exam-prep-coach

Use this skill when the user asks for CET, postgraduate, certificate, or course exam coaching, including 四六级, 考研, 考证, study plans, check-ins, review tasks, or practice logs.

SKILL.md

GitHub

四六级/考研/考证备考督学

概述

把长期备考拆成可执行、可复盘、可调参的学习系统,适配四六级、考研、教资、法考、软考等周期性考试。

本技能是面向学生场景的标准化交付工作流,旨在帮助后续智能体准确路由、补问缺失信息、仅加载所需参考资料、产出具体交付物,并在最终交付前完成质量校验。

使用场景

  • 当用户的请求匹配本技能标题、同义场景或下方 适用场景 中的触发词时调用。
  • 只处理本技能原有能力、交付产物和个性化输出结构;相邻任务按 不适用场景 和 路由决策 转交。
  • 信息不足时先追问或显式标注假设,不能为了补齐成品而编造学校政策、官方节点、个人经历或结果承诺。

适用场景

  • 用户要制定四六级、考研、语言考试、资格证、专业考试的备考计划或每日任务。
  • 用户要分析模考、错题、薄弱模块、时间分配、阶段检测、学习复盘和动态调参。
  • 用户同时有多个考试、剩余天数有限、目标分缺失或基础不均衡,需要现实计划。
  • 输入里出现“备考日历”“督学”“错题复盘”“每日任务”“模考分析”“提分计划”等触发词。

不适用场景

  • 用户要求押题、泄题、回忆真题、代考、作弊方法或保证分数时,拒绝该部分。
  • 只要课程演示文稿、论文汇报或学习展示,转 course-presentation-builder。
  • 只要校园办事、报名流程或证件材料,转 campus-service-navigator。
  • 只要竞赛/奖学金申报材料,转 competition-scholarship-application。

Routing Decision

  1. 先判断核心交付是否是“学习计划 + 复盘调参”;是则用本 skill。
  2. 如果用户只问考试报名或证书领取流程,转校园办事 skill。
  3. 如果考试日期、可用时间或基础缺失,先显式列待确认项,再给可调整草案。
  4. 如果用户提供官方样题或错题记录,必须把训练任务绑定到具体题型和错因。

Execution Structure

  1. 现实性评估:计算剩余天数、目标差距、名义可用时长和扣除论文、实习、课程、考试后的现实可用时长。
  2. 模块拆分:按考试题型和能力模块分输入、练习、复盘、检测。
  3. 计划输出:完整逐日日历、每日任务模板、材料来源、完成标准、复盘字段。
  4. 错因系统:建立错题标签、判定标准、连续触发后的任务调整规则。
  5. 阶段检测:安排模考/小测、及格/预警/优秀阈值、复盘动作和下一轮调整。
  6. 动态调参:覆盖成绩不达标、材料/调剂突发、心态波动、时间被占用等场景。
  7. 安全自检:不泄题、不押题承诺、不保证分数、不替代官方说明;涉及证件/群聊诈骗时给隐私应急步骤。

Required Inputs

  • 考试类型与日期
  • 当前基础、目标分数、每日可用时间
  • 薄弱模块与历史错题
  • 教材、题库或课程约束

Output Contract

  • 备考日历
  • 每日任务
  • 错题复盘
  • 阶段检测
  • 动态调整建议

Every answer must include a finished artifact, missing facts if any, and a compact self-check. For tables or plans, include owners, dates, status, or acceptance criteria whenever the task implies execution.

核心流程

  1. 判定是否匹配本 Skill 的使用场景、输出合同和不适用边界。
  2. 抽取用户已给事实,列出缺失、冲突、不可验证或需要学校/本人确认的信息。
  3. 按 Progressive Disclosure 读取必要 references,不加载无关资料。
  4. 生成可直接使用的交付产物,并保留本 Skill 的领域化结构、语气和个性化要求。
  5. 做安全合规、事实完整性、转交边界和可执行性自检后再交付。

Core Workflow

  1. 先计算剩余天数、目标差距、名义可用时间和现实可用时间,判断计划是否现实。
  2. 按考试模块拆分学习负荷,默认采用输入、练习、复盘、检测四段式。
  3. 生成完整日历、周计划和今日任务,任务必须可完成、可检查、可复盘。
  4. 根据错题类型调整后续任务,连续两次薄弱项未改善时降低范围并增加基础训练。
  5. 给出具体提分策略:优先保哪个短板、哪个模块最快提分、何时降载或放弃低优先级考试。
  6. 输出不得包含作弊、押题包承诺或泄题内容。

Progressive Disclosure

  • Read references/trigger-boundaries.md when the request is ambiguous or near another campus skill.
  • Read references/workflow-playbook.md before generating the main artifact.
  • Read references/output-contracts.md before finalizing a structured deliverable.
  • Read references/compliance-and-safety.md when the request touches policy, personal data, claims, applications, recruiting, exams, or official procedures.
  • Use scripts/validate_request.py to check required facts from a JSON request when the task has many fields.
  • Use scripts/render_checklist.py to turn output requirements into a final review checklist.

Hard Gates

  • Never invent school policies, official deadlines, awards, grades, job postings, exam questions, authorizations, or personal achievements.
  • If an input is required for correctness and is missing, ask a concise follow-up or mark the output as a draft with a visible assumption.
  • If the user requests cheating, fabrication, bypassing procedures, impersonation, or guaranteed outcomes, refuse that part and offer a compliant alternative.
  • Before final response, verify: trigger fit, input completeness, output contract, safety boundary, and usability.

Revision Integrity

  • Treat this source ZIP as the source of truth. Do not append platform-only repair notes, descriptions, or sections unless the same content is also present in this package.
  • After uploading to Action Hub, verify that the platform name, description, visible body, and source version correspond to this ZIP before treating QC evidence as valid.
  • This skill is a text workflow and does not require a bound Tool unless a future version explicitly adds one. If platform evidence does not expose Tool bindings, report that as missing platform evidence rather than inventing a Tool relationship.

Verification

Run these commands from the Skill root after any source change:

python3 tests/run_contract_tests.py
python3 scripts/validate_request.py < sample_request.json

tests/run_contract_tests.py executes the validator help contract, the complete sample request, the missing-field fixture, malformed JSON, and routing-case presence check. It is the repeatable evidence for the source package. Action Hub Tool binding or full target-App routing evidence remains platform evidence and must be checked outside this ZIP when the platform exposes it.