Find engineers by what they've merged. Not what their résumés claim.

GitHub Talent Scout is a skill file your coding agent runs against GitHub's public REST API. Give it a role. It maps the open-source projects that matter, pulls merged pull requests and code reviews, and returns a ranked shortlist. Every claim links to the code behind it.

Plain markdown. Runs locally under your own GitHub token. Nothing goes anywhere except GitHub and your model provider.

A résumé is a claim. A merged PR is a reviewed artifact.

Keyword search finds people who wrote "Kubernetes" on a profile. It can't tell you who fixed the scheduler bug you hit last quarter.

Public GitHub history can. A merged pull request to a serious project has passed review by maintainers who had no reason to be generous. Review comments show how someone reasons about other people's code. Sustained contribution shows judgment over time, not a weekend.

This doesn't replace interviews. It changes who gets one.

Right column is an illustrative example, not a real engineer.

GitHub publishes an API anyone with an account can query. Most hiring teams don't know it exists.

GitHub's REST API is the same documented interface developers use to build integrations. For public data it returns repositories, pull requests, code reviews, and public profile fields as structured JSON. It's free, and you can verify every claim on this page from GitHub's own documentation.

gh api rate_limit --jq '.resources | {core, search, code_search}'

What GitHub allows, per its documentation

GitHub publishes no daily search quota. Limits are enforced per minute and per hour, secondary limits also apply, and GitHub can change any of them. In practice, search is what paces a run, not the 5,000-an-hour core limit.

A sourcing playbook your agent executes with gh.

The skill doesn't ship a scraper or a hidden service. It's an instruction file. It tells your agent what to query, in what order, what to read, and how to score what it finds. Your agent runs standard GitHub CLI and REST calls under your authenticated account.

# 1. Map the ecosystem: which repos matter for this role
gh search repos "query engine" --language=rust --sort=stars --limit 20

# 2. Pull merged work, not activity
gh search prs --repo apache/datafusion --merged \
  --merged-at ">2025-04-01" --limit 100 --json author,title,url,closedAt

# 3. Read the evidence: what changed, who reviewed it
gh pr view <number> --repo apache/datafusion \
  --json files,additions,deletions,reviews

# 4. Public profile context only
gh api users/<login> --jq '{login,name,company,blog,location}'

What counts as evidence. What gets ignored.

Green squares are easy to farm. The skill weights work that another engineer had to review and accept, and discounts activity nobody checked.

Your first search starts with one role spec.

Vague prompts get vague shortlists. Name the repos you know, the modules that matter, a time window, exclusions, and the output you want. Pick an example, copy it, and edit the specifics.

From thousands of contributors to a shortlist you can defend.

The output is a table your hiring team can audit. Each rating carries the PRs that justify it, the volume behind it, and what the data can't tell you.

Powerful research. Honest limits.

The API gives you the data. It doesn't give you the whole market, and it comes with rules.

Five minutes if you already use gh.

You need three things: a GitHub account, an authenticated GitHub CLI, and a coding agent that supports skills.

gh auth status                          # confirm you're logged in
unzip github-talent-scout-skill.zip
mkdir -p ~/.claude/skills
mv github-talent-scout ~/.claude/skills/

# then, inside Claude Code:
# /github-talent-scout  + paste your role spec

gh auth status                          # confirm you're logged in
unzip github-talent-scout-skill.zip
mkdir -p ~/.codex/skills
mv github-talent-scout ~/.codex/skills/

# then, inside Codex:
# $github-talent-scout  + paste your role spec