
Interview Assessment
- 2 installs
- 3 repo stars
- Updated May 19, 2026
- archlizheng/interview-assessment
interview-assessment is a skill that evaluates job-description, resume, and interview materials to score candidate fit, plan interviews, and produce recruiter or candidate reports.
About
This skill supports evidence-based interview assessment for two audiences: recruiters who need to evaluate candidates and plan interviews, and candidates preparing for or reviewing their own interviews. It scores work experience, professional capability, and soft skills against a job description and resume, then produces Markdown reports for fit assessment, interview preparation, and post-interview review. It refuses to score when the JD or resume is missing and returns an insufficient-evidence request instead.
- Evidence-based interview assessment for both recruiters and candidates
- Scores work experience, professional capability and soft skills 0-100 with grounded evidence quotes
- Produces Markdown reports (or JSON) for fit assessment, interview prep, and post-interview review
Interview Assessment by the numbers
- 2 all-time installs (skills.sh)
- Ranked #2,419 of 3,282 Productivity & Planning skills by installs in the Skillselion catalog
- Data as of Jul 28, 2026 (Skillselion catalog sync)
interview-assessment capabilities & compatibility
- Capabilities
- resume screening · interview prep · candidate scoring
- Use cases
- resume · planning
What interview-assessment says it does
Support evidence-based interview assessment for two audiences:
If `jdText` or `resumeText` is missing, do not score or make a conclusion.
npx skills add https://github.com/archlizheng/interview-assessment --skill interview-assessmentAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 2 |
|---|---|
| repo stars | ★ 3 |
| Last updated | May 19, 2026 |
| Repository | archlizheng/interview-assessment ↗ |
What it does
Evaluate a JD and resume, score fit, prepare interview questions, and review interview performance with grounded evidence.
Who is it for?
Scoring candidate-role fit and generating interview prep and post-interview review reports
Skip if: Scoring or concluding when jdText or resumeText is missing (it returns an insufficientEvidence section instead)
When should I use this skill?
when evaluating JD/resume/interview materials or helping a candidate assess role fit and prepare
What you get
Scored fit assessment, targeted interview questions, and a post-interview review grounded in quoted evidence
- candidate fit-assessment report
- interview preparation checklist
- post-interview evaluation report
By the numbers
- scores 3 dimensions (work experience, professional capability, soft skills) on a 0-100 scale
- weighted score computed via scoring-rubric.md
Files
<objective> Support evidence-based interview assessment for two audiences:
audience: recruiter: for HR, hiring managers, and interviewers who need candidate evaluation, interview planning, and post-interview recommendations.audience: candidate: for candidates who need role-fit self-assessment, interview preparation, resume or portfolio evidence improvement, and post-interview review.
Default to Markdown-first delivery. Use outputMode: json only when the user explicitly needs automation, ATS integration, app ingestion, or structured validation. Use outputMode: both when the user asks for both human-readable reports and a machine-readable evaluationBundle. </objective>
<intent_detection> Infer defaults when the user does not specify them:
- If the user asks to evaluate a candidate, screen resumes, prepare interviewer questions, or make a hiring recommendation, use
audience: recruiter. - If the user says "I", asks to prepare for an interview, improve resume match, rehearse answers, or review their own interview, use
audience: candidate. - Use
outputMode: markdownby default.
</intent_detection>
<required_inputs> Minimum required fields:
jdTextresumeText
Optional fields:
interviewTranscriptTextor interview experience notesinterviewerNotesmetadatasuch as candidate name, role, company, round, date, and target languageaudience: recruiter | candidateoutputMode: markdown | json | bothoutputLanguageormetadata.language, such aszh-CNoren
If jdText or resumeText is missing, do not score or make a conclusion. Return an insufficientEvidence section that asks only for the missing material. </required_inputs>
<language_and_filename_policy> Infer output language unless the user explicitly sets outputLanguage or metadata.language.
Language priority:
1. Explicit user instruction, outputLanguage, or metadata.language. 2. The language of the user's request. 3. If the request is mixed, use the dominant language of jdText and resumeText. 4. If still unclear, default to zh-CN.
Rules:
- Write all Markdown report content in the inferred output language.
- Localize section headings while preserving the required template structure and meaning.
- Localize filenames to the same output language.
- Preserve candidate names, company names, role names, product names, and quoted evidence in their original language unless translation is necessary for readability.
- If input materials are in one language but the user asks for another language, write analysis in the requested language and keep evidence quotes in the original language with a short translated explanation when helpful.
</language_and_filename_policy>
<workflow> <stage_1_fit_assessment> 1. Extract must-have skills, role context, seniority, domain expectations, and success signals from the JD. 2. Map resume evidence to the JD. 3. Score in 0-100:
- workExperience
- professionalCapability
- softSkills
4. Compute weighted score using scoring-rubric.md. 5. For each dimension, include:
- score
- 1-3 sentence rationale
- 1-2 grounded evidence quotes from the JD/resume/interview material
6. Apply audience-specific language:
- Recruiter: output
通过 | 待定 | 拒绝and explain hiring risk. - Candidate: output fit level and preparation priority; do not use hiring-decision wording.
</stage_1_fit_assessment>
<stage_2_preparation> 1. Identify uncertainty zones and missing evidence. 2. Produce focus areas with high | medium | low priority. 3. Generate targeted questions and answer strategy:
- behavioral questions
- technical/domain questions
- motivation and expectation questions
4. For recruiter output, include target competency and follow-up hints. 5. For candidate output, include answer angle, evidence to prepare, and weak spots to repair. </stage_2_preparation>
<stage_3_post_interview_review> Run this stage only when interviewTranscriptText, interview notes, or interview experience is provided.
1. Extract objective evidence snippets. 2. Evaluate professional capability, soft qualities, personal style, motivation, and expectation alignment. 3. Output pending concerns and next-step recommendations. 4. Use audience-specific framing:
- Recruiter: final hiring recommendation and process next step.
- Candidate: interview performance review, follow-up strategy, and improvement plan.
</stage_3_post_interview_review> </workflow>
<output_contract> <markdown_mode> Markdown mode is the default. Write Markdown files unless the user asks for chat-only or no files.
For audience: recruiter, use recruiter-report-templates.md:
{候选人姓名}-候选人初评报告.md{候选人姓名}-面试准备清单.md{候选人姓名}-面试后综合评价报告.mdonly when Stage 3 is produced
For audience: candidate, use candidate-report-templates.md:
{候选人姓名}-岗位匹配度自评报告.md{候选人姓名}-候选人面试准备清单.md{候选人姓名}-面试后复盘与跟进建议.mdonly when Stage 3 is produced
For English output, use localized filenames:
- Recruiter:
{candidateName}-candidate-pre-screening-report.md{candidateName}-interview-preparation-checklist.md{candidateName}-post-interview-evaluation-report.md- Candidate:
{candidateName}-role-fit-self-assessment.md{candidateName}-candidate-interview-preparation-checklist.md{candidateName}-post-interview-review-and-follow-up.md
For other languages, translate the filename suffix naturally and keep the candidate name unchanged.
Default directories:
- Recruiter:
{workspaceRoot}/候选人评估报告/ - Candidate:
{workspaceRoot}/候选人面试准备/
Candidate name:
- Use
metadata.candidateNameif present. - Otherwise infer from the resume header only when explicit.
- If still unknown, use
candidateIdand ask for the name in follow-up; do not guess.
</markdown_mode>
<json_mode> When the user explicitly requests outputMode: json or outputMode: both, return a complete evaluationBundle using references/evaluationBundle.schema.md as the optional automation contract.
Rules:
- In
jsonmode, do not write Markdown files unless asked. - In
bothmode, Markdown and JSON must contain the same scores, conclusions, risks, recommendations, and evidence. - JSON is an integration format, not the default HR/candidate deliverable.
</json_mode> </output_contract>
<audience_rules> <recruiter_mode>
- Keep language concise, evidence-based, and suitable for HR or interview panels.
- Use decision vocabulary:
通过 | 待定 | 拒绝. - Include risks, must-verify items, next-round suggestions, and evidence quotes.
- For high-impact decisions (
通过or拒绝), include at least 2 grounded evidence quotes. - Do not invent facts not found in the input.
</recruiter_mode>
<candidate_mode>
- Keep language practical, coaching-oriented, and non-deterministic.
- Do not use phrases such as "建议拒绝该候选人" or imply a guaranteed hiring outcome.
- Use fit and preparation vocabulary:
- 匹配度判断
- 优势证据
- 风险短板
- 回答策略
- 简历与作品集补强建议
- 面试后跟进建议
- Scores are preparation references, not predictions of hiring results.
</candidate_mode> </audience_rules>
<multi_agent_mode> Use only when the user asks for multi-agent or panel-style assessment.
Suggested role split:
- Role A: professional capability
- Role B: soft qualities and communication
- Role C: motivation, culture fit, and risk control
Aggregation:
1. Each role proposes scores, evidence, and concerns. 2. The orchestrator resolves conflicts by evidence quality. 3. Output the same audience-specific Markdown templates. 4. If JSON is requested, include agentVotes and roundtableSummary without removing base fields.
If a requested model is unavailable, use a single available model while preserving the role logic. </multi_agent_mode>
<responsible_use> Follow docs/responsible-use.md for fairness, evidence grounding, and sensitive decision handling. Never invent qualifications, protected-class signals, interview performance, or hiring conclusions beyond the provided evidence. </responsible_use>
Candidate Report Templates
Use these templates for audience: candidate.
Rules:
- Do not use recruiter-only decision wording such as
通过,拒绝,建议拒绝该候选人, or guaranteed hiring predictions. - Scores are preparation references, not hiring outcome predictions.
- Each weakness must include an actionable repair step.
- Each answer strategy should point to evidence the candidate can prepare.
- Use
high | medium | lowto prioritize preparation work. - Localize headings and labels to the inferred output language while preserving this section order and meaning.
1) 岗位匹配度自评报告
# 岗位匹配度自评报告
## 基础信息
- 候选人姓名:
- 目标岗位:
- 目标公司:
- 评估时间:
## 总体匹配度
- 匹配度分数(0-100):
- 匹配度判断: 高匹配 | 中等匹配 | 需要补强
- 主要原因:
## 分项匹配分析
### 工作经验(0-100)
- 分数:
- 优势证据:
-
- 风险短板:
-
- 补强建议:
-
### 专业能力(0-100)
- 分数:
- 优势证据:
-
- 风险短板:
-
- 补强建议:
-
### 软性素质(0-100)
- 分数:
- 优势证据:
-
- 风险短板:
-
- 补强建议:
-
## 简历与作品集补强建议
-
## 准备优先级
- high | medium | low:
## 置信度
- high | medium | low2) 候选人面试准备清单
# 候选人面试准备清单
## 高优先级准备项
1. [准备项] 优先级: high | medium | low
- 为什么重要:
- 需要准备的证据:
- 建议表达角度:
## 可能被追问的问题
1. 问题:
- 面试官想验证:
- 回答策略:
- 可引用经历:
## 行为面试回答素材
- 场景:
- 任务:
- 行动:
- 结果:
- 可量化成果:
## 技术/业务问题准备
- 主题:
- 核心观点:
- 案例证据:
- 易被追问点:
## 反问面试官的问题
- 3) 面试后复盘与跟进建议
Generate only when interview transcript, notes, or interview experience is provided.
# 面试后复盘与跟进建议
## 面试表现总览
- 表现评分(0-100):
- 表现判断: 表达充分 | 部分充分 | 需要明显补强
- 主要依据:
## 表现较好的部分
-
## 可能造成疑虑的部分
- 疑虑点:
- 可能原因:
- 后续补救方式:
## 跟进邮件/消息建议
- 建议发送时间:
- 建议表达重点:
- 可补充材料:
## 下一轮准备重点
1. [准备项] 优先级: high | medium | low
- 需要补强:
- 具体行动:
## 长期改进建议
- Responsible Use
Human review
Recruiter-facing outputs are decision support, not final hiring decisions. A human recruiter or hiring manager should review evidence, context, and job-specific constraints before acting.
Evidence limits
Only use information provided in the JD, resume, transcript, interview notes, or user-supplied context. If evidence is missing, state the uncertainty instead of filling gaps.
Fairness
Do not base conclusions on protected or sensitive attributes such as age, gender, race, ethnicity, religion, disability, marital status, pregnancy, nationality, or other legally protected categories.
Candidate guidance
Candidate-facing reports provide preparation advice and fit analysis. They must not promise interview success, imply a guaranteed offer, or present preparation scores as hiring predictions.
Privacy
Minimize personal data. Prefer anonymized candidate IDs when sharing examples. Do not include private contact details, identity documents, compensation records, or unrelated personal information unless the user explicitly provides them for a legitimate assessment purpose.
使用指南:用脱敏案例跑通 interview-assessment
这份指南面向第一次使用 interview-assessment 的用户。读完后,你应该能用一份 JD、一份简历和一段面试记录,生成招聘方或候选人方的 Markdown 报告。
本文示例改写自一组真实材料,但已做脱敏处理:姓名、公司、年份、薪资、地点和项目细节均经过模糊化。请不要把示例中的人物或公司视为真实对象。
1. 准备输入材料
最小输入只需要两类材料:
jdText:岗位 JDresumeText:候选人简历
如果要生成面试后报告,再补充:
interviewTranscriptText:面试转写、面试官笔记或候选人自述复盘
脱敏示例:
metadata:
candidateName: 吴XX
appliedRole: AI Agent 开发工程师
company: 某智能硬件公司
language: zh-CN
jdText: |
岗位定位:负责 AI Agent 系统从方案到上线的工程落地,重点搭建多模态 AIGC 生产管线,
支持文本、图像、视频内容生成,并推动能力在智能硬件场景中稳定交付。
必备条件:
- 3-5 年 AI 应用或相关后端工程经验
- 有完整多模态 AIGC 生产管线实践
- 熟悉模型/API 调用、异步任务、内容后处理、存储分发、质量回归
- 熟练掌握 Python、JavaScript、Rust、Go 中至少一门
- 熟悉主流 Agent 应用框架并可完成服务化落地
resumeText: |
候选人吴XX,20XX 年毕业于某 211 高校机器人工程相关专业。
20XX 年至 20XX 年在某影像硬件公司担任算法工程师,参与 SLAM、双目视觉、
AIGC 内容生成、视频翻译、剧本生成、漫剧生成等方向。
代表项目:
- 主导漫剧生成系统,覆盖剧情拆分、分镜规划、画面生成、质检修复和结果输出。
- 参与视频翻译链路,覆盖字幕识别、文本翻译、TTS、字幕抹除、音画同步和视频合成。
- 在剧本生成模块中引入 RAG,用向量库和模板召回提升内容风格匹配度。2. 选择用户视角
interview-assessment 支持两种 audience。
招聘方使用:
audience: recruiter
outputMode: markdown适合生成:
- 候选人初评报告
- 面试准备清单
- 面试后综合评价报告
候选人方使用:
audience: candidate
outputMode: markdown适合生成:
- 岗位匹配度自评报告
- 候选人面试准备清单
- 面试后复盘与跟进建议
如果没有显式写 audience,Skill 会根据你的请求自动判断。比如“帮我评估这个候选人”会倾向招聘方;“帮我准备这个岗位面试”会倾向候选人方。
3. 生成招聘方报告
招聘方最小输入:
audience: recruiter
outputMode: markdown
metadata:
candidateName: 吴XX
appliedRole: AI Agent 开发工程师
jdText: |
这里放脱敏后的 JD。
resumeText: |
这里放脱敏后的简历。默认输出:
候选人评估报告/吴XX-候选人初评报告.md候选人评估报告/吴XX-面试准备清单.md
如果再加入面试转写:
interviewTranscriptText: |
面试中候选人介绍了一个多阶段 AIGC 内容生成系统:
输入可以从 idea 或梗概开始,经过剧本生成、人设/场景生成、分镜规划、视频生成和视频合成。
候选人提到系统会用 JSON Schema 保存结构化内容,并区分临时存储和全局存储。
候选人说明 RAG 模块使用向量库和标签化模板召回,并按高分、低分和中间分三类情况处理。
当召回结果低于阈值时,会临时生成风格模板,并在用户接受后动态入库。
候选人也提到自己主要使用 LangChain 做 RAG 或数据库调用,但对 LangGraph、AutoGen 等多智能体编排框架的回答不够充分。会额外输出:
候选人评估报告/吴XX-面试后综合评价报告.md
招聘方报告会使用 通过 | 待定 | 拒绝 等决策词,并给出风险提示、证据摘录和下一轮建议。
4. 生成候选人方报告
候选人方最小输入:
audience: candidate
outputMode: markdown
metadata:
candidateName: 吴XX
appliedRole: AI Agent 开发工程师
jdText: |
这里放目标岗位 JD。
resumeText: |
这里放自己的简历。默认输出:
候选人面试准备/吴XX-岗位匹配度自评报告.md候选人面试准备/吴XX-候选人面试准备清单.md
如果加入面试经历或转写:
interviewTranscriptText: |
面试中重点聊到了漫剧生成系统、RAG 召回策略、视频翻译链路、Agent 编排框架、
智能硬件场景理解、薪资期望和加班制度。会额外输出:
候选人面试准备/吴XX-面试后复盘与跟进建议.md
候选人方报告不会给出“录用/拒绝”式结论。它会聚焦匹配度、优势证据、风险短板、回答策略和后续准备。
完整候选人视角样例见 candidate-output-sample.zh-CN.md。
5. 什么时候使用 JSON
默认使用 Markdown 即可。
只有在以下场景才建议使用 JSON:
- 需要接入 ATS 或招聘系统
- 需要批量评估多个候选人
- 需要网页应用展示结构化字段
- 需要自动校验分数、证据和结论是否一致
示例:
outputMode: both此时 Skill 会同时输出 Markdown 和 evaluationBundle。Markdown 与 JSON 中的分数、证据、风险和建议必须一致。
6. 脱敏建议
公开示例、演示稿、GitHub 文档中建议这样处理:
- 真实姓名:
某真实姓名→吴XX - 公司名称:真实公司 →
某影像硬件公司、某智能硬件公司 - 学校名称:可保留层级,不保留全名,如
某 211 高校 - 年份日期:
真实起止日期→20XX 年至 20XX 年 - 薪资:具体数字 →
低于岗位预算下限、有议价空间 - 地点:具体园区 →
某一线城市科技园区 - 会议转写:删除主持人姓名、会议 ID、精确时间戳和无关闲聊
保留的信息应服务于评估本身,例如项目类型、技术链路、责任边界、证据摘录、风险点和下一步建议。
7. 使用效果
使用这套输入后,招聘方可以获得证据化评估,候选人可以获得准备路径。两种视角使用同一套材料,但输出语言和结论方式不同:
- 招聘方关心:是否推进、风险在哪里、下一轮问什么。
- 候选人关心:自己哪里匹配、哪里短板、下一轮怎么准备。
这就是 interview-assessment 的核心价值:把面试材料变成可行动、可复盘、可交付的 Markdown 文档。
Candidate Input Example
audience: candidate
outputMode: markdown
metadata:
candidateName: 李四
appliedRole: 数据分析师
company: 示例零售
jdText: |
岗位要求:
- 熟练使用 SQL 和 Python
- 能搭建经营分析看板
- 有零售、电商或用户增长分析经验优先
resumeText: |
李四,2 年数据分析经验。
熟悉 SQL、Python、Tableau,曾搭建销售日报和用户留存分析看板。
参与过电商促销复盘,支持运营团队优化投放策略。Expected default output:
候选人面试准备/李四-岗位匹配度自评报告.md候选人面试准备/李四-候选人面试准备清单.md
If interview notes or interview experience are added, also generate:
候选人面试准备/李四-面试后复盘与跟进建议.md
候选人视角输出样例(脱敏)
本样例基于脱敏后的 JD、简历和面试转写生成,适用于 audience: candidate。示例中的姓名、公司、年份、薪资和项目细节均已模糊化。
岗位匹配度自评报告
基础信息
- 候选人姓名: 吴XX
- 目标岗位: AI Agent 开发工程师
- 目标公司: 某智能硬件公司
- 评估时间: 20XX-XX-XX
总体匹配度
- 匹配度分数(0-100): 80
- 匹配度判断: 高匹配
- 主要原因: 你的多模态 AIGC 内容生成、剧本/漫剧生成、视频翻译和 RAG 模块经验,与岗位要求中的“多模态 AIGC 生产管线”“Agent 编排与服务化”“场景化能力落地”高度相关。主要短板是主流 Agent 编排框架、服务化工程、成本控制和智能硬件场景经验需要更明确地展示。
分项匹配分析
工作经验(0-100)
- 分数: 78
- 优势证据:
- 你在 20XX 年至 20XX 年持续参与 AI 应用和 AIGC 内容生成方向,覆盖剧本生成、漫剧生成、视频翻译、图像/视频生成等多个项目。
- 你在面试中说明自己曾从 idea 或梗概开始,设计剧本、人设、分镜、视频生成和视频合成的多阶段链路。
- 风险短板:
- 过往经验主要集中在单一组织和内容生成平台,跨公司、跨业务系统迁移能力需要更主动解释。
- 对目标岗位中的智能硬件生态、设备控制、环境感知和现实世界执行风险,需要准备更具体的迁移逻辑。
- 补强建议:
- 准备一段“从内容生成平台迁移到智能硬件 Agent”的说明,重点讲任务规划、状态管理、失败重试、结果校验和用户介入点。
专业能力(0-100)
- 分数: 83
- 优势证据:
- 你能清楚讲出漫剧生成系统的阶段拆分:剧本生成、人设/场景生成、分镜规划、视频生成、视频合成,并提到 JSON Schema、临时存储、全局存储、QA check 和用户介入。
- 你能描述 RAG 召回策略:标签化模板、向量库检索、高分直选、低分临时生成、中间分候选选择,以及用户接受后的动态入库。
- 你能讲出视频翻译链路:字幕识别、文本翻译、TTS、字幕抹除、情感识别、音画同步和视频合成。
- 风险短板:
- 面试中对 LangGraph、AutoGen 等多智能体编排框架的回答不够聚焦,容易让面试官担心你只是手写流程,而不是掌握成熟编排范式。
- 服务化、监控、CI/CD、并发、成本控制和稳定性指标没有展开,和岗位“生产级工程意识”还有表达缺口。
- 补强建议:
- 准备一张文字版编排图,说明每个 Agent 或模块的输入、输出、状态、失败重试和回滚逻辑。
- 准备一个成本控制案例:例如模型选择、缓存、异步任务、失败重试、人工审核前置、质量回归如何降低无效调用。
软性素质(0-100)
- 分数: 80
- 优势证据:
- 面试环境有干扰时,你能主动请求复述并继续结构化表达,说明沟通稳定性较好。
- 你会主动暴露边界,例如说明部分方向只是探索、部分功能还在开发中、硬件参数需要学习。
- 风险短板:
- 你对薪资、福利、加班制度的提问比较直接,如果时机和表达不够平衡,可能被部分面试官解读为关注点偏待遇。
- 对未知概念能类比思考,但需要更快地回到自己已做项目中的具体实践,避免显得泛化。
- 补强建议:
- 准备一段结尾提问顺序:先问业务目标和团队挑战,再问协作方式,最后问薪资福利等流程问题。
简历与作品集补强建议
- 在简历中补充主语言、框架和服务化能力,例如 Python/JavaScript、LangChain、向量库、API 服务、异步任务、监控和质量回归。
- 将“漫剧生成系统”写成一条端到端案例,按输入、处理链路、状态管理、用户介入、质量检查、输出和指标展开。
- 为 RAG 模块补充可量化信息,例如数据规模、召回阈值、命中策略、人工选择机制、动态入库效果。
- 为视频翻译模块补充你个人负责边界,避免“参与项目”显得责任不清。
- 准备一页作品集式项目说明,展示系统架构图、核心流程和关键 trade-off。
准备优先级
- high: Agent 编排框架与状态管理说明。它最可能成为技术复试的关键追问。
- high: 服务化与生产稳定性。需要补充部署、监控、重试、降级、成本控制和质量回归。
- medium: 智能硬件场景迁移。需要把内容生成经验转译成设备联动、环境感知和结果校验语言。
- low: 学历和行业背景。不是主要短板,但可以用项目深度和业务贴合度抵消。
置信度
- high
候选人面试准备清单
高优先级准备项
1. Agent 编排与状态管理 优先级: high
- 为什么重要: 目标岗位强调 Agent 编排、工具调用、状态管理和容错机制。
- 需要准备的证据: 漫剧生成系统中剧本、人设、分镜、视频生成、合成之间的状态流转,以及失败重试和用户介入点。
- 建议表达角度: 不只说“用了 API”,而要说清楚“每一步的输入输出、状态保存、失败处理和质量检查”。
2. 生产级工程能力 优先级: high
- 为什么重要: JD 明确要求稳定性、可观测性、成本和效率平衡。
- 需要准备的证据: 异步任务、质量回归、badcase 回流、模型选择、缓存、降级、人工审核或自动 QA 的实践。
- 建议表达角度: 用一个项目讲完整闭环:问题、方案、指标、上线后效果、仍需改进的点。
3. 智能硬件场景迁移 优先级: medium
- 为什么重要: 目标公司不是单纯内容平台,而是智能硬件和多设备生态。
- 需要准备的证据: SLAM、双目视觉、3D 灯光估计、视频重打光或硬件相关协作经历。
- 建议表达角度: 强调自己能理解“软件决策影响现实世界执行”的安全边界和校验需求。
可能被追问的问题
1. 你们的多智能体编排到底用了什么框架?如果不用 LangGraph,状态是怎么管理的?
- 面试官想验证: 你是否真正理解 Agent 编排,而不是只串 API。
- 回答策略: 先承认实际项目中使用 LangChain 的范围,再用状态机/数据库/任务队列/QA check 解释编排设计。
- 可引用经历: 漫剧生成系统中全局存储、临时存储、用户介入和阶段性回滚。
2. 如果一个视频生成任务失败,你们如何重试、降级或回滚?
- 面试官想验证: 生产级容错意识。
- 回答策略: 按失败类型拆分:模型失败、质量不达标、用户不满意、资源超限、上下文过长。
- 可引用经历: 分镜生成、改图、生图、字幕处理和视频合成链路。
3. RAG 召回不准时怎么办?
- 面试官想验证: 检索质量、阈值策略和人工介入设计。
- 回答策略: 讲高分、低分、中间分三类处理;低分时临时生成模板,用户接受后动态入库。
- 可引用经历: 模板标签化、Chroma 向量库、LangChain 检索链。
4. 你如何把 AIGC 内容生成经验迁移到智能硬件 Agent?
- 面试官想验证: 场景理解和抽象迁移能力。
- 回答策略: 用“计划、执行、观察、校验、反馈”结构回答。
- 可引用经历: 内容生成链路中的计划、用户确认、质量检查和输出合成。
行为面试回答素材
- 场景: 某 AIGC 内容生成平台需要从简单 idea 生成多集内容,并支持用户介入修改。
- 任务: 设计从剧本生成到视频合成的端到端链路,确保结构化保存、质量检查和阶段性回滚。
- 行动: 使用 JSON Schema 保存剧本、人设、场景和分镜信息;区分临时存储和全局存储;在关键节点加入用户确认和 QA check;针对 RAG 召回设置阈值与动态入库策略。
- 结果: 系统能够支持多阶段内容生成,并让用户在关键节点控制风格、角色、分镜和输出质量。
- 可量化成果: 若真实数字可公开,可补充生成效率、准确率、任务吞吐或人工介入减少比例;公开场景中建议写成区间或脱敏指标。
技术/业务问题准备
- 主题: 多模态 AIGC 管线
- 核心观点: AIGC 生产系统不是单次模型调用,而是意图解析、结构化状态、模型调用、质量检查、用户反馈和结果合成的工程系统。
- 案例证据: 漫剧生成、视频翻译、剧本生成。
- 易被追问点: 成本、并发、失败重试、质量回归。
- 主题: Agent 编排
- 核心观点: 编排的关键是状态、工具边界、失败处理和观察反馈。
- 案例证据: 阶段化生成链路、临时/全局存储、用户介入点。
- 易被追问点: LangGraph/AutoGen 是否使用、为什么不用、如何替代。
- 主题: 智能硬件 Agent
- 核心观点: 设备 Agent 要关注现实世界执行风险,不能只关注文本生成效果。
- 案例证据: 视觉、灯光、视频和硬件相关项目经验。
- 易被追问点: 设备状态观察、动作确认、安全边界、异常处理。
反问面试官的问题
- 团队当前最急的目标是 AIGC 内容生成管线,还是多设备 Agent 编排?
- 当前系统已经有哪些可复用模块:任务队列、设备 API、模型网关、内容审核、日志和指标?
- 复试中是否希望我准备一个具体项目的架构图或编排图?
- 这个岗位前三个月最希望解决的一个工程问题是什么?
面试后复盘与跟进建议
面试表现总览
- 表现评分(0-100): 82
- 表现判断: 表达充分
- 主要依据: 你对漫剧生成、RAG 召回和视频翻译链路讲得比较完整,能说明数据结构、阈值策略、用户介入和工程边界。主要不足是 Agent 编排框架、服务化工程和成本控制没有展开到复试所需深度。
表现较好的部分
- 能从输入、状态、处理链路、质量检查和输出完整描述一个复杂 AIGC 系统。
- 对 RAG 召回策略有分情况处理,不是只停留在“用了向量库”。
- 能坦诚区分已落地、探索中和不了解的部分,可信度较高。
- 对智能硬件场景表现出兴趣,并能追问硬件参数是否影响开发。
可能造成疑虑的部分
- 疑虑点: 多智能体框架回答不够清晰。
- 可能原因: 过往项目更多是工程链路设计,而不是围绕 LangGraph/AutoGen 这类框架展开。
- 后续补救方式: 跟进材料中补一张“项目编排图”,说明你如何实现状态管理、工具调用、失败重试和人工介入。
- 疑虑点: 服务化和稳定性表达不足。
- 可能原因: 面试重点集中在算法和内容生成链路,未主动展开后端工程细节。
- 后续补救方式: 补充一个“生产化能力清单”,覆盖 API 服务、异步任务、监控、成本控制、质量回归和 badcase 回流。
- 疑虑点: 智能硬件经验需要迁移解释。
- 可能原因: 过往主要是内容生成和影像相关项目,目标岗位含多设备联动。
- 后续补救方式: 用“观察-计划-执行-校验-反馈”的 Agent 闭环解释你如何进入硬件场景。
跟进邮件/消息建议
- 建议发送时间: 面试后 24 小时内。
- 建议表达重点: 感谢面试机会,补充自己对岗位中 AIGC 管线和多设备 Agent 的理解,并主动附上一页项目编排说明。
- 可补充材料:
- 漫剧生成系统脱敏架构图
- RAG 召回策略说明
- 视频翻译链路职责边界
- 复试可讲的 3 个技术 trade-off
示例跟进文案:
您好,感谢今天的沟通。面试后我重新梳理了自己过往 AIGC 内容生成链路与贵司岗位的匹配点。
我理解这个岗位的短期重点是把用户输入转化为稳定可交付的多模态内容生成管线;中长期会延伸到多设备 Agent 的计划、执行、观察和校验闭环。
我过往在漫剧生成、剧本生成和视频翻译中做过类似的阶段化链路设计,包括结构化状态保存、用户介入点、质量检查、RAG 召回和动态入库。下一轮我可以重点展开 Agent 编排、失败重试、服务化和成本控制部分。
如果方便,我也可以补充一页脱敏项目架构图,便于技术复试前对齐讨论重点。下一轮准备重点
1. 编排图与代码级说明 优先级: high
- 需要补强: LangChain/LangGraph/AutoGen 相关追问。
- 具体行动: 准备一个最小示例,说明状态节点、工具调用、条件分支、失败重试和人工介入。
2. 生产化能力 优先级: high
- 需要补强: 服务部署、并发、监控、成本控制。
- 具体行动: 用表格列出每个项目中的“线上问题-工程方案-效果指标”。
3. 硬件 Agent 场景理解 优先级: medium
- 需要补强: 多设备联动和现实世界执行安全。
- 具体行动: 准备一个泳池清洁或智能音箱场景,按目标拆解、设备 API、状态观察、异常处理和用户通知讲一遍。
长期改进建议
- 系统学习主流 Agent 编排框架,至少准备一个可讲清楚的 LangGraph 或类似状态图案例。
- 将项目经历从“算法/模型调用”升级为“生产系统设计”表达方式。
- 给每个核心项目补充可量化指标,即使公开版本使用区间或脱敏数字。
- 把硬件相关经历整理成“软硬件协同”叙事,突出自己能理解传感器、设备状态、执行风险和用户体验之间的关系。
Recruiter Input Example
audience: recruiter
outputMode: markdown
metadata:
candidateName: 张三
appliedRole: AI 产品经理
company: 示例科技
jdText: |
岗位要求:
- 3 年以上 AI 产品经验
- 熟悉 B 端产品设计、用户研究和数据分析
- 能与算法、工程、销售团队协作推进落地
resumeText: |
张三,AI 产品经理,4 年 B 端 SaaS 产品经验。
曾负责智能客服产品从 0 到 1,上线后提升客服处理效率 28%。
熟悉需求调研、PRD、数据分析和跨团队项目管理。Expected default output:
候选人评估报告/张三-候选人初评报告.md候选人评估报告/张三-面试准备清单.md
If interview transcript or interviewer notes are added, also generate:
候选人评估报告/张三-面试后综合评价报告.md
interview-assessment
Dual-audience interview assessment and preparation skill for recruiters and candidates.
中文 README
Installation
Recommended:
npx skills add archlizheng/interview-assessmentThen restart Codex to pick up the new skill.
Manual Git install:
git clone https://github.com/archlizheng/interview-assessment.git "${CODEX_HOME:-$HOME/.codex}/skills/interview-assessment"Update an existing manual install:
git -C "${CODEX_HOME:-$HOME/.codex}/skills/interview-assessment" pullUninstall:
rm -rf "${CODEX_HOME:-$HOME/.codex}/skills/interview-assessment"Quick start
Recruiter mode:
Use interview-assessment to evaluate this candidate for recruiter audience.
jdText:
<paste the job description>
resumeText:
<paste the resume>
metadata:
- candidateName: <name>
- appliedRole: <role>
- company: <company>
- language: enCandidate mode:
Use interview-assessment to help me prepare for this role.
jdText:
<paste the job description>
resumeText:
<paste my resume>
metadata:
- candidateName: <name>
- appliedRole: <role>
- language: enWhat it does
This skill turns JD, resume, and optional interview transcript material into Markdown reports.
It supports two audiences:
recruiter: candidate screening, interviewer preparation, and post-interview evaluation.candidate: role-fit self-assessment, interview preparation, answer strategy, and post-interview review.
Default output is Markdown. JSON is optional for automation and app integration.
Inputs
Required:
jdTextresumeText
Optional:
interviewTranscriptTextor interview notesmetadata.candidateNamemetadata.appliedRolemetadata.companyaudience: recruiter | candidateoutputMode: markdown | json | bothoutputLanguageormetadata.language
Defaults:
audienceis inferred from the request.outputModedefaults tomarkdown.- Output language is inferred from the user's request and input materials unless explicitly set.
Recruiter outputs
With JD + resume:
候选人初评报告.md面试准备清单.md
With interview transcript or notes:
面试后综合评价报告.md
Recruiter reports may use hiring recommendation language: 通过 | 待定 | 拒绝.
English filenames:
{candidateName}-candidate-pre-screening-report.md{candidateName}-interview-preparation-checklist.md{candidateName}-post-interview-evaluation-report.md
Candidate outputs
With JD + resume:
岗位匹配度自评报告.md候选人面试准备清单.md
With interview transcript, notes, or interview experience:
面试后复盘与跟进建议.md
Candidate reports must not imply a guaranteed hiring outcome or use recruiter rejection language.
English filenames:
{candidateName}-role-fit-self-assessment.md{candidateName}-candidate-interview-preparation-checklist.md{candidateName}-post-interview-review-and-follow-up.md
Language behavior
The skill localizes both report content and filenames.
Priority:
1. Explicit user instruction, outputLanguage, or metadata.language. 2. The language of the user's request. 3. The dominant language of the JD and resume. 4. zh-CN if still unclear.
Names, companies, role titles, and evidence quotes should remain in their original language unless translation is requested.
Optional JSON mode
Use outputMode: json or outputMode: both only when you need structured automation output.
The optional contract is documented in references/evaluationBundle.schema.md.
Examples
- examples/recruiter-input.md
- examples/candidate-input.md
- docs/usage-guide.zh-CN.md
- examples/candidate-output-sample.zh-CN.md
Responsible use
Hiring decisions should remain human-reviewed and evidence-based. Candidate-facing guidance is preparation support, not a guarantee of interview results. See docs/responsible-use.md.
interview-assessment
面向招聘方和候选人方的双视角面试评估与准备 Skill。
English README
安装方式
推荐方式:
npx skills add archlizheng/interview-assessment安装完成后,重启 Codex 以加载新的 skill。
手动 Git 安装:
git clone https://github.com/archlizheng/interview-assessment.git "${CODEX_HOME:-$HOME/.codex}/skills/interview-assessment"更新手动安装版本:
git -C "${CODEX_HOME:-$HOME/.codex}/skills/interview-assessment" pull卸载:
rm -rf "${CODEX_HOME:-$HOME/.codex}/skills/interview-assessment"快速开始
招聘方模式:
使用 interview-assessment,以 recruiter audience 评估这个候选人。
jdText:
<粘贴岗位 JD>
resumeText:
<粘贴候选人简历>
metadata:
- candidateName: <候选人姓名>
- appliedRole: <岗位名称>
- company: <公司名称>
- language: zh-CN候选人模式:
使用 interview-assessment,帮我准备这个岗位的面试。
jdText:
<粘贴岗位 JD>
resumeText:
<粘贴我的简历>
metadata:
- candidateName: <姓名>
- appliedRole: <岗位名称>
- language: zh-CN它能做什么
这个 Skill 可以根据 JD、简历,以及可选的面试转写或面试记录,生成可直接使用的 Markdown 文档。
它支持两类用户:
recruiter:适用于 HR、招聘负责人、面试官,用于候选人初筛、面试准备和面试后评估。candidate:适用于候选人本人,用于岗位匹配度自评、面试准备、回答策略设计和面试后复盘。
默认输出 Markdown。JSON 只在自动化、ATS 集成、网页系统接入或批量评估场景中作为可选输出。
输入
必填:
jdTextresumeText
可选:
interviewTranscriptText或面试记录metadata.candidateNamemetadata.appliedRolemetadata.companyaudience: recruiter | candidateoutputMode: markdown | json | bothoutputLanguage或metadata.language
默认规则:
audience会根据用户请求自动判断。outputMode默认为markdown。- 输出语言会根据用户请求和输入材料自动判断,也可以显式指定。
招聘方输出
当输入 JD + 简历时,默认生成:
候选人初评报告.md面试准备清单.md
当额外提供面试转写、面试官笔记或面试记录时,还会生成:
面试后综合评价报告.md
招聘方报告可以使用招聘决策措辞:
通过待定拒绝
英文输出时,文件名会本地化为:
{candidateName}-candidate-pre-screening-report.md{candidateName}-interview-preparation-checklist.md{candidateName}-post-interview-evaluation-report.md
候选人方输出
当输入 JD + 简历时,默认生成:
岗位匹配度自评报告.md候选人面试准备清单.md
当额外提供面试转写、面试经历或复盘记录时,还会生成:
面试后复盘与跟进建议.md
候选人方报告不能使用招聘方拒绝或录用式措辞,也不能承诺面试结果。它的重点是:
- 匹配度判断
- 优势证据
- 风险短板
- 回答策略
- 简历与作品集补强建议
- 面试后跟进建议
英文输出时,文件名会本地化为:
{candidateName}-role-fit-self-assessment.md{candidateName}-candidate-interview-preparation-checklist.md{candidateName}-post-interview-review-and-follow-up.md
语言行为
这个 Skill 会同时本地化报告内容和文件名。
语言判断优先级:
1. 用户显式要求、outputLanguage 或 metadata.language 2. 用户请求本身使用的语言 3. JD 和简历的主要语言 4. 仍不明确时默认使用 zh-CN
候选人姓名、公司名、岗位名、产品名和证据摘录默认保留原文。若用户要求翻译,或为了理解需要说明,可以在保留原文证据的同时补充简短译文或解释。
可选 JSON 模式
仅在需要结构化自动化输出时使用:
outputMode: jsonoutputMode: both
JSON 契约见 references/evaluationBundle.schema.md。
默认情况下,不需要输出 JSON,也不需要用户理解 evaluationBundle。
示例
- examples/recruiter-input.md
- examples/candidate-input.md
- docs/usage-guide.zh-CN.md
- examples/candidate-output-sample.zh-CN.md
负责任使用
招聘方输出是决策辅助,不应替代人工判断。候选人方输出是准备建议,不构成录用承诺或结果预测。
更多说明见 docs/responsible-use.md。
Recruiter Report Templates
Use these templates for audience: recruiter.
Rules:
- Use hiring-decision vocabulary only in recruiter mode:
通过 | 待定 | 拒绝. - Every score must be 0-100 and match the score used in the final recommendation.
- Each scoring dimension must include score, rationale, and 1-2 grounded evidence quotes.
- High-impact decisions (
通过or拒绝) require at least 2 evidence quotes. - Do not add claims that are not supported by JD, resume, transcript, or interviewer notes.
- Localize headings and labels to the inferred output language while preserving this section order and meaning.
1) 候选人初评报告
# 候选人初评报告
## 基础信息
- 候选人ID:
- 候选人姓名:
- 应聘岗位:
- 评估时间:
## 综合岗位胜任力评分
- 总分(0-100):
### 工作经验(0-100)
- 分数:
- 评分说明:
- 关键证据:
-
-
### 专业能力(0-100)
- 分数:
- 评分说明:
- 关键证据:
-
-
### 软性素质(0-100)
- 分数:
- 评分说明:
- 关键证据:
-
-
## 初步结论
- 建议结果: 通过 | 待定 | 拒绝
- 结论原因:
## 风险提示
-
## 必验项
- [ ]
## 置信度
- high | medium | low2) 面试准备清单
# 面试准备清单
## 重点考察方向
1. [方向] 优先级: high | medium | low
- 为什么要考察:
- 建议观察点:
## 建议问题清单
1. 问题:
- 对应能力项:
- 追问建议:
## 必验项(必须确认)
- [ ] 技术/业务关键能力
- [ ] 过往项目真实性与角色贡献
- [ ] 动机与期望匹配度
## 风险追问项
- 风险主题:
- 追问问题:
- 判定标准:3) 面试后综合评价报告
Generate only when interview transcript, interviewer notes, or interview experience is provided.
# 面试后综合评价报告
## 综合结论
- 最终建议: 通过 | 待定 | 拒绝
- 综合分:
- 主要依据:
## 专业能力
- 与岗位匹配程度:
- 过往行业经验:
- 技能维度:
- 结论摘要:
## 软性素质
- 沟通表达能力:
- 逻辑思维能力:
- 抗压力:
- 学习力:
- 结论摘要:
## 个人风格
- 特征标签:
- 证据说明:
## 应聘动机与期望
- 应聘动机:
- 对岗位/薪酬/发展预期:
- 匹配度判断:
## 待定顾虑点
-
## 后续流程建议
- 建议动作: 进入下一轮 | 补充面试 | 拒绝并归档
- 下一步负责人:
- 建议完成时限:
## 证据摘录
-
## 置信度
- high | medium | lowOptional Evaluation Bundle Schema
This file is an optional JSON contract for automation, ATS integration, app ingestion, batch evaluation, or validation workflows.
The GitHub skill defaults to Markdown output. Use this schema only when the user explicitly requests outputMode: json or outputMode: both.
1. Purpose
This document defines a unified schema for single-agent and multi-agent interview evaluation workflows that need machine-readable output.
2. Input: candidatePack
{
"version": "1.0",
"candidateId": "C-2026-0001",
"positionId": "P-2026-0010",
"jdText": "string, required",
"resumeText": "string, required",
"interviewTranscriptText": "string, optional in pre-screening stage",
"interviewerNotes": "string, optional",
"metadata": {
"candidateName": "string, optional",
"appliedRole": "string, optional",
"audience": "recruiter | candidate",
"outputMode": "json | both",
"outputLanguage": "zh-CN | en | other BCP-47 language tag",
"language": "zh-CN",
"interviewRound": "screening | technical | final | optional",
"createdAt": "ISO-8601 timestamp",
"source": "manual-upload | ats-sync | api"
}
}Input validation rules
jdTextandresumeTextmust not be empty.interviewTranscriptTextis required for post-interview reporting.- When transcript length is too short, output must include
insufficientEvidence. - This schema is only for
outputMode: jsonoroutputMode: both.
3. Output: evaluationBundle
{
"version": "1.0",
"candidateId": "C-2026-0001",
"decisionPolicy": {
"passMin": 75,
"holdMin": 55
},
"preScreening": {},
"interviewChecklist": {},
"postInterviewReport": {},
"markdownBundle": {
"preScreening": "string, Markdown 正文,对应初评报告结构",
"interviewChecklist": "string, Markdown 正文,对应面试准备清单结构",
"postInterview": "string, Markdown 正文,对应面试后综合评价结构"
},
"traceability": {
"evidenceQuotes": [],
"insufficientEvidence": []
}
}Optional dual delivery rule (JSON + Markdown)
- Canonical machine-readable payload: all scored fields and structured lists live under
preScreening,interviewChecklist,postInterviewReport, andtraceability. - Human-readable artifacts: when
outputMode: bothis requested, populatemarkdownBundlewith Markdown documents whose headings and bullet structure follow the audience-specific templates in this skill. WhenoutputMode: jsonis requested,markdownBundlemay be omitted or left empty. Content must be derived only from the same JSON objects; do not add factual claims in Markdown that are absent from JSON. - Agent/chat response order: emit a fenced
jsonblock containing the fullevaluationBundle, then list any Markdown files written.
In default Markdown mode, this schema is not required.
4. preScreening object
{
"overallScore": 0,
"subScores": {
"workExperience": 0,
"professionalCapability": 0,
"softSkills": 0
},
"dimensionEvidence": {
"workExperience": {
"score": 0,
"rationale": "string, 1-3 sentences: why this score for this dimension",
"evidenceQuotes": ["string, 1-2 quotes grounded in jdText/resumeText/transcript"]
},
"professionalCapability": {
"score": 0,
"rationale": "string",
"evidenceQuotes": ["string"]
},
"softSkills": {
"score": 0,
"rationale": "string",
"evidenceQuotes": ["string"]
}
},
"decision": "通过 | 待定 | 拒绝",
"decisionReason": "string",
"risks": ["string"],
"confidence": "high | medium | low"
}dimensionEvidence rules
- Required for new pre-screening outputs: each of
workExperience,professionalCapability, andsoftSkillsmust be present. - `score` (0–100) must equal the same key under
subScores(single source of truth for display: usesubScoresfor weighted overall calculation;dimensionEvidence.*.scoremirrors it for per-dimension narrative). - `evidenceQuotes`: must be attributable to inputs (resume/JD/transcript); duplicate or cross-reference entries in
traceability.evidenceQuotesas needed. - Backward compatibility: legacy bundles may omit
dimensionEvidence; agents implementing this schema should populate it for all new pre-screening reports.
5. interviewChecklist object
{
"focusAreas": [
{
"topic": "string",
"whyItMatters": "string",
"priority": "high | medium | low"
}
],
"questionList": [
{
"question": "string",
"targetCompetency": "string",
"followUpHints": ["string"]
}
],
"interviewerPreparation": {
"mustVerify": ["string"],
"redFlagsToProbe": ["string"]
}
}6. postInterviewReport object
{
"overallScore": 0,
"professionalCapability": {
"roleFit": 0,
"industryExperience": 0,
"skillsDepth": 0,
"summary": "string"
},
"softQualities": {
"communication": 0,
"logicalThinking": 0,
"stressTolerance": 0,
"learningAgility": 0,
"summary": "string"
},
"personalStyle": ["外向开朗", "沉稳踏实", "积极向上", "目标清晰"],
"motivationAndExpectation": {
"motivation": "string",
"expectationAlignment": "high | medium | low"
},
"concernsPendingValidation": ["string"],
"processRecommendation": {
"decision": "通过 | 待定 | 拒绝",
"nextStep": "进入下一轮 | 补充面试 | 拒绝并归档",
"why": "string"
},
"confidence": "high | medium | low"
}7. Multi-agent compatibility
- Single-agent and multi-agent modes must output the same
evaluationBundleshape. - Multi-agent mode can add
agentVotesandroundtableSummaryfields, but cannot remove base fields. - When
markdownBundleis present, it must reflect the merged final scores and narrative, not per-role draft unless explicitly requested.
Scoring Rubric
1. Shared scoring model
Use 0-100 scores for:
- Work experience: 30%
- Professional capability: 40%
- Soft skills: 30%
Formula:
overallScore = workExperience*0.3 + professionalCapability*0.4 + softSkills*0.3
Dimension guidance:
- Work experience: role relevance, ownership level, outcome impact, continuity, and stability signals.
- Professional capability: JD skill match, domain transferability, complexity handled, problem-solving depth, and technical judgment.
- Soft skills: communication clarity, collaboration signals, learning initiative, resilience, and growth evidence.
2. Recruiter decision thresholds
Use only for audience: recruiter:
通过: overall score >= 75待定: 55 <= overall score < 75拒绝: overall score < 55
Every recruiter conclusion must explain:
- why this score
- what evidence supports it
- what uncertainty remains
- what must be verified next
3. Candidate fit levels
Use for audience: candidate; do not present these as hiring predictions:
高匹配: fit score >= 75中等匹配: 55 <= fit score < 75需要补强: fit score < 55
Candidate reports must include preparation priority:
high: likely interview blocker or major evidence gapmedium: important differentiator or likely follow-up topiclow: useful polish, optional expansion, or low-risk improvement
4. Post-interview scoring
Suggested aggregation:
- Professional capability group: 55%
- Soft qualities group: 35%
- Motivation and expectation alignment: 10%
Professional capability group:
- Role fit
- Industry experience transferability
- Skill depth
Soft qualities group:
- Communication ability
- Logical thinking
- Stress tolerance
- Learning agility
5. Confidence rules
high: conclusions supported by 3+ strong evidence pointsmedium: conclusions supported by 2 evidence pointslow: key dimensions rely on assumptions, missing material, or weak evidence
6. Evidence rules
- Every dimension score under 60 or above 85 must include at least one explicit evidence quote.
- Every high-impact recruiter recommendation (
通过or拒绝) must include at least 2 evidence quotes. - Every candidate weakness must include a concrete preparation or repair step.
- Do not invent facts. If evidence is absent, state the uncertainty and ask for the missing material.
7. Risk flags catalog
- Stability risk: frequent job changes or weak commitment signals.
- Adaptability risk: weak change handling or fixed-mindset signs.
- Culture-fit risk: values or collaboration mismatch.
- Capability-gap risk: core JD skills missing or weakly evidenced.
- Motivation mismatch risk: expectations not aligned with role scope, compensation, or growth path.
Related skills
FAQ
What inputs does it require?
The minimum required fields are jdText and resumeText; if either is missing it will not score and returns an insufficientEvidence section asking only for the missing material.
Who is it for?
Two audiences: recruiters (HR, hiring managers, interviewers) for candidate evaluation, and candidates for role-fit self-assessment and interview prep.