SlopGuard
a GitHub App that quarantines AI slop PRs and issues
Last verified:
What is SlopGuard?
SlopGuard is a GitHub App designed to help maintainers detect and triage AI-generated low-effort pull requests and issues, commonly called 'AI slop.' AI slop refers to machine-generated content pushed to a repository with little or no human review—PRs and issues created by LLMs that appear plausible but miss context, add nothing valuable, and mostly exist to farm activity or contribution stats.
The tool scores every contribution from 0 to 100 based on a slop score, provides the reasons behind the score, and includes a provenance trail showing generator hints, prompt fingerprints, and leaked assistant phrases like 'As an AI model.' At or above a user-set threshold, SlopGuard adds a 'quarantine' label and a review comment. Below the threshold, it stays silent. The threshold is adjustable via a slider, and PRs reclassify between quarantined and passed in real time.
SlopGuard is built specifically for maintainers of open-source and private repositories who struggle with fake PRs eating their time and endless triage notifications. It respects contributors by never auto-closing anything—a human always makes the final call. The app installs via one-click GitHub App installation (gh app install), requires no Action YAML, no CI config, and no secrets to wire. It works in heuristics-only mode with zero API keys and still achieves 100% precision on a golden set of 25 cases.
SlopGuard pricing
Pricing model: Freemium
A free tier is available. Paid plans are Pro ($19/month), Team ($99/month), and Enterprise ($299/month). Payments are processed by Polar.
SlopGuard pros
- One-click GitHub App installation with no CI config needed
- Adjustable slop-score threshold slider for real-time reclassification
- 0 to 100 slop score with detailed reasons for every contribution
- Provenance tagging showing generator hints and prompt fingerprints
- Never auto-closes PRs—human always makes final decision
- Works in heuristics-only mode with zero LLM API keys required
- Achieves 100% precision on golden set of 25 labeled cases
- 92% recall catching 12 of 13 real slop cases heuristics-only
- F1 score of 96% combining precision and recall
- Policy-as-code with thresholds, labels, allowlists in SLOP_POLICY.yml
- No database required—state lives in GitHub labels and issues
- Review comment templates customizable in repo configuration
- Self-host option available under Commons Clause license
- Flags leaked assistant phrases like 'As an AI model'
- Zero false alarms on golden set validation
- Respects first-time contributors by not closing on pattern alone
- Single-pass review of real contributions after filtering slop
SlopGuard cons
- Only labels and comments—does not block merges automatically
- Heuristics-only mode may miss some sophisticated AI slop
- Requires maintainers to manually review quarantined PRs
- No behavioral pattern analysis of binaries
- Golden set validation limited to only 25 cases
- Self-host source available but requires manual setup
- No built-in notification system for quarantined items
- Policy file must be reviewed like any other code change
- Threshold slider is manual—no automatic learning
- Only supports GitHub repositories, not other platforms
Frequently asked questions about SlopGuard
What is AI slop?
AI slop is low-effort, machine-generated content pushed to a repo with little or no human review: PRs and issues spun up by an LLM that look plausible but miss the context, add nothing, and mostly exist to farm activity or contribution stats.
Does SlopGuard auto-close PRs?
No. SlopGuard never auto-closes anything. It only adds a quarantine label and a review comment at or above the threshold. A human always makes the final call on whether to close a PR.
How does the slop score work?
Every contribution gets a 0 to 100 slop score with the reasons behind it and a provenance trail. At or above your set threshold, SlopGuard adds a label and review comment. Below it, the contribution stays silent.
Can I adjust the threshold?
Yes. You drag a slider to move the slop-score line, and PRs reclassify between quarantined and passed in real time. You set the line and a human always has the final word.
Does SlopGuard require an LLM API key?
No. SlopGuard works in heuristics-only mode with zero API keys and still hits 100% precision on the golden set of 25 cases.
How do I install SlopGuard?
Install via one-click GitHub App installation using 'gh app install'. No Action YAML, no CI config, and no secrets to wire. You can install on a single repo or an entire organization.
What is provenance tagging?
Provenance tagging flags generator hints, a prompt fingerprint, and leaked assistant phrases like 'As an AI model' to show why a contribution was flagged as AI slop.
Where does SlopGuard store its state?
SlopGuard has no database. State lives in GitHub labels and issues. You can also self-host the entire thing for your own use.
What is the accuracy of SlopGuard?
Against a hand-labelled golden set of 25 cases (13 real slop, 12 legit contributions) at default threshold 50: 100% precision (zero false alarms), 92% recall (caught 12 of 13 slop cases), and 96% F1 score.
What license is SlopGuard under?
SlopGuard is MIT licensed. The source is available under the Commons Clause for self-hosting for your own use.