Back to Details

kf-delegate-specialist

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Delegate Specialist

委托专业子Agent

Use a focused sub-agent to apply a domain-appropriate method, work with an isolated evidence set, or provide an independent judgment on a specific specialty. This is especially useful when a task needs substantial non-engineering knowledge that is not central to the parent agent's working context.
A sub-agent has the same underlying platform capabilities. A specialist framing can surface more relevant concepts and reasoning patterns, but it does not guarantee additional knowledge, credentials, correctness, or authority.
使用专注的子Agent来应用适合特定领域的方法、处理独立的证据集，或针对特定专业领域提供独立判断。当任务需要大量非工程类知识，而这些知识并非父Agent工作核心时，这种方法尤为有用。
子Agent具备与父Agent相同的底层平台能力。专业视角的设定可以挖掘出更相关的概念和推理模式，但并不保证其拥有额外知识、资质、正确性或权威性。

When delegation helps

委托适用场景

  • The user explicitly requests a specialist, sub-agent, delegation, or independent domain review.
  • A task contains a clearly separable specialty with its own domain method, evidence, and deliverable.
  • A non-engineering question would benefit from dedicated research or analysis without pulling a large domain corpus into the parent agent's working context.
  • Independent analysis would reduce confirmation bias or protect the parent agent's working context.
  • The available sub-agent mechanism and current policy permit delegation.
Continue directly when the work is ordinary, tightly coupled to the main task, smaller than the handoff overhead, or cannot be safely separated.
  • 用户明确请求专家、子Agent、委托服务或独立领域审查。
  • 任务包含一个清晰可分离的专业领域，该领域拥有自身的方法、证据和交付成果。
  • 非工程类问题需要专门的研究或分析，且无需将庞大的领域语料引入父Agent的工作环境。
  • 独立分析有助于减少确认偏差，或保护父Agent的工作环境不受干扰。
  • 现有的子Agent机制和当前政策允许进行委托。
若工作属于常规任务、与主任务紧密耦合、规模小于交接成本，或无法安全分离，则直接继续执行即可。

Choose a mode

选择委托模式

  • Perspective: apply a domain framework or vocabulary to supplied material without external research or side effects.
  • Research: gather and synthesize evidence. Prefer authoritative primary sources for current, disputed, regulated, or high-stakes claims, and record relevant publication, effective, or access dates.
  • Independent review: evaluate an existing conclusion or artifact without being primed with the parent's preferred answer. Look for counterevidence and plausible alternatives.
  • Execution: perform a bounded change or action only with explicit authorization, exclusive ownership, and appropriate validation.
Use the fewest specialists needed. Do not create recursive delegation chains by default.
  • 视角模式：将领域框架或词汇应用于提供的材料，无需进行外部研究或产生副作用。
  • 研究模式：收集并整合证据。对于时效性强、存在争议、受监管或高风险的主张，优先使用权威的原始来源，并记录相关的发布日期、生效日期或获取日期。
  • 独立审查模式：在不预先告知父Agent偏好答案的情况下，评估现有结论或成果。需寻找反证和合理的替代方案。
  • 执行模式：仅在获得明确授权、拥有专属权限并经过适当验证的情况下，执行有限的变更或操作。
尽量使用最少的专业子Agent。默认情况下，不要创建递归委托链。

Workflow

工作流程

  1. Define the exact specialty, select a mode, and state which domain method, evidence base, or independent viewpoint makes delegation useful.
  2. Isolate one concrete objective that can be completed independently.
  3. Prepare a handoff containing:
    • objective and expected deliverable
    • domain role and useful analytical method, not an inflated persona
    • minimum relevant context, known facts, and evidence
    • source quality, recency, citation, and tool expectations
    • allowed files, tools, and side effects
    • constraints, non-goals, and compatibility requirements
    • validation expectations and stopping condition
  4. Name the specialist after the work, such as 
    tax-rule-researcher
    , 
    database-migration-reviewer
    , or 
    accessibility-verifier
    ; avoid inflated or vague personas.
  5. Spawn the specialist with the handoff. Keep perspective, research, and review modes read-only; for execution, establish exclusive file or action ownership.
  6. Continue useful parent work that does not overlap while the specialist runs.
  7. Require the result to distinguish sourced facts, inferences, assumptions, and unresolved questions. It should cite material evidence, surface counterevidence or alternatives, and state confidence and applicability limits.
  8. Review the returned reasoning, sources, changes, and validation. Challenge unsupported conclusions, check time-sensitive claims against current evidence, and reconcile conflicts with repository or user-provided facts.
  9. Integrate only the necessary result, run parent-level verification, and report what was delegated and how it was validated.
  1. 明确具体的专业领域，选择委托模式，并说明为何该领域的方法、证据库或独立视角使得委托具有价值。
  2. 确定一个可独立完成的具体目标。
  3. 准备交接材料，包含：
    • 目标和预期交付成果
    • 领域角色和实用分析方法，而非夸大的人设
    • 最少的相关背景、已知事实和证据
    • 来源质量、时效性、引用要求和工具预期
    • 允许使用的文件、工具和可产生的副作用
    • 约束条件、非目标需求和兼容性要求
    • 验证标准和终止条件
  4. 根据工作内容为专业子Agent命名，例如
    tax-rule-researcher
    、
    database-migration-reviewer
    或
    accessibility-verifier
    ；避免使用夸大或模糊的人设名称。
  5. 携带交接材料启动专业子Agent。视角、研究和审查模式保持只读；执行模式需建立专属文件或操作权限。
  6. 在专业子Agent运行期间，继续执行父Agent中不重叠的有效工作。
  7. 要求返回结果区分已查证事实、推论、假设和未解决问题。结果需引用实质性证据、呈现反证或替代方案，并说明可信度和适用范围限制。
  8. 审查返回的推理过程、来源、变更和验证结果。质疑无依据的结论，核查时效性主张是否符合当前证据，并协调与知识库或用户提供事实的冲突。
  9. 仅整合必要的结果，执行父Agent层面的验证，并报告委托内容及验证方式。

Constraints

约束条件

  • Delegation does not expand the user's authorization, filesystem scope, external permissions, or approval boundaries.
  • Do not send secrets, irrelevant private data, or unnecessary repository context.
  • Do not let multiple agents edit the same files concurrently.
  • Treat specialist output as evidence to review, not an automatically correct answer.
  • Do not treat a confident tone, specialist title, or unsourced domain terminology as evidence.
  • The parent agent remains responsible for the final result and must disclose unresolved uncertainty.
  • When a domain requires licensed, regulated, or human expert judgment, explain that limitation instead of presenting a sub-agent as a substitute.
  • 委托不会扩大用户的授权范围、文件系统权限、外部权限或审批边界。
  • 不得发送机密信息、无关的私人数据或不必要的知识库背景。
  • 禁止多个Agent同时编辑同一文件。
  • 将专业子Agent的输出视为待审查的证据，而非自动正确的答案。
  • 不得将自信的语气、专业头衔或无来源的领域术语视为有效证据。
  • 父Agent仍对最终结果负责，必须披露未解决的不确定性。
  • 当领域需要持证、受监管或人类专家判断时，需说明这一限制，而非将子Agent作为替代方案。