Policy Gates
Core rules that PRs must satisfy before reaching reviewers
Require Linked Issue
PRs must reference a GitHub issue (e.g., "Closes #123")
Require Tests for Code Changes
PRs that modify source code must include corresponding test files
Require PoC for Security Reports
Security-labeled issues must include a proof of concept or reproduction steps
Allow Refactors Without Tests
Skip test requirement for PRs labeled "refactor" with no behavior changes
Intent & Risk
Checks that read what a PR says it does, and what it could break
PR Coherence
AI compares the description with the diff and rates it clear, thin, mismatched or slop
Wild Goose Guard
Flags PRs that drift from the project purpose you set. Stays off until a purpose is written
Blast Radius for Infrastructure
Terraform changes to prod, IAM, networking or shared modules must include an impact statement
Spice Rack Auto-Approve
Approves low-risk PRs when every other check passed. It only approves and never blocks
Only Hold Bots & Agents to This Recipe
Run a recipe only for authors matching patterns such as
*[bot] or copilot*
Size Limits
Prevent oversized PRs from consuming reviewer bandwidth
AI Detection
Configure sensitivity for detecting AI-generated code submissions
Slop Sensitivity
How aggressively should PRKitchen flag potential AI-generated content?
Low
Permissive
Only flag obvious AI-generated submissions (score > 80). Best for trusted teams with established contributors.
Medium
Balanced
Flag suspicious patterns and AI signals (score > 50). Good balance of signal and noise for most teams.
High
Strict
Strict detection, flag any AI-like patterns (score > 30). Recommended for open-source repos with high external contribution volume.
Additional Settings
Draft high-confidence AI submissions
Convert PRs with a slop score above 90 to draft. Drafted PRs appear in Blocked Items for review.
Require human justification for flagged PRs
Authors of flagged PRs must provide a written explanation before the PR can proceed.