thrifty
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseThrifty Task Delegation
低成本任务委托
Use this skill to delegate tasks to opencode/glm-5, a cost-effective model ideal for:
- Simple research and information gathering
- Code exploration and file searches
- Document processing and summarization
- Routine automation tasks
- Parallel processing of independent subtasks
使用该技能将任务委托给opencode/glm-5，这是一款高性价比模型，非常适合以下场景：
- 简单研究与信息收集
- 代码探索与文件搜索
- 文档处理与摘要生成
- 常规自动化任务
- 独立子任务的并行处理
Usage Pattern
使用模式
Invoke glm-5 via the task tool:
task({
description: "Brief task summary",
prompt: "Detailed instructions for the agent",
subagent_type: "general" // glm-5 is the default model
})通过task工具调用glm-5：
task({
description: "Brief task summary",
prompt: "Detailed instructions for the agent",
subagent_type: "general" // glm-5 is the default model
})Cost Guidelines
成本指南
| Task Type | Recommended Model | Est. Cost |
|---|---|---|
| Simple search/explore | glm-5 (default) | Low |
| Code review | code-review-glm | Low |
| Complex architecture | oracle | Medium |
| Deep debugging | code-review-opus | Higher |
| 任务类型 | 推荐模型 | 预估成本 |
|---|---|---|
| 简单搜索/探索 | glm-5（默认） | 低 |
| 代码审查 | code-review-glm | 低 |
| 复杂架构设计 | oracle | 中 |
| 深度调试 | code-review-opus | 较高 |
Best Practices
最佳实践
- Break large tasks into smaller chunks - glm-5 handles focused tasks well
- Provide clear, specific instructions - reduces iteration and context usage
- Use for parallel work - delegate multiple independent tasks simultaneously
- Verify complex outputs - have a stronger model review critical results
- Batch similar operations - group related file operations together
- 将大型任务拆分为更小的模块 - glm-5擅长处理聚焦型任务
- 提供清晰、明确的指令 - 减少迭代次数和上下文使用量
- 用于并行工作 - 同时委托多个独立任务
- 验证复杂输出结果 - 让更强大的模型审查关键结果
- 批量处理相似操作 - 将相关的文件操作分组处理
Example Delegations
委托示例
File exploration:
task({
description: "Find auth-related files",
prompt: "Search the codebase for all files related to authentication. Return a list of file paths and brief descriptions of what each does.",
subagent_type: "general"
})Documentation summarization:
task({
description: "Summarize API docs",
prompt: "Read docs/api.md and provide a 3-bullet summary of the main endpoints and their purposes.",
subagent_type: "general"
})Parallel processing:
// Delegate multiple independent tasks concurrently
const tasks = [
task({ description: "Analyze utils.ts", prompt: "...", subagent_type: "general" }),
task({ description: "Analyze helpers.ts", prompt: "...", subagent_type: "general" }),
task({ description: "Analyze constants.ts", prompt: "...", subagent_type: "general" })
];
await Promise.all(tasks);文件探索：
task({
description: "Find auth-related files",
prompt: "Search the codebase for all files related to authentication. Return a list of file paths and brief descriptions of what each does.",
subagent_type: "general"
})文档摘要生成：
task({
description: "Summarize API docs",
prompt: "Read docs/api.md and provide a 3-bullet summary of the main endpoints and their purposes.",
subagent_type: "general"
})并行处理：
// Delegate multiple independent tasks concurrently
const tasks = [
task({ description: "Analyze utils.ts", prompt: "...", subagent_type: "general" }),
task({ description: "Analyze helpers.ts", prompt: "...", subagent_type: "general" }),
task({ description: "Analyze constants.ts", prompt: "...", subagent_type: "general" })
];
await Promise.all(tasks);When NOT to Use glm-5
不适合使用glm-5的场景
- Complex multi-step architectural decisions
- Security-critical code review
- Novel problem-solving requiring creativity
- Tasks requiring deep contextual understanding
For these, use , , or other specialized subagents instead.
oraclecode-review-opus- 复杂的多步骤架构决策
- 安全关键型代码审查
- 需要创造力的创新性问题解决
- 需要深度上下文理解的任务
对于这些场景，请改用、或其他专用子代理。
oraclecode-review-opus