fix(scheduler): append actionable hint to MCP utility-model errors

[Battleplus]

这次改了什么

当 schedule_set_pre_run_hook 因为所有 utility-model 候选全部失败时，原始错误信息直接返回给 MCP 调用方，没有任何操作指引。Agent 无法知道可以通过直接传 script 参数绕过失败的生成链。

本修复为 4 种 utility-model 错误码追加可操作提示，明确建议传 script 绕过生成，解决 #3317。

变更类型

• feat 新功能
• fix 缺陷修复
• refactor / perf 重构或性能优化
• docs / test / chore 文档、测试或工程维护
• 其他：

范围

• 关联 Issue：MCP 只传 description 时，前置钩依靠隐藏 utility 链生成且失败不可操作 #3317
• 本 PR 包含：
  • packages/lizi-mcps/src/scheduler/errors.ts：为 4 种 utility-model 错误码追加可操作提示
  • packages/lizi-mcps/src/scheduler/errors.test.ts：更新已有测试 + 新增提示验证测试
• 明确不包含：不修改错误分类逻辑；不修改 scheduler 核心流程

怎么验证的

1. 运行 npx vitest run packages/lizi-mcps/src/scheduler/errors.test.ts，8/8 测试通过
2. 验证每种错误码（ALL_CANDIDATES_FAILED、NO_CANDIDATE、TIMEOUT、EMPTY_RESPONSE）的错误消息都包含 "passing the script parameter" 提示
3. 验证原始错误信息仍然保留在提示之后

风险

风险极低——仅向现有错误消息追加文本，无控制流变更，无 breaking change。

[greptile-apps]

Greptile Summary

This PR improves scheduler MCP diagnostics by appending an actionable direct-script workaround to all utility-model generation errors.

• Preserves the existing structured utility-model error codes and original diagnostics.
• Adds code-specific hints explaining that the caller can bypass generation through the script parameter.
• Updates parameterized tests to cover all four utility-model error codes and verify the recovery guidance.

Confidence Score: 5/5

The PR appears safe to merge because the new guidance matches the exposed MCP parameter and direct script input bypasses utility-model generation.

The change only extends utility-model error messages with fixed recovery text, preserves structured codes and original diagnostics, and covers every affected code with focused tests.

Important Files Changed

Filename	Overview
packages/lizi-mcps/src/scheduler/errors.ts	Appends accurate direct-script recovery guidance while preserving existing error classification and diagnostic text.
packages/lizi-mcps/src/scheduler/errors.test.ts	Verifies code preservation, original diagnostic retention, and recovery hints for all four utility-model error variants.

_{Reviews (1): Last reviewed commit: "fix(scheduler): append actionable hint t..." | Re-trigger Greptile}

[MagicLizi]

格式门未通过（P1）

• .github/PULL_REQUEST_TEMPLATE.md: Description 缺段落「这次改了什么」「怎么验证的」「风险」。请按仓库 PR 模板补齐后再推一次。

本次只拦模板完整性，未进入代码审查。

[MagicLizi]

Reviewed. Utility-model errors now carry an actionable script bypass hint; tests cover the four codes; no P0/P1.

[MagicLizi]

合了。工具模型挂掉时不再只丢一行诊断，调用方能直接看见可以用 script 绕开生成——这种小提示能少绕不少圈子。