harness2: upstream gap fill — trivial selection fix, session budget, run_tag validation, preflight check

- select_final trivial condition now k==1 and single-candidate pool,
  regardless of whether a selector is provided (previously gated on
  selector is None, so wired runs always paid for a useless selector
  task at k=1)
- Budget gains max_sessions gate (probe + stage-attempt sessions;
  fmt2 corrections and selector/reflector sessions exempt), the
  backstop for unpriced models where the USD gate never binds
- validate_run_tag (charset, no leading '-', no s<N> seed prefix)
  enforced at run_improvement_via_tasks and the CLI
- CLI --check preflight (base harness, source dir, prompt, tag,
  control-plane /health probe) without creating tasks
- ModeResult.to_dict now includes rollout summaries so a lost probe
  is visible in full_result instead of only on the server
- CLI plan counts the selector only when k>1

cloud/harness2_tasks.py

@@ -62,6 +62,7 @@ parallel 一轮 = 1 base + 2 候选 + 1 终版探针 + 3 编辑 + 1 选择 +
 from __future__ import annotations
 
 import json
+import re
 import time
 import urllib.error
 import urllib.request
@@ -91,6 +92,67 @@ DEFAULT_TASK_TIMEOUT_SECONDS = 3600
 DEFAULT_POLL_SECONDS = 10
 TRACE_MAX_CHARS = 50000
 
+_TAG_RX = re.compile(r"[A-Za-z0-9_.][A-Za-z0-9_.\-]*\Z")
+_SEED_TAG_RX = re.compile(r"s\d+(-|\Z)")
+
+
+def validate_run_tag(tag: str) -> None:
+    """校验 run_tag 形状（上游 schedule.validate_tag 的对照补齐）。
+
+    run_tag 会进幂等键与路径段：空串表示“调用方稍后生成”，不
+    校验；非空时只允许字母数字与 _ . -、首字符不能是 -、不能
+    以 s + 数字开头（上游用 s<n> 段表示 seed，撞名会把两轮的
+    产物算到同一命名空间）。
+    """
+    if not tag:
+        return
+    if not _TAG_RX.match(tag):
+        raise ValueError(
+            f"run_tag {tag!r} must match [A-Za-z0-9_.-] with no leading '-' "
+            "and no '/'")
+    if _SEED_TAG_RX.match(tag):
+        raise ValueError(
+            f"run_tag {tag!r} must not start with a seed segment like 's1'")
+
+
+def preflight_checks(*, base_harness_dir: str | Path, source_dir: str | Path | None = None,
+                     prompt: str = "", run_tag: str = "",
+                     base_url: str | None = None) -> tuple[list[str], list[str]]:
+    """建任务前的本地 + 可达性预检（2026-10-29 对照补齐）。
+
+    返回 (errors, warnings)：errors 非空时调用方不得建任务；
+    warnings 只告警（health 探针失败、run_tag 为空待生成等不
+    拦任务）。本函数不读密钥、不建任务，--check 可离线跑。
+    """
+    errors: list[str] = []
+    warnings: list[str] = []
+    base_dir = Path(base_harness_dir)
+    agents = base_dir / "AGENTS.md"
+    if not base_dir.is_dir():
+        errors.append(f"base-harness-dir 不存在或不是目录：{base_harness_dir}")
+    elif not agents.is_file() or not agents.read_text(
+            encoding="utf-8", errors="replace").strip():
+        errors.append(f"base-harness-dir 里没有可用的 AGENTS.md：{base_harness_dir}")
+    if source_dir is not None and str(source_dir) != "":
+        if not Path(source_dir).is_dir():
+            errors.append(f"source-dir 不存在或不是目录：{source_dir}")
+    if not (prompt or "").strip():
+        errors.append("需要 --prompt 或 --prompt-file（任务提示为空）")
+    if run_tag:
+        try:
+            validate_run_tag(run_tag)
+        except ValueError as exc:
+            errors.append(str(exc))
+    if base_url:
+        try:
+            req = urllib.request.Request(base_url.rstrip("/") + "/health")
+            with urllib.request.urlopen(req, timeout=10) as resp:
+                if resp.status != 200:
+                    warnings.append(f"控制面 health 非 200：{resp.status}")
+        except Exception as exc:
+            warnings.append(f"控制面 health 探针失败（不拦任务）：{type(exc).__name__}: {exc}")
+    return errors, warnings
+
 
 def unwrap_result_text(raw: str) -> str:
     """把服务端 result_text 的 JSON 信封拆开，只取 agent 的最终回复。
@@ -709,6 +771,7 @@ def run_improvement_via_tasks(
     mode: str = "parallel",
     library_root: Path | None = None,
     budget_usd: float | None = None,
+    max_sessions: int | None = None,
     model: str | None = None,
     effort: str | None = None,
     approval: str | None = None,
@@ -739,10 +802,11 @@ def run_improvement_via_tasks(
     selector = CodexSelector(client=client, project_id=project_id, run_tag=run_tag, **shared)
     reflector = CodexReflector(client=client, project_id=project_id, run_tag=run_tag, **shared)
     library = Library(library_root, domain) if library_root is not None else None
+    validate_run_tag(run_tag)
     return improve_task(
         adapter, editor, selector=selector, reflector=reflector,
         work_dir=work_dir, k=k, mode=mode, library=library,
-        auto_apply=False, budget_usd=budget_usd)
+        auto_apply=False, budget_usd=budget_usd, max_sessions=max_sessions)
 
 
 __all__ = [
@@ -750,5 +814,5 @@ __all__ = [
     "TRACE_MAX_CHARS", "TaskState", "TaskClient", "build_multipart",
     "HttpTaskClient", "CompletedTask", "run_task",
     "CodexProbeAdapter", "CodexEditor", "CodexSelector", "CodexReflector",
-    "run_improvement_via_tasks",
+    "run_improvement_via_tasks", "validate_run_tag", "preflight_checks",
 ]