docs

SlopGuard documentation

Install, slash commands, scoring, and how to change behavior with .github/SLOP_POLICY.yml.

config & behaviorfig.04
SLOP_POLICY.yml

Quick start

Install the GitHub App on a repo or org and you are done. Public repos are free and it works out of the box with no config. New PRs and issues get scored; at or above the threshold it adds the slop-quarantine label and a review comment with the reasons. It never auto-closes.

Install the GitHub App

Slash commands

Maintainers reply with a comment. Every action stays with a human.

/slop approve
Clears the quarantine and marks it a real contribution (slop-cleared).
/slop reject
Your explicit action: closes the PR as AI slop.
/slop false-positive
Opens a tuning issue and clears the quarantine.

Scoring and thresholds

Every contribution gets a score from 0 (clean) to 100 (almost certainly slop). By default 50+ is quarantined and 85+ is phrased more strongly. For how the score is built, see How it works.

SLOP_POLICY.yml configuration

Drop a .github/SLOP_POLICY.yml file in your repo to change behavior. Everything is optional; you only override what you set.

.github/SLOP_POLICY.ymlexample
# .github/SLOP_POLICY.yml  (all fields optional)
version: 1
enabled: true

scan:
  pull_requests: true
  issues: true

thresholds:
  quarantine: 50        # score at/above this gets a label + comment
  high_confidence: 85   # phrased more strongly (still never auto-closes)

allowlist:
  authors: [dependabot[bot], renovate[bot]]
  paths: ["docs/**", "**/*.md"]

heuristics:
  weight: 0.4           # blended with the LLM score (llm weight = 1 - this)

llm:
  enabled: true
  provider_order: [anthropic, grok, openai]

Key options

enabled
Master switch to pause SlopGuard without uninstalling.
scan
Whether to scan pull requests, issues, or both.
thresholds
The quarantine and high_confidence score cutoffs.
labels
Names for the quarantine, cleared, and high-confidence labels.
allowlist
Skip trusted authors or paths (globs) entirely.
heuristics
Rule-score weight (0-1). Tune to reduce false positives.
llm
Whether to use the LLM judge and the provider order.
comment_template
The PR comment format. Supports variables like {{score}}.
notify
Slack/Discord/webhook alerts on quarantine (Team plan).