Claude's Code Dashboard Tracks 19M+ AI-Generated Commits on GitHub
Claude's Code is a dashboard that tracks commits generated by Claude Code on GitHub public repositories. The creator estimates around 19 million commits, though this only includes public repos with signed commits (via author field or commit trailers).
Technical Implementation
The dashboard is built with Next.js, Recharts for graphing, and PostgreSQL for the database. The developer initially considered BigQuery for analytical scale but switched to PostgreSQL due to cost concerns with small writes and frequent reads for deduplication.
The ingestion pipeline evolved from an under-engineered start to a bare-bones but capable ETL system. The main challenge was GitHub's API rate limits: 30 requests/minute on the search API and 5000 requests/hour on the GraphQL API per access token.
The pipeline splits work between:
- Search workers that write basic commit info to a table, paging and splitting to find commits
- Enrichment workers that read from that table and fill in additional info like lines added/deleted and repo information
There's currently some lag in reading commits as the system is still pulling historical data, which explains why recent dates show lower commit counts and some repos don't have language data yet.
Current Statistics
The dashboard shows:
- Total commits: 14,856,060
- Lines added: +29,053,921,169
- Lines deleted: -12,157,124,674
- Active repos: 800,832
- New repos this week: 71,517 original repos (non-forks) with their first observed Claude Code commit
- Net code delta: +29.7B lines (added minus deleted)
Language Distribution
Top 3 languages by commit count:
- TypeScript: 35.3% (6.99M commits)
- Python: 19.2% (3.8M commits)
- JavaScript: 10.3% (2.05M commits)
Early Adoption
The earliest observed public-era Claude Code commits appear around February 24, 2025. One example is moinmir/ClashOfCans with the commit message: "Change initial game setup to always have exactly 1 correct can. This improves the starting condition by ensuring players always begin with exactly one can in the correct position, making the initial game state more consistent." This commit includes the trailer: Co-Authored-By: Claude <[email protected]>
Recent Activity Examples
The system activity log shows recent commits with Claude Code signatures:
- ErikEvenson/architect: "RAG index: add web UI controls, background tasks, and upload indexing" with
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]> - mearley24/AI-Server: "fix: remove IPv6 disable sysctl from vpn service" with
Co-Authored-By: Claude Opus 4.6 <[email protected]> - PawelDziaduch/grafik: "feat: add user clearing function and detailed logging" with
Co-Authored-By: Claude Haiku 4.5 <[email protected]>
The developer notes the project isn't 100% complete and wants to improve ingestion and extract more from the data. They're open to suggestions for additional dashboard features or data to track.
📖 Read the full source: HN AI Agents
👀 See Also
120 Prompt Patterns Tested: 8 That Actually Work for Claude Code
A 3-month empirical test of 120 prompt patterns for Claude Code yields 8 actionable commands and 5 validation prompts. Key patterns: L99 (cuts hedging), /ghost (removes AI voice), OODA (structured reasoning), ULTRATHINK (deep reasoning), HARDMODE (constraint debugging).
Murmur: An Open-Source Cron Daemon for Automating Claude Code Sessions
Murmur is a cron daemon that schedules and automates Claude code sessions using a HEARTBEAT.md file for configuration.
n8n-mcp-lite: MCP server reduces token usage by 80% for Claude with n8n workflows
A new open-source Model Context Protocol server called n8n-mcp-lite helps Claude reason about n8n automation workflows while reducing token usage by approximately 80%. The tool addresses the token-heavy nature of visual node automations by providing targeted workflow scanning and surgical updates.
wearehere browser extension scans sites for tracking and privacy risks
wearehere is a browser extension that scans websites across ten categories including cookies, trackers, device fingerprinting, and dark patterns, then scores them based on privacy risks. It's under 200KB, runs locally in the browser, and also comes as an npm package for integration with AI agents via barebrowse MCP server.