网有NETHAVE
AGENT SKILL

no-slop-pr

Scores GitHub pull requests for AI slop and blocks the ones an agent wrote, in the spirit of No Sloptober. Use when the user runs /no-slop-pr, or asks to check, gate, block, flag or reject AI generated, LLM written, agent authored or sloppy pull requests.

License 许可
Unlicense
最近更新
2026-11-07
Stars
★ 21
Forks
9

01 前言 Front Matter

name
no-slop-pr
description
Scores GitHub pull requests for AI slop and blocks the ones an agent wrote, in the spirit of No Sloptober. Use when the user runs /no-slop-pr, or asks to check, gate, block, flag or reject AI generated, LLM written, agent authored or sloppy pull requests.
allowed-tools
[ "Bash", "Read", "AskUserQuestion" ]

02 资源文件 15

03 SKILL.md 正文

No Slop PR

No Sloptober asks for one month without LLM tools. This skill guards the other side of the door: pull requests written by an agent do not get merged.

Inputs

  • A PR number or URL: python3 .claude/skills/no-slop-pr/slop_check.py 42
  • A PR from another repo: python3 .claude/skills/no-slop-pr/slop_check.py 42 --repo owner/name
  • All open PRs: loop over gh pr list --state open --json number --jq '.[].number'
  • Local files, no GitHub: python3 .claude/skills/no-slop-pr/slop_check.py --text body.txt --diff change.diff

Add --json for machine output. The script exits 1 on BLOCK and 0 otherwise, so it can gate CI.

Signals

Every signal, weight and threshold lives in rules.json. Read it, never guess.

Signal Why it matters
AI co-author trailer The commit itself admits an agent wrote it
Generated-with footer Same admission, in the PR body
LLM tell phrases delve, robust, seamless, comprehensive, this PR introduces
Emoji bullets and headings Checkmark and rocket bullet lists are an agent template
Em dash storm Humans rarely type the long dash, models love it
Boilerplate section headings Summary, Key Changes, Test Plan scaffolding
Comment cruft in added code Onarheim: agents add comments that explain nothing

Score is capped at 100. BLOCK at block_at, SUSPECT at suspect_at, PASS below.

Steps

  1. Run the script for every PR the user named.
  2. Print one line per PR: number, verdict, score, the top reasons.
  3. For SUSPECT, quote the lines that matched so a human decides. Never block on SUSPECT alone.
  4. For BLOCK, ask the user before touching GitHub. Only after a yes run gh pr review <n> --request-changes --body "<reasons>". Never close, merge or push.
  5. The review body lists the reasons from the script and one line: No Sloptober: please rewrite this PR yourself.

Rules

  • The script is a filter, not a judge. Say so when reporting.
  • Never write the review body with emojis, em dashes or any phrase from rules.json.
  • Run the tests after changing rules.json: python3 -m unittest discover -s .claude/skills/no-slop-pr/tests.

04 同仓库更多 Skills

diegopacheco
diegopacheco/ai-playground

word-count

Counts the words, lines and characters of a text or a file.

文档处理
★ 21
diegopacheco
diegopacheco/ai-playground

slugify

Turns a title or a sentence into a URL slug, lowercase words joined by hyphens with accents removed.

★ 21
diegopacheco
diegopacheco/ai-playground

json-format

Validates a JSON file and pretty prints it with two space indentation, or reports where the JSON is invalid.

★ 21
diegopacheco
diegopacheco/ai-playground

commit-message

Writes a conventional commit message for the staged git changes.

★ 21
diegopacheco
diegopacheco/ai-playground

skills-gateway

Gateway and router to a catalog of skills that are not registered directly. Use it before any task a specialized skill could handle, such as text, JSON or git chores. It finds the right skill and loads its instructions.

★ 21
diegopacheco
diegopacheco/ai-playground

decision-log

Record, list, show and supersede architecture decisions that persist across sessions in the shared agent infra state store. Use when the user says "record this decision", "log an ADR", "what did we decide about", "list decisions", or "supersede decision".

★ 21