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 score-fit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。
score-fit
Inputs
- Job description (URL, pasted text, or rows from
find-roles). profile/facts.json,profile/prefs.yml.- Rubric:
resources/rubric.md.
Steps
- Extract from the JD: title, level, season/start, location/remote, must-have skills, nice-to-haves, years required, degree/grad-date window, sponsorship language ("must be authorized", "no sponsorship", "CPT/OPT ok").
- Check hard blockers first (rubric §Blockers): work auth vs
work_authorization, grad-date window, senior level. Any blocker → cap score at 2 and say why. - Score each dimension in the rubric (skills, experience/level, domain, location/logistics, growth). Cite the facts.json id (e.g.
exp1,proj1) that supports each match. - List gaps: missing must-haves, and for each a cheap fix (course, small project, reframe existing fact — only if true).
- Compute weighted score (1.0–5.0, one decimal). Recommend: ≥4 apply now; 3.5–3.9 apply + tailor; 3–3.4 apply if low effort; <3 skip.
- Output:
Company — Title — Score 4.1/5 — Apply now
Matches: Python/FastAPI (exp1), ML (proj1)
Gaps: Spark (nice-to-have) → mention pandas scale; not a blocker
Years: "0-1 yrs" ✓ Sponsorship: Canada role, citizen ✓- Offer to update tracker (
track-pipeline) and tailor resume for ≥min_score_to_apply.
Rules
- Be honest; don't inflate scores. Never assume skills not in facts.json.
- Flag uncertain sponsorship as "unknown — ask recruiter", not as OK.

