Slopper: A GitHub Action to Filter AI-Generated Contributions

The term AI slop has been circulating among open source project maintainers for months now: pull requests with generic code, commit messages devoid of context, documentation that reads like a ChatGPT summary, and changes that don't match the repository's style. Anyone who has managed a moderately popular GitHub project over the past year knows what we're talking about. Now there's a concrete tool to address it.

Slopper is a GitHub Action released by developer `malvads` and discussed this week on Hacker News. Its goal is straightforward: automatically analyze incoming pull requests for signals indicating that code was generated by an LLM without proper care, and block or flag them before the maintainer has to review them manually.

What Slopper does exactly

Slopper integrates into any GitHub repository's CI/CD pipeline as just another workflow action. When a pull request arrives, the action inspects the diff, the PR message, and optionally the commit messages looking for patterns associated with slop: generic phrases like "this function improves performance", repetitive code structures with no adaptation to the project's context, or comments that seem generated to tick a box rather than to explain.

The outcome can be configured so the PR is blocked outright, gets a high-priority review label, or simply receives an automatic comment flagging the detected signals. This last option is the most respectful toward legitimate contributors who use AI assistants as support but review and adapt the code before submitting it.

The repository includes configuration examples and allows adjusting detection sensitivity, which matters because the core problem with any detector of this type is the false positive rate.

Why this problem matters now

The proliferation of code assistants based on LLMs, Claude Code included, has drastically lowered the barrier to contributing to open source projects. That has an obvious upside, but it has also driven up the volume of low-quality contributions sent with barely any human review. Some maintainers of popular projects have publicly reported that the percentage of PRs that simply don't fit, that ignore contribution guidelines, or that duplicate existing functionality has grown noticeably over the past year.

The problem isn't that someone uses an LLM to write code, but that they submit it without understanding or adapting it. That distinction matters, and Slopper tries to operate in that space: it doesn't penalize AI use, but rather lack of care in the contribution.

For projects with few maintainers and many incoming contributions, having an automatic filter at the gate can save hours of weekly triage. For large projects with review teams, it can serve as a pre-classification layer before assigning human reviewers.

Limitations worth keeping in mind

Any detection system based on text heuristics will have its limits. A developer who uses Claude Sonnet 4.6 to generate a patch, reviews it thoroughly, and adapts it to the project's style will produce code unlikely to trigger Slopper's alerts. Conversely, someone who writes in a very formal style or isn't a native English speaker might see their PR flagged unfairly.

The repository is still young and, at the time it appeared on Hacker News, had little traction in comments. There are no public data yet on accuracy rates in real environments. That doesn't invalidate it, but it does suggest deploying it first in observation mode before using it to automatically block PRs.

It's also worth asking whether the underlying problem, contributors who don't read the project guidelines, is better solved with technical tools or with explicit social friction: more demanding PR templates, mandatory checklists, or simply a well-written `CONTRIBUTING.md`.

---

From our perspective, this is a reasonable response to a real problem that many maintainers haven't yet voiced openly. The existence of the tool doesn't mean it's the definitive solution, but at least it puts the conversation on the table with something concrete to point to.