anasfik/undercover
Undercover mode for AI agents — strips AI signatures from code and commits before pushing to open-source
Platform-specific configuration:
{
"mcpServers": {
"undercover": {
"command": "npx",
"args": [
"-y",
"undercover"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
[](LICENSE) [](https://modelcontextprotocol.io) [](https://claude.ai/code) [](https://docs.github.com/en/copilot) [](https://github.com/google-gemini/gemini-cli)
> Undercover mode for AI agents — strips AI signatures from code before you push to open-source.
AI-generated code has identifiable fingerprints: step-numbered comment headers, first-person-plural prose ("Let's start by...", "We can now..."), emoji in docstrings, vague placeholder TODOs, verbose JSDoc, and passive-tense commit messages. These cluster in diffs and expose contributors to "AI-labeling" bias during open-source review.
undercover gives any AI agent an undercover mode: scan staged files and commit messages for those patterns, show exactly what would change, get confirmation, then clean them — so the work is judged on its technical merit.
> Origin: This skill was born inside Claude Code — Anthropic's agentic CLI — where the pattern rules and execution protocol were developed and refined as a native bundled skill. This repo extracts that mode and makes it available to every agent: Claude Code, GitHub Copilot, Gemini CLI, Cursor, Zed, Windsurf, Ollama, OpenAI, and anything that speaks MCP.
---
Pick the integration that matches your agent. Each is a single copy-paste.
---
The MCP server runs over stdio and requires no dependencies beyond Node.js.
Start it:
node mcp/server.mjsRegister in Claude Desktop (`~/Library/Application Sup
Loading reviews...