
||
task=f"[单元测试] {module} 模块核心逻辑测试",
|
||
files_dirs=",".join(sorted(files)),
|
||
done_when="测试通过",
|
||
test_required=True,
|
||
)
|
||
graph.add_node(unit_test_node)
|
||
|
||
# 添加依赖边：实现任务 → 测试任务
|
||
for tid, node in graph.nodes.items():
|
||
if tid.startswith("T-TEST-"):
|
||
continue
|
||
# 检查是否属于同一模块
|
||
node_files = set(node.files_dirs.split(",")) if node.files_dirs else set()
|
||
if node_files & files: # 有交集，说明是同一模块的任务
|
||
graph.add_edge(Edge(source=tid, target=interface_test_id, kind="dependency"))
|
||
graph.add_edge(Edge(source=tid, target=unit_test_id, kind="dependency"))
|
||
|
||
|
||
def enter_mode(project_root: Path) -> dict:
|
||
paths = _paths(project_root)
|
||
_ensure_dirs(paths)
|
||
payload = {"enabled": True, "updatedAt": now_iso(), "projectRoot": str(project_root)}
|
||
atomic_json_write(paths["state"], payload)
|
||
return {"state_path": str(paths["state"])}
|
||
|
||
|
||
def parallel_review_mode(project_root: Path, todo_path: Path) -> dict:
|
||
paths = _paths(project_root)
|
||
_ensure_dirs(paths)
|
||
|
||
# 三阶段门控：仅在 confirmed 阶段允许写入 execution-plan.json
|
||
gate = ArcPhaseGate(paths["state"])
|
||
if not gate.can_write_plan():
|
||
return {
|
||
"blocked": True,
|
||
"reason": f"arc phase is '{gate.current_phase}', must be 'confirmed' before generating plan",
|
||
"currentPhase": gate.current_phase,
|
||
}
|
||
|
||
review = build_parallel_review(todo_path)
|
||
review_json_path = paths["reviews_dir"] / "parallel-review.json"
|
||
review_md_path = paths["reviews_dir"] / "parallel-review.md"
|
||
atomic_json_write(review_json_path, review.to_dict())
|
||
review_md_path.write_text(render_review_markdown(review), encoding="utf-8")
|
||
|
||
# 构建 DAG（包含边界测试任务注入）
|
||
graph, done_when_violations = _build_graph_from_todo(todo_path)
|
||
for edge_info in review.to_dict().get("edges", []):
|
||
graph.add_edge(Edge(source=edge_info["source"], target=edge_info["target"],
|
||
kind=edge_info.get("kind", "dependency")))
|
||
|
||
# INV-16: 弱模型安全审计 — 扫描所有 TODO 任务的危险词和宽泛文件范围
|
||
safety_warnings = []
|
||
for nid, node in graph.nodes.items():
|
||
if node.status == "TODO":
|
||
safety_warnings.extend(
|
||
_audit_task_safety(nid, node.task, node.files_dirs, node.done_when)
|
||
)
|
||
|
||
_export_task_graph_json(graph, paths["task_graph_json"])
|
||
|
||
# 生成执行计划
|
||
selected_tasks = review.parallel_groups[0].task_ids if review.parallel_groups else []
|
||
execution_plan = {
|
||
"generatedAt": now_iso(), "projectRoot": str(project_root),
|
||
"todoPath": str(todo_path), "planSource": "airarc-post-plan-review",
|
||
"parallelReview": review.to_dict(),
|
||
"selectedTasks": selected_tasks,
|
||
"parallelGroups": [g.to_dict() for g in review.parallel_groups],
|
||
"conflicts": [c.to_dict() for c in review.conflicts],
|
||
"serializationPoints": review.serialization_points,
|
||
"doneWhenViolations": done_when_violations, # P1-19.1: Done When 不含"测试通过"的任务
|
||
"boundaryTestTasks": [n.id for n in graph.nodes.values() if n.test_required],
|
||
"safetyWarnings": safety_warnings, # INV-16: 弱模型安全警告
|
||
}
|
||
atomic_json_write(paths["execution_plan_json"], execution_plan)
|
||
|
||
markdown_lines = [
|
||
"# AirArc Execution Plan", "",
|
||
f"- Generated: `{execution_plan['generatedAt']}`",
|
||
f"- Plan Source: `{execution_plan['planSource']}`", "",
|
||
"## Selected Tasks",
|
||
]
|
||
for tid in selected_tasks:
|
||
markdown_lines.append(f"- `{tid}`")
|
||
markdown_lines.extend(["", "## Parallel Groups"])
|
||
for g in review.parallel_groups:
|
||
markdown_lines.append(f"- `{g.name}`: {', '.join(g.task_ids)} — {g.reason}")
|
||
paths["execution_plan_md"].write_text("\n".join(markdown_lines) + "\n", encoding="utf-8")
|
||
|
||
# 保留 arcPhase 字段，避免 phase 被 reset 为 discussing
|
||
existing_state = safe_json_load(paths["state"]) or {}
|
||
state_payload = {**existing_state, "enabled": True, "updatedAt": now_iso(),
|
||
"projectRoot": str(project_root)}
|
||
atomic_json_write(paths["state"], state_payload)
|
||
return {"json_path": str(review_json_path), "markdown_path": str(review_md_path),
|
||
"execution_plan_json_path": str(paths["execution_plan_json"]),
|
||
"parallel_group_count": len(review.parallel_groups),
|
||
"conflict_count": len(review.conflicts),
|
||
"done_when_violations": done_when_violations,
|
||
"boundary_test_task_count": len([n for n in graph.nodes.values() if n.test_required])}
|
||
|
||
|
||
def incremental_replan_mode(project_root: Path, todo_path: Path, previous_graph_path: Path | None = None) -> dict:
|
||
"""增量重规划：产出 PlanDelta 并喂回 prev graph，再写 plan-delta.json。"""
|
||
paths = _paths(project_root)
|
||
_ensure_dirs(paths)
|
||
|
||
review = build_parallel_review(todo_path)
|
||
new_graph, _ = _build_graph_from_todo(todo_path)
|
||
for edge_info in review.to_dict().get("edges", []):
|
||
new_graph.add_edge(Edge(source=edge_info["source"], target=edge_info["target"],
|
||
kind=edge_info.get("kind", "dependency")))
|
||
|
||
delta = PlanDelta()
|
||
if previous_graph_path and previous_graph_path.exists():
|
||
prev_graph = TaskGraph.load(previous_graph_path)
|
||
delta = new_graph.diff(prev_graph)
|
||
# 关键：把 delta 喂回去给 prev graph，保留已调度状态
|
||
if previous_graph_path.exists():
|
||
prev_graph.apply_delta(delta)
|
||
_export_task_graph_json(prev_graph, previous_graph_path)
|
||
else:
|
||
delta.added_tasks = list(new_graph.nodes.values())
|
||
delta.edge_changes.added = list(new_graph.edges)
|
||
|
||
atomic_json_write(paths["plan_delta_json"], {
|
||
"generatedAt": now_iso(),
|
||
"removedTasks": delta.removed_tasks,
|
||
"addedTasks": [{"id": n.id, "task": n.task, "filesDirs": n.files_dirs,
|
||
"doneWhen": n.done_when, "writeSet": n.write_set}
|
||
for n in delta.added_tasks],
|
||
"modifiedTasks": [{"id": n.id, "task": n.task, "filesDirs": n.files_dirs,
|
||
"doneWhen": n.done_when, "writeSet": n.write_set}
|
||
for n in delta.modified_tasks],
|
||
"edgeChanges": {
|
||
"added": [{"source": e.source, "target": e.target, "kind": e.kind}
|
||
for e in delta.edge_changes.added],
|
||
"removed": [{"source": e.source, "target": e.target, "kind": e.kind}
|
||
for e in delta.edge_changes.removed],
|
||
},
|
||
})
|
||
|
||
_export_task_graph_json(new_graph, paths["task_graph_json"])
|
||
|
||
log = EventLog(event_log_path(project_root))
|
||
log.emit("arc.replanned", {"delta_added": len(delta.added_tasks),
|
||
"delta_removed": len(delta.removed_tasks),
|
||
"delta_modified": len(delta.modified_tasks),
|
||
"edges_added": len(delta.edge_changes.added),
|
||
"edges_removed": len(delta.edge_changes.removed)})
|
||
|
||
return {"plan_delta_json_path": str(paths["plan_delta_json"]),
|
||
"task_graph_json_path": str(paths["task_graph_json"]),
|
||
"added_count": len(delta.added_tasks),
|
||
"removed_count": len(delta.removed_tasks),
|
||
"modified_count": len(delta.modified_tasks),
|
||
"edges_added": len(delta.edge_changes.added),
|
||
"edges_removed": len(delta.edge_changes.removed)}
|
||
|
||
|
||
def main(args) -> None:
|
||
project_root = Path(args.project).expanduser().resolve()
|
||
sub = args.sub or "status"
|
||
|
||
if sub == "enter":
|
||
result = enter_mode(project_root)
|
||
print("airplan_mode=arc")
|
||
print(f"state_path={result['state_path']}")
|
||
elif sub == "parallel-review":
|
||
tpath = Path(args.todo).expanduser().resolve() if args.todo else get_todo_path(project_root)
|
||
result = parallel_review_mode(project_root, tpath)
|
||
print("airplan_mode=arc")
|
||
print(f"json_path={result['json_path']}")
|
||
print(f"parallel_group_count={result['parallel_group_count']}")
|
||
print(f"conflict_count={result['conflict_count']}")
|
||
elif sub == "incremental-replan":
|
||
tpath = Path(args.todo).expanduser().resolve() if args.todo else get_todo_path(project_root)
|
||
prev = _paths(project_root)["task_graph_json"]
|
||
result = incremental_replan_mode(project_root, tpath, prev)
|
||
print("airplan_mode=arc")
|
||
print(f"plan_delta_path={result['plan_delta_json_path']}")
|
||
print(f"added={result['added_count']} removed={result['removed_count']}")
|
||
else:
|
||
paths = _paths(project_root)
|
||
state = safe_json_load(paths["state"]) or {}
|
||
print(f"airplan_mode=arc")
|
||
print(f"enabled={state.get('enabled', False)}")
|