PRKitchen is the quality gate that catches AI slop, enforces your standards, and protects your main branch — automatically, before the merge.
Vague prompt. No context. Paste straight into a PR. It compiles. It doesn't belong. Major open-source projects are adding explicit AI Slop bans. Repeat offenders are being removed. Maintainers are exhausted. One bad batch can permanently damage a contributor's reputation.
Structured prompt, clear context, human review before the PR opens — that code doesn't just pass PRKitchen. It sails through. PRKitchen is the gate that tells the difference.
AI assistants exploded how fast code gets written. Everything downstream — review, merge, release — barely moved. The orders pile up faster than the pass can plate them.
The study pegs the elasticity of substitution between AI output and human effort at just 0.25 — AI and human reviewers are strong complements, not substitutes. More generated code doesn't mean less review. It means more.PRKitchen widens the bottleneck: it sends the slop back before it ever reaches a human, so the review you do still counts.
It doesn't stop at the merge. Across four major app marketplaces, the same study found a moderate rise in new apps but no increase in total usage. More software shipped — it just didn't reach more people.
Figures from a 2026 NBER study, “Writing Code vs. Shipping Code” (Demirer, Musolff & Yang, NBER w35275) — interactive coding-agent effects across the production chain, measured over 100,000+ developers.
All dishes included. Served fresh on every PR.
Dual-layer heuristic + GPT-4o-mini scoring. Set your own thresholds per repo.
Catches PRs in the right repo but chasing work you never asked for. You state the project's purpose; off-mission PRs go back to the kitchen.
Pass, draft, or close — PRKitchen acts. It doesn't leave a comment and hope.
Plain YAML in your repo or point-and-click in the dashboard. No plugins.
Fourteen check types in one declarative config. Toggle each one independently.
Every repo gets its own recipe. One size fits none.
Import proven recipes from other teams. Rate, fork, and customize.
This is all it takes to run a full kitchen.
recipes:
- id: "ai-slop-filter"
name: "🔍 AI Slop Detection"
checks:
- type: "slop_score"
threshold_medium: 70 # Flag for review
threshold_high: 100 # Return to sender
actions:
low: "pass" # 🟢 Clean plate, serve it
medium: "draft" # 🟡 Needs seasoning
high: "close" # 🔴 Back to the kitchen
- id: "wild-goose-guard"
name: "🪿 Wild Goose Guard"
checks:
- type: "wild_goose" # Right repo, wrong intent
project_purpose: |
A pre-PR gatekeeper that filters AI slop.
Out of scope: unrequested features, scope creep.
actions:
off_mission: "draft" # Chasing a wild goose? Back to the kitchen
- id: "small-plates"
name: "🍱 PR Size Limits"
checks:
- type: "size_limit"
max_files: 10
max_lines_added: 400 # A meal, not a banquet
actions:
exceed: "draft"
- id: "mise-en-place"
name: "📋 Linked Issue Required"
checks:
- type: "linked_issue"
actions:
missing: "draft" # No ticket, no tableNo plugins. No scripting. Plain YAML or use the dashboard UI. 🧑💻
Leaves a comment. Hopes someone reads it. Repeats.
Same rules for every repo, every team, every project. Reheat and serve.
Tells you about the code you're writing, not whether the PR should merge.
Relies entirely on developers reading, remembering, and caring. Under deadline pressure: skipped.
Zero lines of code to write. Six checks in one config. Three actions that enforce. PRKitchen is the pre-merge quality gate that actually sends bad PRs back to the kitchen.
of code to write
in one config
that enforce
Start free with 1 repo. No credit card required.