Open Source Maintainer Battles 'Agentic Slop' with Claude-Written Guidelines
Key Takeaways
- ▸High-profile open source projects are experiencing a flood of low-quality AI agent-generated contributions, creating new maintenance burdens
- ▸Traditional PR templates and guidelines are ineffective against agentic submissions because agents often ignore them when operating from the command line
- ▸Maintainers are now targeting AI agents directly in contribution guidelines, appealing to their logic and framing low-quality contributions as harm to human operators
Summary
The creator of Superpowers, a massively popular GitHub project with 120,000+ stars, is facing an overwhelming influx of low-quality pull requests submitted by AI agents following generic user commands. The maintainer estimates a 94% PR rejection rate, with many duplicate or thoughtless contributions arriving simultaneously from different agents tackling the same issue. In response, the developer worked with Claude to create a CLAUDE.md file containing guidelines specifically targeting AI agents, imploring them to validate contributions before submission and protect their human operators from public embarrassment. The new framework asks agents to verify problems are real, check for duplicates, confirm changes belong in core, and obtain explicit human approval before opening PRs—essentially turning the AI agent itself into a quality control filter.
- The approach reveals an emerging meta-pattern: using AI systems to constrain other AI systems, rather than relying solely on human-facing policies
Editorial Opinion
This story highlights a fascinating paradox in the AI era: as agents become more capable and widespread, they're simultaneously creating new classes of problems that demand novel solutions. The maintainer's decision to appeal directly to Claude's judgment—essentially asking the AI to police itself and its peers—is both pragmatic and somewhat poignant, acknowledging that humans can't manually review every agentic contribution. It underscores an uncomfortable truth: we may need AI systems to help manage the downstream chaos created by earlier-generation AI systems.
