Live README badge
Paste one line in your GitHub profile or project README. Updates within 5 minutes of any XP change. Wakatime-style — ~1M badges in the wild.
Copy my badge →Works with the stack you already use
Native integration
Client
Client
BYOO OTEL
BYOO OTEL
BYOO OTEL
BYOO OTEL
BYOO OTEL
BYOO OTEL
Distribution
Distribution
Get started in 60 seconds
Two commands. No config files.
Install the CLI, run scoreflow init once, and a Stop hook wires into Claude Code. No Docker. No sidecar. No SDK to import.
~/dev
$ npm i -g @scoreflow/cli $ scoreflow init ✔ Signed in as diego@pdata.org ✔ Stop hook installed at ~/.claude/settings.json ✔ You're live. Keep working — your first score lands Monday.
1
Install the CLI
Single npm package. Works on macOS, Linux, and WSL.
2
Run scoreflow init
One-time setup: OAuth to your team and wire the Claude Code Stop hook.
3
Keep working
Every prompt + tool call flows to Scoreflow. First score drops Monday.
Or install as a Claude Code plugin
One line inside Claude Code adds the MCP server + four slash commands (status, leaderboard, profile, explain).
/plugin marketplace add 8infinitelabs/scoreflow /plugin install scoreflow@scoreflow
Prefer opencode? codex? aider? Any OpenTelemetry CLI connects with three env vars — see the BYOO OTEL guide.
Grab your CLI token →How it works
Three moving parts. Nothing you can game.
Measurement that engineers respect because it's rooted in their real tooling, not a survey.
01
Ingest the signal, not the content
OpenTelemetry from Claude Code / Copilot / Cursor, GitHub webhooks for PR outcomes, admin APIs for per-user attribution. Event metadata only — never prompt or output text.
02
Score with bands, not monotones
Four components — Effectiveness, Craftsmanship, Outcome, Breadth — each using sweet-spot bands so gaming the metric collapses the score. Weekly snapshot per team + rolling 28-day windows.
03
Deliver where devs live
MCP server so devs see their score inside Claude Code. Weekly email digest for managers. Web dashboard for deep dives. Live CLI hooks that show XP + level-ups in the terminal.
The Craftsmanship Score
Four components. Each one rewards judgment.
Weights are published. Bands are documented. If a signal can be gamed, the band punishes the gaming so the score doesn't.
Your character sheet
Level up as you ship. Respectful. Earned.
Every profile is a living character sheet: 10 tiers from Karpathy to Turing, 11 skill tracks, 22 achievements. XP derives from CCS so the bands stay anti-gaming. Share it, compare it, unlock the next tier.
100 levels · 10 tiers
From Karpathy to Turing
1-10KarpathyNeural nets from scratch
11-20AltmanProductized the LLM era
21-30HassabisDeepMind, AlphaFold
31-40LeCunCNNs, Meta FAIR
41-50HintonGodfather of deep learning
51-60TorvaldsLinux, Git
61-70KnuthThe Art of Computer Programming
71-80HopperFirst compiler, coined "bug"
81-90LovelaceFirst published algorithm, 1843
91-100TuringTuring machine, the foundation
11 skill tracks · each 1-100
Every track levels on its own.
Daily XP splits across tracks based on the tools you actually used. Each track has its own 1-100 curve (flatter than character levels — specializing is encouraged). Here's how to grind each one:
File CraftEdit / Write / MultiEdit — surgical diffs
ResearchGrep / Glob / Read / Web* — find the signal
AgentsTask / Agent / SubAgent — delegate work
AutomationBash / Shell — scripts that do the loop
Project MgmtTodoWrite / TaskCreate — track the plan
AnalysisPlan Mode / ExitPlanMode — think first
SynthesisSession spans ≥3 tool categories
WritingPrompts with ≥200 char avg — drafting prose
ConversationTool-less prompts — rubber-ducking
TeachingPublish a craftable peers adopt
CreativityFirst-time (tool × modality) combo
Level curves
Respectful of your time.
Character level is steep: 10 × n^1.7 XP — reaching Turing (Lv 100) takes years of sustained craft. Skills are flatter: 3 × n^1.4 — a specialist can reach Skill Lv 60 in ~3 months. The tier breadcrumb stays meaningful across the whole journey.
Crafting (Phase 2)
Ship artifacts your team actually uses.
Authoring a Skill, Rule, MCP, Prompt, Workflow, or Agent earns XP. Rarity climbs as peers adopt it: common → uncommon → rare → epic → legendary. Legendary = real evidence you shipped something the team rallied around.
Viral by design
Every surface is a share surface.
Scoreflow doesn’t buy ads. It ships in six share loops — README badges, OG cards, quizzes, PR comments, monthly Wrapped, public team pages. Each is live. Click any to see.
Pixel-art character sheet
Every /u/<handle> unfurls with a JRPG-style status screen — level, tier, XP bar, skill tracks, achievements. Designed for the X share moment.
See a live profile →No-signup quiz
Ten questions. Returns a CCS estimate and an archetype (Vibecoder → Master). Built to seed HN and Reddit threads — "what did you get?" as a conversation hook.
Take the quiz →PR comment on every push
Install the Action in any repo. On every opened PR it posts the author's CCS as a comment. Every reviewer becomes an impression — the CodeRabbit loop.
Install the Action →
Monthly wrapped drop
On the 1st of every month you get a personal recap page — XP earned, levels crossed, unlocks. Spotify Wrapped pattern but monthly so it never goes silent.
See a sample wrapped →Team leaderboard with k=5
Every team gets a public page. Aggregates unlock at 5 opted-in members — never individual ranks, always bands. Strava-club pride without the shame.
Open the General team →Privacy by default
Built for engineering leaders who have to defend the tool in the all-hands.
Privacy is a design constraint, not a feature toggle. These four pillars are non-negotiable and shipped from day one.
No prompt content stored
We record metadata — length, token counts, acceptance, tool name. Never the prompt text or the output. Optional premium tier adds customer-side encryption for teams that want richer signals.
Team aggregates by default
Every manager-facing view is team-level. Individual rankings require explicit opt-in from the user being shown. Enterprise mode enforces team-only with a configurable k-anonymity floor (default k=5).
EU data residency
All stored events + scores live on Hetzner Frankfurt. Identity (WorkOS), payments (Lemon Squeezy), and email (Resend EU) run on edge providers — never the hot path of ingestion or storage.
Works-council mode
Enterprise contracts include a works-council profile: redaction below k=5, DPIA templates, audit logs on every individual-level view, delete-on-offboarding with retention policies per regulation.
Pricing
Free for everyone during MVP.
We're proving the value prop first. Team Pro + Enterprise are on the roadmap — pricing will be feature-based (connectors, SSO, retention), never per-dev or per-event. Signing up now locks your founding-user status.
All yours
Everyone
$0
while we find the right price
Full scoring, gamification, and team features. Use it freely while we earn your trust.
- ✓Unlimited seats per team
- ✓All scoring + gamification (CCS + XP + levels + achievements)
- ✓Claude Code CLI + MCP + weekly digest
- ✓Global leaderboard (opt-in, default yes for community teams)
- ✓Public profile on scoreflow.dev/u/<handle>
- ✓Anthropic Analytics connector (Claude.ai web/desktop/mobile)
Coming soon
Team Pro
Coming soon
Premium features for teams ready to invest in AI craftsmanship — pricing under review.
- ✓SSO + SCIM
- ✓Extended retention (12mo+)
- ✓Custom branding on /u/<handle>
- ✓Team Pro Badge
- ✓Priority support
- ✓Early access to new scoring components
Enterprise
Custom
annual, invoice
For 50+ devs with works-council-friendly visibility, SLA, and DPA.
- ✓Works-council mode with k-anonymity floor
- ✓DPIA templates + Data Processing Agreement
- ✓Self-hosted option (Hetzner or your infra)
- ✓Dedicated onboarding
- ✓SLA + audit trail export
FAQ
Questions engineering leaders actually ask.
If there's something not covered here, email diego@infinitelabs.co — every question so far has shaped the product.
Won't this just Goodhart itself? Devs will game whatever you score.
Every component is banded, not monotone — spamming accepts, rubber-stamping, token-bursting, or artificial breadth all collapse the score. The anti-gaming flags are published in packages/scoring/src/anti-gaming.ts so the mechanism is auditable.
Do you store our prompts or generated code?
No. We store metadata only: tool name, length, accepted/rejected, timestamps, modality. Never prompt or output text. An opt-in premium tier can ingest richer signals under a customer-side encryption model — same default stays off.
How is individual vs. team visibility handled?
Community teams default to opt-in-with-default-yes: users see their own level publicly and participate in the global leaderboard unless they flip it off. Enterprise mode defaults to team-aggregate-only with a k-anonymity floor (k=5 default) — any view below that is redacted automatically.
What do you need to get started?
A GitHub account and permission to install the ScoreFlow GitHub App on the repos you want scored. For tools like Claude Code / Cursor / Copilot we can attribute via OTEL traces or admin API — whichever your stack supports.
Is this really not a subscription?
Team Pro is a one-time $149 lifetime license up to 50 seats. Enterprise is annual with a custom quote because of the SSO/SCIM/works-council work. Community stays free forever for teams of 5.
Where does the data live?
Hetzner Frankfurt (EU). Identity via WorkOS, payments via Lemon Squeezy (MoR, handles EU VAT), email via Resend EU — those are edge services, never hot-path ingestion or storage.