Skip to content

SKILL.md

Latest commit

 

History

History
138 lines (105 loc) · 4.98 KB

SKILL.md

File metadata and controls

138 lines (105 loc) · 4.98 KB
name llm-intern-skill
description Use when polishing, diagnosing, tailoring, or exporting resumes for LLM, RAG, Agent, Agentic RL, post-training, pretraining, AIGC, search/ranking, multimodal, AI backend, or LLM algorithm internships from raw resume text, a materials folder, and/or a target job description. Audits evidence, maps JD fit, enforces truth boundaries, writes polished and targeted resumes, generates interviewer-style grilling questions, answer cards, evidence-upgrade plans, and optional open-source project recommendations without fabricating experience.

LLMInternSkill

Use this Skill when the user wants resume polish, resume diagnosis, JD tailoring, project packaging, interview preparation, or final resume export for LLM-related internship applications.

Core rule:

Do not fabricate. Diagnose first, polish second.

Inputs

Preferred input folder:

materials/
├── target_jd.txt
├── resume.md / resume.pdf
├── projects/
├── code/
├── notes/
├── papers/
├── awards/
└── other/

If the user only provides a JD and no materials, ask the intake questions from templates/intake.md.

If the user only asks for resume polish, run a lightweight version:

raw resume line -> claim extraction -> evidence/risk check -> polished wording -> interview risk

Main Workflow

  1. Decide the mode

    • Resume polish only: use skill-references/resume-polish.md.
    • JD tailoring: use skill-references/jd-analysis.md and skill-references/resume-tailoring.md.
    • Full materials folder: run the complete workflow below.
    • Interview prep only: use skill-references/interview-grilling.md and skill-references/answer-cards.md.
    • Project Scout only: use skill-references/project-scout.md.

  2. Read the target JD when present

    • Use skill-references/jd-analysis.md.
    • Detect role type: RAG, Agent, Agentic RL, post-training, pretraining, LLM app, LLM algorithm, search/ranking, AIGC, multimodal, backend AI, infra, or mixed.
    • Load the matching role file under skill-references/roles/ when relevant.

  3. Audit the materials folder when present

    • Use skill-references/materials-audit.md.
    • Extract projects, claims, evidence, missing evidence, and unclear ownership.

  4. Set truth boundaries

    • Use skill-references/truth-boundary.md.
    • Classify content as 可以写, 谨慎写, 补证据后写, 不能写, or 无法判断.

  5. Build the evidence contract

    • Use skill-references/evidence-contract.md.
    • Every strong claim needs evidence, risk, safe wording, and interview proof.

  6. Generate polished / targeted resume

    • Use skill-references/resume-polish.md for line-level polish.
    • Use skill-references/resume-tailoring.md.
    • Produce conservative, standard, and stronger-after-evidence bullets.
    • Generate a targeted full resume draft when enough information exists.
    • If the user wants a PDF-ready resume, use templates/resume-latex/bill-ryan-elegant-zh_CN/resume-zh_CN.tex as the LaTeX base.

  7. Generate interview grilling

    • Use skill-references/interview-grilling.md.
    • Ask interviewer-style questions based on JD gaps and resume claims.

  8. Generate answer cards

    • Use skill-references/answer-cards.md.
    • For high-risk questions, produce dangerous / passable / strong answers.

  9. Create upgrade plan

    • Use skill-references/upgrade-plan.md.
    • Split into half-day, 1-day, 3-day, and 1-week evidence upgrades.

  10. Optional Project Scout

  • Use skill-references/project-scout.md when the user's evidence is weak or they ask for projects to learn.
  • Recommend projects only as learning/reproduction/modification opportunities, not as fake experience.
  1. Assemble final pack
  • Use templates/final-pack.md.

Output Files

When writing files, prefer this structure:

output/
├── 01_jd_analysis.md
├── 02_materials_audit.md
├── 03_truth_boundary.md
├── 04_evidence_contract.md
├── 05_resume_polish.md
├── 06_targeted_resume.md
├── 07_interview_grilling.md
├── 08_answer_cards.md
├── 09_upgrade_plan.md
├── 10_project_scout.md
└── 11_final_pack.md

If the user wants only an answer in chat, still follow the same section order.

Fit Verdict

Always give one:

strong fit
weak fit
risky fit
not recommended

Explain the verdict with:

  • JD must-haves.
  • User evidence.
  • Gaps.
  • Highest interview risk.
  • Fastest useful upgrade.

Non-Negotiables

  • Never invent internships, production status, metrics, user scale, model training, ranking gains, or ownership.
  • Do not write "主导" when evidence only supports "参与".
  • Do not write "上线" when evidence only supports demo, local run, or internal trial.
  • Do not write open-source learning as work experience unless the user actually reproduced, modified, and documented it.
  • If materials are insufficient, ask questions or produce a conservative report instead of polished fiction.