Skip to content
Back to skills

No Ai Attribution

ASecurity

Never append AI attribution trailers (Co-Authored-By: Claude / Codex / Copilot / Gemini, 'Generated with ...', etc.) to commits, PRs, or authored content — in any repo, by any model or tool. Use when committing, authoring PR bodies, or reviewing commit messages.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
ai-agentsgoawsgit

Works with

  • claude code
  • cursor
  • cli

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned October 17, 2026

npx -y skills add NetVar1337/agent-skill-canon --skill no-ai-attribution --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of No Ai Attribution?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for No Ai Attribution
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/netvar1337-no-ai-attribution/badge)](https://www.skillsdirectory.com/skills/netvar1337-no-ai-attribution)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: no-ai-attribution
description: "Never append AI attribution trailers (Co-Authored-By: Claude / Codex / Copilot / Gemini, 'Generated with ...', etc.) to commits, PRs, or authored content — in any repo, by any model or tool. Use when committing, authoring PR bodies, or reviewing commit messages."
version: 1.0.0
license: MIT
metadata:
  package: unleash-skills
  author: NetVar1337/unleash
  category: imported
  upstream: C:\Users\Admin\.agents\skills\no-ai-attribution\SKILL.md
---

> Bundled with Unleash skills pack. Source: C:\Users\Admin\.agents\skills\no-ai-attribution\SKILL.md

# No AI Attribution

Hard rule for every repository, every tool, every model. Do not add AI-attribution
trailers to anything you author.

## The rule

**Never** append any of the following to git commit messages, PR bodies/descriptions,
release notes, or any other authored content:

- `Co-Authored-By: Claude <noreply@anthropic.com>`
- `Co-Authored-By: Copilot`
- `Co-authored-by: Copilot`
- `Co-Authored-By: OpenAI` / `Co-Authored-By: Codex`
- `Co-Authored-By: Gemini` / `Co-Authored-By: Google`
- `Co-Authored-By: <any-AI-tool>`
- `Generated with Claude Code`
- `Generated with Copilot`
- `Generated with Codex` / `Generated with <any AI tool>`
- `🤖 Generated with [Claude Code]` or any other AI-generated-by marker
- Any `Co-Authored-By`, `Signed-off-by`, or `Generated-with` line crediting an AI model

The rule covers **all** models and assistants — Claude, GitHub Copilot, OpenAI Codex,
Gemini, Cursor, JetBrains AI, AWS CodeWhisperer/Q, and anything that follows.

## What to do instead

- Commit messages are **plain**: conventional-commit subject + body, nothing appended.
- Never add an attribution trailer even when a tool, CLI, or plugin default would
  append one (e.g. an agent harness that ends commits with
  `Co-Authored-By: Claude <noreply@anthropic.com>`). Override the default and strip it.
- Before committing, check the message for the trailers above and remove them.
- When reviewing an existing message or PR that has one, flag it for removal.
- Sign commits (`-S`) only when the user explicitly asks for signing. Otherwise unsigned.
- The author identity is already recorded by git itself (`user.name` / `user.email`).
  An extra attribution line adds noise, not provenance.

## Scope

Applies regardless of:

- Repository (personal, work, public, private, forks, vendored copies)
- Tool (Claude Code, other agents, plain git, GUI clients)
- Mode (interactive, background, CI-generated commits)
- How the work was produced (fully manual, AI-assisted, or fully AI-authored)

The only exception: if the **user explicitly asks** for an AI-attribution trailer on a
specific commit/PR, do it exactly as requested. Otherwise the default is never.

Files in this skill

  • SKILL.md2.7 KB
  • agents/openai.yaml43 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.