Coolify's maintainers were manually closing more than 120 low-quality, AI-generated pull requests every month. That volume is what drove peakoss to build Anti-Slop, an open-source GitHub Action that automates the triage. Coolify — a self-hostable PaaS alternative with over 50,000 GitHub stars — provided the test bed: peakoss reviewed more than 130 AI-generated PRs by hand before distilling the patterns into 31 configurable check rules. The action runs in under 15 seconds and closes flagged PRs automatically before they hit a maintainer's review queue.
The tool analyzes pull requests across multiple dimensions: branch names, titles, descriptions, PR template adherence, commit messages, file changes, user account signals, and contributor history. All behavior is controlled through 54 configuration options, though the defaults are designed to work without modification for most projects. Repository owners, members, and collaborators are automatically exempt from checks. The configurable max-failures threshold — defaulting to 4 — allows maintainers to tune sensitivity, with the idea that a single check failure should not automatically close a PR from a legitimate new contributor. The project positions itself explicitly as "anti-slop, not anti-AI," stating that high-quality AI-assisted contributions should pass through unaffected.
Several of Anti-Slop's default thresholds have drawn scrutiny for their potential to act as proxy signals against legitimate contributors rather than AI-generated ones specifically. The min-global-merge-ratio default of 30 percent — requiring that at least 30 percent of a contributor's historical PRs across GitHub have been merged — creates a structural disadvantage for early-career developers who have not yet accumulated a merge history. Combined with a minimum account age of 30 days, a minimum profile completeness score, and a cap of two emoji in PR descriptions, a new contributor with genuine code to offer but a sparse GitHub presence could accumulate enough check failures to trip the default threshold without any AI involvement. The tool's exemptions cover established project insiders but offer no special handling for first-time external contributors, the population most likely to fail multiple proxy-signal checks simultaneously.
The project is pre-v1.0.0 and the author has flagged the possibility of breaking changes before a stable release, recommending that users pin to a specific version tag or commit SHA. Anti-Slop treats AI-generated output as a signal quality problem to be filtered at the infrastructure layer rather than the human review layer. That framing is sound, but it only holds if the defaults are tuned to the project's context. Deployed without adjustment or public disclosure, the default configuration raises real questions about which contributors get filtered along with the slop. The source repository is available at github.com/peakoss/anti-slop.