A Claude Agent Skill
that audits and optimizes a resume against automated LLM resume-screening
agents, reproducing the exact 120-point rubric from HackerRank's open-source
interviewstreet/hiring-agent.
Modern screeners are not keyword ATS filters. They read the resume PDF, pull the candidate's live GitHub via API, classify every repo as open-source vs personal, and score four weighted categories plus bonuses and deductions. This skill teaches Claude to see what the agent sees and to fix the highest-leverage gaps, using real signals and honest framing only.
- Reproduces the agent's GitHub classification (which caps the open_source score) with a zero-dependency Python script.
- Scores a resume against the full reverse-engineered rubric.
- Returns a gap table sorted by point-swing, plus concrete edits.
- Refuses to fabricate experience, PRs, stars, or metrics.
Claude Code / claude.ai / API: place this folder so SKILL.md sits at the
skill root (for Claude Code, drop it in ~/.claude/skills/resume-radar/).
The skill auto-activates on prompts like "simulate my resume score" or "help me
beat the resume bot".
python scripts/github_enrich.py <github-username-or-url>
Set GITHUB_TOKEN to raise the GitHub API limit from 60/hr to 5000/hr. Then ask
Claude to audit, simulate, or optimize a resume.
You: "Simulate the hiring-agent score for my resume and tell me where I'm losing points."
Claude runs the enrichment script, which reports the same classification the screener computes:
{
"username": "example-dev",
"public_repos": 23,
"open_source_repos": 0,
"self_project_repos": 15,
"all_repos_are_self_project": true,
"open_source_cap_triggered": true
}Then it scores the resume and returns a gap table sorted by point-swing:
| Category | Score | Ceiling | Highest-leverage fix |
|---|---|---|---|
| open_source | 8 / 35 | +25 | land one merged PR to a 1000+ star repo (personal repos are capped at ~10) |
| self_projects | 23 / 30 | +7 | add a working live-demo URL to each project |
| production | 20 / 25 | +5 | quantify scale/impact of shipped work |
| technical_skills | 9 / 10 | +1 | already near max |
| bonus | 6 / 20 | +3 | add the portfolio/blog URL that is missing from the resume |
Every suggestion is a real, honest action; the skill never invents experience.
| Path | Purpose |
|---|---|
SKILL.md |
Skill instructions and workflow (loaded when triggered) |
references/RUBRIC.md |
Full 120-point rubric (loaded on demand) |
scripts/github_enrich.py |
Reproduces the agent's live GitHub classification |
scripts/test_github_enrich.py |
Offline unit tests for the script |
evaluations.json |
Behavioral evals for the skill |
This skill optimizes what a candidate genuinely has: it surfaces real merged pull requests, adds working demo links, cleans up a cluttered GitHub, and ensures real profile URLs are present. It never invents experience or contributions. The screening agent reads the live repo and a human reads the PDF, so dishonest signals get caught.
Rubric reverse-engineered from the MIT-licensed
interviewstreet/hiring-agent
by HackerRank. This project is not affiliated with or endorsed by HackerRank.
MIT