AI product builder and founder of Clipwise and Xod, based in San Francisco. Replit Ambassador since January 2026. I ship full-stack products solo with AI coding tools (React frontends, Python backends, MCP servers, agent pipelines), then teach other people to do the same: 200+ hackathons, hands-on workshops across the Bay Area, and mentoring non-technical founders.

The pre-flight and coordination layer for open-source contribution. It answers the questions no other tool answers: is this issue secretly taken, does this repo's policy allow your PR, who else is racing you. Timeline X-Ray deep-verifies issues against their timelines; friction scores and an Agent Policy Index (505 repos mapped) decode each repo's gates; PR Radar clusters duplicate and competing PRs across a repo's whole queue; PR Brief writes an evidence-verified review brief for any PR. Ships as a CLI and an MCP server with 27 tools, born from vetting real issues by hand, then automated end to end. 965 tests, CI green on Node 20 and 24.

Release v0.15.0

Set up your agent in one command: npx @reneturcios/greenlight init --agent claude-code (also Cursor, Copilot, Windsurf, Cline). All four toolkit packages are on npm under @reneturcios/ and listed in the official MCP Registry as io.github.CloudCorpRecords/greenlight, prwatch, claimcheck and gatefile.

It also runs the pipeline after the verdict: claim comments with repo conventions baked in, CI pre-flight that runs a repo's checks locally before you push, CI failure triage (yours vs pre-existing), review-response triage (real findings vs bot noise, drafts only), a maintainer guard Action with opt-in Autopilot labels, repo digests that arbitrate duplicate-PR races, a PR babysitter that watches your open PRs, honest base-vs-head test verification, and evidence packs that cite every claim they make.

Newest: PR Radar, which clusters a repo's open PRs into duplicate and competing sets with a labeled keep heuristic (its first live run found two PRs closing the same issue that nobody had connected), and PR Brief, a one-screen verified brief for any PR: change shape, CI base-vs-head, findings classified against evidence. Plus the Claim Protocol, an open spec for claiming issues with expiring leases and a public queue, with a live Claim Ledger and a hosted pre-flight API. And optional AI claim detection: with a Fireworks key set, a cheap model reads the comments the patterns cannot parse and flags hidden claims with the quote as evidence (it can only ever make a verdict more careful, never more confident). Field data lives on the Contribution Friction Index and the Agent Policy Index.

Greenlight runs my own contribution campaign: 88 PRs authored in real open-source repos so far, 26 merged, every one of them vetted by the tool first.

TypeScript MCP CLI GitHub API

Greenlight is one of four tools I built out of a single day of real open-source contribution work, one per stage of the lifecycle:

• prwatch: after you submit. Ranks your open PRs by what needs attention (changes requested, unresolved threads, real failing checks, conflicts, reviewer silence), with external blockers like Vercel authorization classified separately
• claimcheck: while you work. Finds stale issue claims aging with no PR, and parked fixes: finished work on fork branches that never became a PR
• gatefile: for maintainers. Extracts a repo's hidden contribution gates from its docs into a machine-readable contribution-gates.json, with a drift validator for CI

Adjacent to the toolkit: pkgverdict — pre-install verdicts for npm packages (GO / RISKY / FAKE / MISSING), built for humans and AI agents.

Both flagship projects are MIT-licensed and open to contributors of every level, including first-timers.

• ⭐ Star or watch Xod and TrainerBoi to follow along, and follow me here on GitHub for new builds.
• 🐣 Make your first PR by picking a good first issue in Xod or in TrainerBoi. Comment on it and I'll help you get set up.
• 💬 Ask, share or suggest in Xod Discussions or TrainerBoi Discussions.
• 🐛 Found a bug? Open an issue on Xod or TrainerBoi.

• Clipwise — MCP-first AI video clipping: submit a long video, get captioned vertical clips back, each moment scored for hook, story and payoff. Agents can drive the whole pipeline over MCP. Live: clipwise.replit.app
• SudoHired — AI platform for automated technology recruiting
• TaskCato — an accessible agent harness that lets everyday users run AI agents on projects and workflow automation
• Agentcies — an AI agent platform for small businesses, with infrastructure for agent-to-agent communication
• mcp-directory — searchable directory of MCP servers with a live in-browser playground: connect to any server and call its tools, no install needed
• agent-evals — open eval harness + leaderboard: point it at any AI agent, get scored on tool-use, accuracy, and latency
• mcp-job-tracker — MCP server over a job-application tracker: ask Claude about your pipeline, find stale leads, check reply status
• resume-sync — CLI that regenerates your resume's Projects section from GitHub pinned repos, as Markdown or DOCX
• AI-Course-Creator — a complete course on building software with AI, the same material I teach in workshops
• system-monitor — macOS system monitor with a built-in agentic AI assistant (Ollama), live CPU, memory, disk, network and battery views

• IBM ContextForge — #181 (merged): moved prompt/resource argument forms onto the shared accessible Field component, with the tests and i18n cleanup that go with it
• Sage — #302 (merged): extended the docs link checker to cover the knowledge hub, with regression tests; the maintainer independently re-verified it
• pdfcn (shadcn-labs) — six registry blocks and components in review, each claimed, built and review-fixed the same day: Business Letter, Resume, Cover Letter, Project Proposal, Quarterly Report, Addresses
• Reticle — #1620 (in review): fixed the destructive-action guard judging selectable items by their whole text, with red-first regression tests and a changelog entry per the repo's house rules
• Reticle — #1633, #1634, #1635 (in review): upload through hidden dropzone inputs, a pre-writable selector in the split-text hint, and geometry assertions (fully in viewport, overlap, clipped text) for element predicates; #1636 makes a failed browser launch name the exact Chromium path it checked
• Firecrawl — #5141 (in review): fixed the map and extract search builders corrupting URLs by stripping www. from anywhere in the string instead of only the hostname, with a shared helper and unit tests for the corrupted cases
• Firecrawl — #5164 (in review): made the crawler's JS link-filter fallback enforce the same host scoping as the native filter, so crawls cannot silently wander off-site when the native filter fails
• FastMCP — #5757 (in review): wired the per-server timeout in MCP client configs into the session read timeout, so a config that looks capped no longer runs unlimited
• Squidbrake — #91 (merged): added a taint signal so approvers see when an agent edits its own setup right after reading outside content
• Plus more in review across Go, Python and TypeScript projects: all my pull requests

• 🏗️ Build — full-stack TypeScript/Python apps, MCP servers, AI agent systems, local-first AI apps
• 🎓 Teach — hands-on workshops on building with Replit, Cursor, and Claude; AI course creator
• 🤝 Mentor — help non-technical builders ship with AI

• 🏆 2024 NAB Show Product of the Year (AI/ML category), as a builder on ARwall's ARFX XR/AI Filmmaking Suite
• 📰 Profiled in The San Francisco Standard (July 2025) for AI-assisted building and San Francisco's hackathon community

TypeScript React Python FastAPI MCP MLX PyTorch Electron Stripe Claude OpenAI Cursor Replit

• 🌐 Portfolio: reneturcios.replit.app
• 💼 LinkedIn: linkedin.com/in/reneturcios
• ✉️ Email: [email protected]
• 📍 San Francisco, CA — available now for Forward Deployed, Applied AI and DevRel roles, and for freelance AI builds