Code of Conduct

nmg.labs is committed to providing a safe, inclusive, collaborative and intellectually honest environment for all builders. We expect respect, professional integrity and sportsmanship from every individual who participates in the Forge hackathon series.

1. Safety & Anti-Harassment
   We enforce a strict zero-tolerance policy against any form of harassment, discrimination or abusive behavior. This includes, but is not limited to, offensive verbal comments, written remarks, physical intimidation, disruption of pitches or inappropriate physical contact related to gender, sexual orientation, disability, race, religion or technical choice.
2. Originality & Academic Honesty
   The logic, architecture and core custom features of your submitted Claude Code Skills must be your own work. Forking an existing open-source repository or copy-pasting code written entirely by another builder without attribution is strictly prohibited and constitutes grounds for immediate disqualification. meaningful extension of public frameworks is encouraged but must be clearly documented.
3. Mandatory Official Stack
   Every builder must work on the one official Forge Sprint 01 stack: Claude Code driven by an open model through Ollama, launched with ollama launch claude. You may run the model either way, both are free: Track A on Ollama's cloud free tier, or Track B fully local on your own machine. Use a tool-trained model from the Builder Handbook. This is the only permitted setup. Using a hosted Anthropic API key, Claude Pro, Groq, Cerebras, Google Antigravity, Gemini, OpenAI, Cursor or any other provider or routing tool is not allowed, and submissions produced on any other stack will not be judged. Full setup instructions are in the Builder Handbook.
4. The Triangle of Attribution (Anti-AI-Cheating Rules)
   To prevent builders from misrepresenting fully AI-generated templates as their own custom software solutions, every Forge submission must supply three verifiably incremental proof points:
   • Saved Chat Logs (`agent-log.md`): An unedited saved record of your prompts and agent loops with Claude Code showing the incremental development.
   • Incremental Git History: A public GitHub repository commit timeline displaying at least 10 separate, logical commits spaced out over the 6-hour build window (no single giant "initial commit").
   • 90-Second Walkthrough Video: A live, unedited, single-take Loom screen recording showing you personally running the agent end-to-end on your local Claude Code and Ollama setup against a fresh test case.
5. The 8 Disqualification Triggers
   A participant or submission is automatically disqualified if any of the following parameters are true at or after the submission lock:
   • The submitted GitHub repository is private at the submission deadline.
   • The Loom walkthrough video is missing, highly edited or features a speaker other than the registered builder.
   • The `agent-log.md` conversation file is missing from the repository root.
   • The git commit history displays a single initial upload instead of incremental phases.
   • The skill cannot be run on the auto-tester by the judging panel within 5 minutes of setup.
   • The Code Reading score in the finalist round is 0 (the builder cannot explain any block of their own code).
   • Plagiarism analysis reveals a 75%+ verbatim code overlap with any public skill repositories.
   • Violation of the physical conduct rules at the venue (harassment, intoxication or property damage).
6. Reporting Conduct Issues
   If you experience, witness or suspect a breach of this Code of Conduct, report it immediately to our staff or email us confidentially at labs@nmgdigital.com. We take all reports seriously and will investigate and act swiftly to protect participant safety.