| name | interview-prep |
| description | Research a company + job description + the candidate's CV, then generate a polished, bilingual (English/中文) single-file HTML interview-preparation report with a Figma-style design system. Use this skill whenever the user shares a job posting URL/JD text, a CV/resume, or asks to prepare for an interview (面试准备/面经/求职), asks "why company / why role" prep, self-introduction scripts, behavioral (STAR) or technical question banks, or multi-round interview planning — even if they don't explicitly say "report". Also trigger for requests like "help me get ready for my interview at X" or "分析这个JD和我的简历匹配度".
|
Interview Prep Report Generator
Turn (JD + CV + company research) into one self-contained, bilingual HTML report the
candidate can study from. The report is the deliverable; research quality is what makes
it worth reading.
Workflow
0. Resolve the workspace (optional — usually there isn't one)
This skill is portable and stateless by default. It does not require any folder
layout, and it must never ask the user to create one.
Resolve <workspace> per references/workspace.md. Three outcomes:
- The user named one → read
<workspace>/candidate-profile.md and any matching
interviews/<date>-<company>-<role>.md before asking the user anything.
- The project obviously has one (a folder containing
candidate-profile.md) → same.
- None → chat-only mode. This is the normal case. Take everything from the
conversation. Do not create folders. Do not treat this as missing setup.
Only if a workspace exists, write back at the end (history table, new intel, absolute
dates). Full contract — including what must never be written to disk — is in
references/workspace.md.
1. Gather inputs
The CV is required; the JD is optional (a URL works). Both may arrive as pasted
text, an attachment, or a file path — all three are equally valid. Never insist on
files, and never copy the user's CV anywhere; read it where it is.
- CV: extract education, skills, projects (with metrics), work history,
LinkedIn/GitHub URLs, soft skills, interests.
- JD: fetch the job URL (or use pasted text). Extract: company, team, location,
responsibilities, required vs. desirable skills, process/closing date.
- If a profile already exists in a workspace, reuse it and only ask for the new JD.
- Ask two quick multiple-choice questions (use the question tool if available;
otherwise ask in chat) before writing:
- What type is this interview? — ① HR/recruiter screen ② technical interview
- How long is it? — ① 30–45 min ② 60+ min
Use the answers to add a "THIS INTERVIEW" battle card at the top of the report
(interviewer type, must-ask list, per-answer time budget, this round's single goal,
which report sections to prioritize) and to weight the content: HR screen → Core 3 +
salary/visa/logistics dominate; technical → project deep-dives + technical bank
dominate; 30–45 min → answers ≤90s, lead with the point; 60+ → prepare follow-up
chains and more questions to ask. If the user doesn't know, note that in the card
and prepare for the most likely first round.
2. Research the company
Run web searches (in parallel where possible) covering, at minimum:
- Company overview: size, brands, founding story, ownership, recent financial results.
- Strategy & tech: digital/AI strategy, recent partnerships, the hiring team's mandate,
budget signals (cost-cutting vs. investment — infer honestly, never invent numbers).
- People: LinkedIn presence of the team/department, likely hiring manager profile.
- Candidate experience: Glassdoor/Seek/Indeed interview reviews — number of rounds,
format (phone/Teams/panel), typical questions.
- Values & culture: official purpose/values (quote them — they appear in interviews).
- Local interview culture for the company's country (see
references/interview-culture.md — read it when the interview is outside the
candidate's home culture, e.g. a Chinese candidate interviewing in New Zealand).
Cite every claim in the report's Sources section. Mark inferences as inferences
(e.g. "budget signal: reset year + new hires ⇒ selective investment") — a candidate who
repeats an invented fact in an interview gets burned.
3. Analyze fit
Build a requirement-by-requirement match table: JD requirement → strongest CV evidence
→ status (strong / partial / gap). For every gap, write a mitigation line the candidate
can actually say (adjacent experience, project work, learning plan). Never leave a gap
without a talking point.
4. Write the report
Read references/report-spec.md for the required sections, and use
assets/report-template.html as the design-system starting point (Figma-style: light
neutral canvas, Inter/system font stack, sidebar TOC, cards, pill tags, no external
dependencies). Fill in real content — never ship placeholder text.
Non-negotiables:
- Single self-contained .html file — inline CSS/JS, system fonts, works offline.
- Bilingual: every substantive element exists in both languages with an EN / 中文 / 双语
toggle. English is what they'll say; Chinese is how they'll understand and memorize.
Spoken scripts (self-intro, answers) must be natural spoken English, not translated prose.
- The three core questions come first: ① 60-second self-introduction (word-for-word
script, timed), ② Why this company, ③ Why this role. These are asked in ~90% of
screenings; the candidate should be able to rehearse them from the report alone.
- Multi-round planning: cover at least screen → hiring-manager (behavioral+technical)
→ final/panel, with per-round prep even if the company's actual process is unknown.
- Behavioral answers use STAR and must reference the candidate's real CV stories —
map each likely question to a specific project/job from the CV.
- Technical section matches the JD, not a generic list (an ETL-heavy JD gets
ETL/ELT, data modelling, SQL, pipeline-debugging questions with model answers).
5. Deliver
Deliver the report to the user — as a file in the conversation, or written to
<workspace>/reports/<company>-<role>-report.html if a workspace exists. Name it
<company>-<role>-report.html either way.
Then summarize in chat: match assessment, top 3 risks/gaps, and the single
highest-leverage thing to rehearse first.
Bundled resources
Everything the skill needs is inside this folder. Drop skills/interview-prep/ into
any project and it works — no sibling folders required.
references/workspace.md — how to resolve an optional workspace, and what must never
be written to disk. Read at the start of every run.references/report-spec.md — required report sections + quality bar. Read before writing.references/interview-culture.md — NZ/AU/Western interview culture notes for
Chinese-background candidates. Read when relevant.references/question-banks.md — behavioral + technical question seeds by role family
(data/AI, software, product). Read while drafting sections 5–6 of the report.assets/report-template.html — design-system skeleton (colors, layout, language
toggle). Copy and fill; don't redesign from scratch.
