An evidence-based platform to find great developers, measure where you stand, and grow from there.
English · 中文
🔍 Score a GitHub profile · 🏆 Discover developers · 🤖 Install the GitHub App · 📚 Architecture
- Measure. Enter a GitHub handle to get a deterministic 0–100 score across six dimensions, a level and rating, the risk signals behind it, and an optional AI-written report.
- Discover. Browse the developer board, language/org/repo directories, project boards and the talent pool, or put two developers head to head.
- Share. Every profile has a live README badge and a light/dark developer card.
- Maintain. The ghfind Review GitHub App labels new issues and PRs with the author's score band.
- Integrate. The same data is available through a REST API, an MCP server, SDKs and a CLI.
-
Deterministic. Six dimensions — account maturity, original projects, contribution quality, ecosystem impact, community influence and activity authenticity — plus capped risk rules for farming patterns (star or fork farming, self-merged PRs, PR floods, bots). Same evidence and version, same score.
-
Hard-to-fake signals win. Merged PRs into real projects and sustained activity weigh more than buyable stars and followers. Only public GitHub data is used, so mostly-private work is underrated.
-
Levels and ratings.
Lv = floor(score). Ratings step every 10 points from 30:Score < 30 30 40 50 60 70 80 90+ Rating Qi Condensation Foundation Est. Core Formation Nascent Soul Deity Transformation Spatial Tempering Body Integration Grand Ascension -
AI never changes the number. The report model writes tags and prose from the fixed score.
-
Open and verifiable. The formula, risk rules and rating ladder are the public module
scoring/(TypeScript, Go and a Python port, locked together by shared fixtures). The SDKs (0.2+) recompute any published scan locally:ghfind verify scan.json. -
Super newcomers. Young accounts with strong, evidenced work can earn up to 4 extra maturity points, fading to zero by age 3 — see the rule.
Install ghfind Review and new issues and pull requests arrive labelled with the
author's public score band (review: labels at 40, 70 and 90; review: no-score
when no score is available). By default it doesn't comment, and it never blocks
or closes anything — a profile score is a starting point for triage, not a
verdict on the contribution. Labels are created automatically; no workflow or
secret is needed.
Full reference: ghfind.com/docs.
- REST —
GET https://ghfind.com/api/score/{username}returns a deterministic score with no auth and no LLM. OpenAPI 3.1: ghfind.com/openapi.json. - MCP — Streamable HTTP at ghfind.com/mcp for Claude, Cursor or any MCP client.
- SDKs + CLI —
@hikariming/ghfind(npm) ·ghfind(PyPI) · the Go CLI. - For agents — ghfind.com/llms.txt links every machine-readable surface.
ghfind is built from three repositories. This one is public:
| Repository | Visibility | What it holds |
|---|---|---|
FindAny/ghfind (this) |
public | The scoring module (scoring/), the JS and Python SDKs (packages/), the Go CLI (cmd/ghfind), agent skills, and every CI/CD and deploy workflow |
FindAny/ghfind-server |
private | API, GitHub collection, storage, Feed, Workers and the workspace configuration |
FindAny/ghfind-frontend |
private | Astro site, React components, messages and assets |
- What you can verify. Every production score is computed by the code in
scoring/. Thedeploybranch records each production release: itscomponents.manifest.jsonnames the exactghfindrelease (and so the scoring code) that is live, next to the commits of the private components. - How it ships. Each repository publishes
vX.Y.Zreleases. The latest verified release of all three is assembled into one workspace and deployed to Cloudflare from this repository. Design: docs/architecture.md; procedures: docs/runbook.md; Chinese deployment overview: docs/deployment.zh.md.
Requires Node 22, pnpm 11, Go 1.23+ and Python 3.9+.
pnpm install
pnpm --filter @ghfind/scoring typecheck && pnpm --filter @ghfind/scoring test
pnpm scoring:boundary # purity, credential and import gate
(cd scoring && go test ./...) # Go port
(cd packages/ghfind-py && python3 -m pytest -q)
(cd packages/ghfind-js && npm ci && npm run typecheck && npm test)
go test ./internal/agentcli && go build -o bin/ghfind ./cmd/ghfindAny formula change updates the TypeScript, Go and Python ports and their shared
fixtures in the same pull request; see scoring/README.md.
Pull requests (forks included) run the scoring CI; the full application suite runs
after merge on main and on every production release. See
CONTRIBUTING.md and AGENTS.md.
- Repository release: a maintainer tags
vX.Y.Zon a greenmain. The release workflow re-checks CI for that exact commit, marks it verified (rel/vX.Y.Z), publishes the GitHub release and starts a production composition. - Scoring package:
scoring/vX.Y.Ztags publish@ghfind/scoringand the Go module (.github/workflows/scoring-release.yml). - SDKs:
@hikariming/ghfindandghfindare built frompackages/.
Choose "Use your own model" on the site and enter a base URL, API key and model for any OpenAI-compatible provider. The key stays in your browser's localStorage, is sent only with your request and is never stored on the server.
Scores and commentary are generated automatically from public GitHub data only. They describe public activity, are not a judgement of a person, are not findings of fact and must not be used for harassment.
Sponsorship helps cover GitHub API, LLM and hosting costs. It never affects any
score, rank or risk conclusion — scores come only from the public
scoring/ module — and sponsor placements are kept separate from
ranking data.
GNU AGPL-3.0. You may use, modify and self-host this project; if you
offer a modified version as a network service, AGPL requires you to publish your
modifications under the same license. The scoring formula began as a port of the
open-source Claude skill github-account-value and has since evolved here in
scoring/, which is now its source of truth.
Trademark: the "ghfind / 毒舌 GitHub 评分" name, logo and domain are not covered by the license. Please don't use them to impersonate the official site.

