SlopGuard
Pricing
- Free Tier
- Public repositories free
Summary
Every week you spend thirty seconds on a PR that turns out to be pure LLM output — polished title, confident summary, zero grasp of your codebase — and you cannot just close it on suspicion without burning the one real first-timer in the batch. SlopGuard scores every incoming PR and issue against heuristics, labels the likely machine-made ones with the specific reasons, and leaves the close button entirely in your hands.
The tool installs as a GitHub App with no Action YAML, no CI config, and no secrets to wire. Each contribution gets a 0–100 slop score derived from heuristics only — no LLM API calls — and at or above your configured threshold it adds a quarantine label plus a review comment listing the exact signals, such as leaked chat-assistant phrases or prompt fingerprints. Below the threshold it stays silent. You reply with slash commands to approve, reject, or flag a false positive. The vendor states the golden-set benchmark sits at 100% precision and 92% recall — every flagged item was real slop, and the single miss was slop that slipped through, not a genuine contributor wrongly quarantined.
Bottom line: Pick this if you are an open-source maintainer drowning in AI-generated PRs and need triage help that will never auto-close a real contributor; skip it if your team needs programmatic API access, self-hosted deployment outside the Commons Clause terms, or detection that adapts beyond static heuristics as generation patterns shift.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Heuristics-only scoring with no external LLM calls, so detection runs without API keys, per-call costs, or a third-party model availability dependency — the queue keeps moving even when OpenAI is down.
- 100% precision on the vendor's labelled golden set, meaning every contribution it flags is real slop and no genuine first-time contributor gets a quarantine label by mistake — the risk you take by not using it is missed slop, not burned contributors.
- Per-repository threshold configuration via a slider, so a high-traffic org repo and a small side project can run at different sensitivity levels without separate installs or config files.
- Provenance trail attached to each flagged item — leaked phrases, prompt fingerprints, and the specific signals — so when you review a quarantined PR you are not just seeing a score, you are seeing exactly why it was flagged.
- One-click GitHub App install with no Action YAML or secrets to wire, so a maintainer can have it running on a new repo in under a minute without touching CI configuration.
Cons
Sign in to edit- Detection is bounded by a static heuristic ruleset, so when LLM output patterns shift — shorter prompts, less boilerplate, better title generation — recall degrades silently until someone updates the rules manually. Teams processing high volumes of slop that evades the current heuristics have no model-retraining path and no feedback loop beyond the slash commands; at that point they evaluate classifier-backed alternatives.
- There is no API, which means a team that wants to pull slop scores into a separate dashboard, feed them into a Slack alert, or trigger any downstream automation has no supported integration path. The label-and-comment output is the only interface. Teams that need scores as data rather than GitHub UI annotations will be screen-scraping labels or abandoning the tool for a solution with a query endpoint.
- Self-hosting is gated behind Commons Clause terms, which permits personal use but blocks commercial redistribution. An organization that wants to run SlopGuard on internal infrastructure for a commercial product and control the full deployment will hit a licensing wall and need either a separate commercial agreement with the vendor or a different tool.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- GitHub
- API Available
- No
- Self-Hosted
- No
- Last Updated
- 2026-06-11T08:23:59.823Z
Best For
Who it's for
- Open-source maintainers receiving high volumes of AI PRs
- Teams wanting provenance tracking without auto-closing
- Repositories that prefer heuristics over LLM calls
What it does well
- Triage incoming PRs and issues for AI-generated content
- Quarantine suspected slop while preserving human review
- Tag provenance and leaked phrases on machine-generated contributions
- Configure thresholds and policies per repository
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Compare SlopGuard
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is SlopGuard free?
- SlopGuard is a paid tool. No permanent free tier is offered.
- Is SlopGuard open source?
- No — SlopGuard is a closed-source tool. Source code is not publicly available.
- What platforms does SlopGuard support?
- SlopGuard is available on: GitHub.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
SlopGuard sits between your GitHub inbox and your review queue. When a PR or issue arrives, the app scores it against a heuristic ruleset — looking for leaked chat-assistant boilerplate, prompt fingerprints, generic auto-generated titles, and empty descriptions — and returns a 0–100 slop score with the specific signals attached. At or above the threshold you configure, it labels the contribution and leaves a review comment. Below it, nothing happens. You respond via slash commands: `/slop approve`, `/slop reject`, or `/slop false-positive`. Nothing is ever auto-closed.
The defining design decision is that it runs entirely on heuristics, with no LLM key required. That removes per-call API costs and avoids the latency and availability risk of routing every PR through an external model. The tradeoff is that the detection ceiling is fixed to the heuristic set — the system does not learn from your corrections in the way a model-backed classifier would, and as generation patterns evolve the recall figure the vendor reports against its 25-case golden set will need manual ruleset updates to hold.
Public repositories get the GitHub App install at no cost, according to the vendor’s site. Self-hosting is source-available under Commons Clause, which permits personal use but restricts commercial redistribution. There is no API, so any workflow that requires reading SlopGuard’s scores programmatically — to feed a dashboard, trigger a downstream automation, or integrate with a separate triage system — has no supported path. Teams that outgrow label-and-comment-only output and want to route flagged PRs into a wider workflow will be building around the tool, not with it.
Installation is one click from the GitHub App directory with no secrets, no Action YAML, and no CI configuration needed. The threshold slider is adjustable per repository, and the vendor’s live demo shows PRs reclassifying between quarantined and passed in real time as the line moves — which means you can tune aggressiveness per project without touching config files.