使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/gaket/talent-agent --skill tailor-resume开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
tailor-resume
Inputs
- Target JD (and
score-fitoutput if available). profile/cv.mdandprofile/facts.json(source of truth).
Steps
- Extract the JD's top 6–8 keywords (languages, frameworks, ML/data tools, responsibilities).
- Select content: education first for students; then the 2–3 most relevant experiences/projects by keyword overlap. Drop weak items to stay on one page.
- Rewrite bullets: action verb + what + tool + result. Reuse metrics only if present in facts.json. Mirror JD wording only where the fact supports it (e.g. "REST API" if FastAPI service exists).
- Reorder the skills line to put JD-matching, evidenced skills first. Never add a skill not in
facts.json.skillsor a project stack. - Verify: produce a mapping table — each bullet → source fact id (
exp1,proj1…). Any bullet without a source is removed or flagged as[NEEDS FACT]and you ask the user. - Output to
output/resume-<company>-<role>.md(default) or.tex(simple one-column LaTeX, no exotic packages). Keep to one page: ≤ ~550 words, ≤ 4 bullets per entry. - Show a short diff summary: what moved up, what was cut, keywords covered / still missing (missing ones go back to
score-fitgaps, not into the resume).
Rules
- No invented experience, employers, dates, metrics, or skills. Rephrasing is fine; fabrication is not.
- Keep contact info and education dates exactly as in facts.json.
- ATS-friendly: no tables/columns/images in the resume itself; standard headings.

