VisiData Sets Disclosure Rules for AI-Assisted Open-Source Contributions

Summary

VisiData maintainer Saul Pwanson outlines how the project is handling the growing number of LLM-assisted pull requests. He says he uses Claude Code and Claude Opus for review, ideation, and software work, while acknowledging both strong results and failures. The project separates human and bot identities: AI-generated comments and commits use the GitHub account @saulbert, while Pwanson writes personally as @saulpw. Contributors are asked to disclose any non-trivial AI-generated code or text, and AI comments must carry a footnote. VisiData’s pull-request template now asks contributors to estimate the balance between human effort and AI involvement through levels 0 to 10. The scale ranges from no AI use, through human-directed assistance and code generation, to autonomous or uninvited bots receiving little or no human attention. At levels 4 and above, contributors must identify the model and version used, and bot commits must identify the human operator. All levels from 0 through 8 are potentially welcome when properly disclosed, but contributions with less human attention receive more skeptical treatment. A human must still vouch for every pull request in good faith, test it, and confirm that it works at least minimally. Pwanson argues that provenance helps maintainers judge the likely review burden, since a contribution generated quickly can take much longer to salvage. Autonomous level-9 and level-10 bots are restricted to explicitly authorized, limited-scope tasks; uninvited bots are treated as spam or potentially hostile activity. The approach is intended to increase useful functionality while preserving trust, accountability, and maintainability in the project.