SmallClaw 在个人模式与组织模式下的四个实际使用案例 Four Practical Use Cases of SmallClaw in Personal Mode and Organization Mode

先理解 SmallClaw 的工作方式 Start with How SmallClaw Works

用户安装并运行 SmallClaw 后，可以直接在 Agent Console 中操作；如果电脑保持运行，也可以通过 WhatsApp、Telegram 等已接入的 messaging tools 远程下达任务。无论命令从哪里发出，任务执行、项目文件读取和结果存储都发生在这台本地设备上。 After SmallClaw is installed and running, the user can operate it directly in the Agent Console. If the machine stays on, tasks can also be sent remotely through connected messaging tools such as WhatsApp or Telegram. No matter where the instruction comes from, execution, file access, and result storage all happen on the same local device.

案例一：Personal Mode 下的留学申请 Use Case 1: Study Abroad Applications in Personal Mode

在 Personal Mode 里，SmallClaw 不是替用户做最终判断，而是把一个长期申请项目组织起来，并持续执行其中的具体工作。用户先创建一个项目，例如 2027 Fall US Master Application - Geoff，然后建立对应的 Work Items。 In Personal Mode, SmallClaw does not replace the user’s final judgment. It organizes a long-running application project and keeps executing the concrete work inside it. The user starts with a project such as 2027 Fall US Master Application - Geoff and then creates the relevant work items.

• Applicant ProfileApplicant Profile
• School List ResearchSchool List Research
• Requirements ComparisonRequirements Comparison
• SOP DraftingSOP Drafting
• Deadline TrackingDeadline Tracking
• Scholarship SearchScholarship Search
• Email CommunicationEmail Communication

接着，用户把成绩单、语言成绩、简历、作品集、实习证明、推荐人信息和以往文书草稿放进项目目录，再让 SmallClaw 先建立 applicant profile，而不是直接问“我适合哪些学校”。例如可以要求它读取现有材料，提取 GPA、专业背景、研究兴趣、预算要求和国家偏好，并把结果分成 confirmed information 与 pending clarification。 Next, the user places transcripts, language scores, CVs, portfolios, internship records, recommender information, and older statement drafts into the project directory. Instead of asking “Which schools fit me?”, the first step is to build an applicant profile from those files, separating confirmed information from pending clarification.

有了 applicant profile 之后，学校池研究、要求对比和分层筛选都可以直接建立在同一项目之上继续推进。用户可以让 SmallClaw 只读取学校官网项目页，生成 school comparison table，提取学制、学费、语言要求、GRE 要求，再把项目分成 reach、match、safe，并标记 over-budget、research-heavy 或 career-oriented。 Once the applicant profile exists, school research, requirements comparison, and portfolio tiering can continue inside the same project. SmallClaw can build a school comparison table from official program pages only, extract tuition and requirements, and classify programs into reach, match, and safe while marking over-budget, research-heavy, or career-oriented options.

文书阶段也不是从零开始。系统可以先读取重点学校官网与申请说明，为不同学校生成独立的 SOP outline，再基于项目目录中的 CV 和经历材料输出例如 UCL_SOP_v1.md 这样的文件。之后每一轮修改都在同一项目中继续，只需要说明新的目标，比如压缩字数、增加 user research 经历或减少空泛的职业规划。 The writing stage also does not restart from scratch. SmallClaw can first read target program pages and application notes, produce separate SOP outlines, and then generate files such as UCL_SOP_v1.md from the project’s CV and experience materials. Later edits stay in the same project and only need the new goal, such as shortening the text or emphasizing a specific research experience.

与此同时，Deadline Tracking 这个 Work Item 可以持续扫描申请截止时间，并把项目目录中已有与缺失的材料整理成 application timeline 和 missing materials list。这样，整个留学申请不会散落在聊天、表格和不同文件夹里，而是作为一个持续运行的项目被 SmallClaw 保存在本地。 At the same time, the Deadline Tracking work item can keep scanning deadlines and checking which materials already exist in the project directory, producing an application timeline and a missing materials list. The whole application process no longer lives across scattered chats, spreadsheets, and folders. It stays as one local project that SmallClaw can keep advancing.

案例二：Personal Mode 下的医学信息整理 Use Case 2: Medical Information Organization in Personal Mode

这个场景里，SmallClaw 也不是用来替代医生做诊断或治疗决策，而是用于管理病例资料、建立时间线、提取检测信息、检索公开资料，并准备沟通材料。用户先创建一个项目，例如 Patient-A Precision Oncology Review，并建立以下 Work Items。 In this scenario, SmallClaw is not used to replace a doctor’s diagnosis or treatment decisions. It is used to manage records, build a timeline, extract test results, search public information, and prepare communication materials. The user might begin with a project such as Patient-A Precision Oncology Review and create the following work items.

• Case TimelineCase Timeline
• Pathology ExtractionPathology Extraction
• Genomic Report ReviewGenomic Report Review
• Treatment HistoryTreatment History
• Literature SearchLiterature Search
• Clinical Trial SearchClinical Trial Search
• MDT BriefingMDT Briefing

把病理报告、基因检测报告、影像摘要、出院小结、既往用药记录和会诊意见放进项目目录后，第一步通常不是直接找文献，而是先建立 case index，并按日期、来源机构、文档类型和重要字段为资料建档。随后，SmallClaw 可以基于这些材料整理 chronological timeline，提取首诊、确诊、影像、治疗开始与结束、疾病进展和基因检测等关键事件。 After pathology reports, genomic reports, imaging summaries, discharge notes, prior treatments, and consultation notes are added to the project directory, the first step is usually not literature search. It is to build a case index and classify documents by date, source, type, and important fields. From there, SmallClaw can build a chronological timeline of diagnosis, imaging, treatment, progression, and genomic testing events.

接下来，Pathology Extraction 与 Genomic Report Review 会分别提取 tumor type、stage、grading、immunohistochemistry markers、driver mutations、resistance-associated mutations、TMB、MSI、fusion events 和 copy number changes，并保留原始报告来源。Treatment History 则把既往治疗按 line of therapy 整理成结构化记录，包括药物组合、起止时间、疗效描述、不良反应和停药原因。 Next, Pathology Extraction and Genomic Report Review can pull out tumor type, stage, grading, immunohistochemistry markers, driver mutations, resistance-associated mutations, TMB, MSI, fusion events, and copy number changes while preserving source references. Treatment History organizes prior therapies by line of treatment, including drug combinations, timing, response notes, adverse events, and reasons for stopping.

等到 Case Timeline、Genomic Summary 和 Treatment History 形成之后，后续的 Literature Search 与 Clinical Trial Search 就不需要重新解释病例背景。系统可以直接基于现有摘要去搜索公开指南、研究资料和招募中的试验，并输出 literature summary、trial list，最后进一步整理成 MDT briefing 和 Questions for Oncologist。新报告加入时，也只需要更新项目中的相应摘要，而不是从头重做整个流程。 Once the Case Timeline, Genomic Summary, and Treatment History are in place, Literature Search and Clinical Trial Search do not need the whole case to be re-explained. SmallClaw can search public guidance, research, and recruiting trials from the existing summaries, produce a literature summary and trial list, and then roll them into an MDT briefing and a question set for the oncologist. When a new report arrives, only the affected summaries need updating instead of restarting the full workflow.

案例三：Organization Mode 下的留学申请服务 Use Case 3: Study Abroad Service Workflow in Organization Mode

到了 Organization Mode，SmallClaw 的重点就不再是单个用户持续发任务，而是让一个项目内部的多个角色围绕同一目标自动移交、自动讨论、逐步形成阶段标准，并把结果提交给 Supervisor。一个典型项目可以是 2027 Fall Graduate Application Service - Student A。 In Organization Mode, the focus shifts from one user continuously issuing tasks to multiple roles inside one project automatically handing off work, discussing trade-offs, forming stage standards, and submitting results to a supervisor. A typical project might be 2027 Fall Graduate Application Service - Student A.

• Applicant Intake CoordinatorApplicant Intake Coordinator
• School Research AnalystSchool Research Analyst
• Strategy PlannerStrategy Planner
• Document SpecialistDocument Specialist
• Submission CoordinatorSubmission Coordinator
• SupervisorSupervisor

项目启动后，Applicant Intake Coordinator 先建立 Applicant Profile，确认哪些背景信息完整、哪些仍待补充，然后把结果移交给 School Research Analyst 和 Strategy Planner。前者负责收集项目、整理 school comparison table 和申请要求，后者则读取这些结果，对学校分层、预算匹配、方向偏差和优先级作出判断。 After the project starts, the Applicant Intake Coordinator builds the applicant profile and confirms what information is complete or still missing, then hands the result to the School Research Analyst and Strategy Planner. The analyst gathers programs and requirements, while the planner reads those outputs and judges tiering, budget fit, direction mismatch, and priorities.

关键在于，这时系统不会机械地一路往下做，而是会出现角色之间的讨论。比如研究员给出初版学校池后，策略角色可能提出“学校过于偏研究型”“保底项目不足”“两所学校明显超预算”“需要补充一年制英国或澳洲项目”等问题。经过讨论，项目内部会形成一个当前阶段的 working standard，例如学校池必须覆盖 reach、match、safe 三层，总项目数控制在 15 到 18 个，至少包含预算可控项目和明确偏就业导向项目。 The key point is that the system does not proceed mechanically from there. Discussion happens between roles. After the analyst proposes an initial school pool, the strategy role may flag issues such as “too research-heavy,” “not enough safety schools,” “two options clearly exceed budget,” or “one-year UK or Australia programs should be added.” From that discussion, the project forms a stage-level working standard.

学校池稳定后，重点学校会移交给 Document Specialist 生成不同版本的 SOP outline；Submission Coordinator 则维护 deadline、材料完整性和外部邮件。如果它发现推荐信缺失或 deadline 逼近，就会把阻塞项回传给前面的角色，并更新 missing materials list 和 application timeline。最后，项目是否“准备完成”不是由某一个角色单方面宣布，而是由各角色形成 completion consensus，再提交给 Supervisor 决定接受、退回或继续讨论。 Once the school pool stabilizes, the target list is handed to the Document Specialist for school-specific SOP outlines, while the Submission Coordinator manages deadlines, completeness, and external communication. If recommendation letters are missing or deadlines are close, the blocker is passed back to earlier roles and recorded in the missing materials list and application timeline. In the end, readiness is not declared by one role alone. It is formed as a completion consensus and then submitted to the Supervisor for acceptance, revision, or another discussion round.

案例四：Organization Mode 下的医学信息服务 Use Case 4: Medical Information Service Workflow in Organization Mode

在医学信息服务场景下，Organization Mode 同样体现为一个病例项目在多个角色之间自动流转、自动讨论，并在推进过程中逐步收敛目标标准。一个典型项目可以是 Patient-A Precision Oncology Case Review。 In medical information services, Organization Mode appears in the same way: a case project flows automatically across multiple roles, triggers internal discussion, and gradually converges on a stage goal. A representative project could be Patient-A Precision Oncology Case Review.

• Case Intake CoordinatorCase Intake Coordinator
• Timeline ManagerTimeline Manager
• Pathology and Genomics AnalystPathology and Genomics Analyst
• Literature and Trial ResearcherLiterature and Trial Researcher
• Clinical Briefing CoordinatorClinical Briefing Coordinator
• SupervisorSupervisor

当用户上传一批病例资料后，Case Intake Coordinator 先建立 Case Index，标注文档日期、来源和类型，并检查是否缺少关键材料。随后，Timeline Manager 建立 chronological timeline，Pathology and Genomics Analyst 从病理与基因检测报告中提取 tumor type、stage、driver mutations、resistance mutations、TMB、MSI 等结构化信息，生成 Pathology Summary 与 Genomic Summary。 Once a batch of records is uploaded, the Case Intake Coordinator builds the Case Index, labels dates, sources, and document types, and checks for missing key materials. Then the Timeline Manager builds a chronological timeline, while the Pathology and Genomics Analyst extracts structured findings such as tumor type, stage, driver mutations, resistance mutations, TMB, and MSI into pathology and genomic summaries.

这之后，Literature and Trial Researcher 会读取时间线、检测摘要和治疗史，检索公开指南、研究资料和临床试验。但与此同时，角色之间也会围绕“当前阶段做到什么程度才算可交付”展开讨论。比如有人认为缺少某次影像结果，有人认为真正关键的缺口是最新一次基因检测，有人指出没有最新检测结果就无法判断现有 trial list 是否仍然相关。经过讨论，项目内部会形成阶段标准，例如先完成 Case Timeline、Pathology Summary、Genomic Summary 和最近两线治疗史，再决定是输出 interim briefing 还是进入正式的 clinical trial search。 After that, the Literature and Trial Researcher reads the timeline, molecular summaries, and treatment history to search public guidance, studies, and trials. At the same time, roles may debate what level of completeness counts as deliverable for the current stage. One role may insist an imaging result is missing, another may argue the real gap is the latest genomic report, and another may point out that without that report the trial list may no longer be relevant. From this discussion, the project forms a stage standard.

如果试验检索中发现某个 eligibility 条件无法从现有材料判断，系统会自动触发回流：Researcher 把缺口回传给 Case Intake Coordinator，后者检查原始资料或生成 missing information item；如果仍无法确认，Clinical Briefing Coordinator 就把它加入 Questions for Oncologist。新报告上传后，也会触发完整更新链，从 Case Index、Timeline、Genomic Summary 到 literature summary、trial list 和 briefing 全部联动更新。必要时，项目状态甚至会从 “ready for supervisor review” 回退为 “research update required”。 If a trial search surfaces an eligibility condition that cannot be determined from the current materials, the system triggers a feedback loop: the researcher sends the gap back to the Case Intake Coordinator, who checks the source files or creates a missing-information item; if it still cannot be resolved, the Clinical Briefing Coordinator adds it to the Questions for Oncologist. When a new report is uploaded, a full update chain follows across the Case Index, timeline, genomic summary, literature summary, trial list, and briefing, and the project may even move backward from “ready for supervisor review” to “research update required.”

最终，和留学服务案例一样，“工作完成”并不是由单个角色宣布，而是由各角色汇总本轮标准、交付物和残留风险，形成 completion consensus，再提交给 Supervisor 做最终判断。 As in the study-abroad service example, “work completed” is not declared by one role alone. The roles combine their stage standards, deliverables, and remaining risks into a completion consensus and submit that to the Supervisor for a final decision.

Personal Mode 与 Organization Mode 的区别 The Difference Between Personal Mode and Organization Mode

这四个案例对应了 SmallClaw 的两种基本使用方式。在 Personal Mode 中，用户的参与程度更高。项目通常由一个用户持续主导，用户负责提出任务、决定下一步、检查结果并继续发出后续指令。SmallClaw 负责执行具体操作、整理材料、生成中间结果，并把这些结果保存在项目中供下一步继续使用。 These four examples map to SmallClaw’s two basic operating modes. In Personal Mode, the user participates more directly. One user usually leads the project, sets the next task, checks results, and keeps issuing follow-up instructions. SmallClaw handles the concrete execution, organizes materials, generates intermediate outputs, and keeps them inside the project for the next step.

在 Organization Mode 中，用户不需要亲自推动每一个中间步骤。任务进入项目后，系统可以在不同角色之间自动移交，角色之间可以围绕目标持续讨论、补充约束、修正理解，并逐步形成阶段标准和完成共识。用户更多是在项目开始时设定目标与边界，在关键节点查看结果、修正方向，或者以 Supervisor 的方式进行最终确认。 In Organization Mode, the user does not need to personally push every intermediate step. Once work enters the project, the system can hand it across roles automatically. Those roles can discuss goals, add constraints, correct understanding, and gradually form stage standards and completion consensus. The user mainly defines goals and boundaries at the beginning, checks results at key points, and confirms or redirects as a supervisor.

结尾 Closing

无论是哪一种模式，这四个案例的共同点都很明确。SmallClaw 不是一次性回答一个问题，而是在一个持续存在的项目环境中，围绕文件、Work Items、角色和结果，把工作一步一步往前推进。真正被保存下来的，不只是聊天记录，而是可复用的项目资产、阶段标准和执行结果。 Across both modes, the shared pattern is clear. SmallClaw is not designed to answer a question once. It advances work step by step inside a persistent project environment built around files, work items, roles, and results. What gets preserved is not only conversation, but reusable project assets, stage standards, and execution outputs.