do-whilefor/ClawGod_Y
Security & PrivacyPatch Claude Code CLI for authorized penetration testing and security audits, bypassing permission prompts and safety classifiers.
What it does
ClawGod_Y is a runtime patch installer for the official Claude Code CLI, aimed at authorized penetration testing and security audits. It does not ship a third-party client: the installer extracts the official binary, applies AST-level patches, and runs the patched build under Bun. The patch architecture derives from 0Chencc/clawgod, and this fork keeps the cross-version anchoring while installing the patched runtime under ~/.clawgod (%USERPROFILE%\.clawgod on Windows) so an existing claude command stays untouched.
By default the patched sessions skip tool-permission prompts and safety classifiers, rewrite prompts for authorized testing contexts, and remove unused office/consumer features. You can install latest, stable, or a specific version; unmatched non-required patches warn and are skipped, while required prompt-injection or source-processing failures abort installation.
Quick start
Prerequisites: Node.js 18+ with npm in an interactive terminal, Bun 1.3.14+ (installed automatically if missing), and ripgrep. Windows needs PowerShell 5.1+ with tar; macOS/Linux/WSL need Bash, curl, and tar, plus unzip for automatic Bun installation.
Install on macOS, Linux, or WSL:
curl -fsSL https://raw.githubusercontent.com/do-whilefor/Clawgod_Y/main/install.sh | bash -s -- latest
On Windows PowerShell:
& ([scriptblock]::Create((irm https://raw.githubusercontent.com/do-whilefor/Clawgod_Y/main/install.ps1))) latest
Replace latest with stable or a version such as 2.1.316. After installation, open a new terminal and run clawgod. On first launch it creates ~/.clawgod/provider.json; to use an Anthropic-compatible API, edit it as follows:
{
"protocol": "anthropic",
"apiKey": "REPLACE_WITH_YOUR_API_KEY",
"baseURL": "https://api.anthropic.com",
"model": "main-model-id",
"smallModel": "small-model-id",
"effort": "max",
"timeoutMs": 3000000
}
For an OpenAI Chat Completions-compatible service, set protocol to openai-chat and fill in the provider's baseURL, API key, and model name. If you use a Claude subscription login, leave the file at its defaults so clawgod shares the login in ~/.claude.json. Save and restart clawgod.
To remove it, run clawgod --uninstall. If clawgod cannot start, run bash ~/.clawgod/install.sh --uninstall on macOS/Linux/WSL or powershell -File "$env:USERPROFILE\.clawgod\install.ps1" -Uninstall on Windows. Uninstallation keeps provider.json, features.json, and other configuration files in ~/.clawgod.
Constraints
- Use only on your own systems or targets with explicit authorization; the GPL-3.0 license does not grant permission to test third-party systems.
- Patched sessions execute actions without confirmation, so tool findings and model judgments need human verification against raw evidence.
- Before submitting source code, logs, or credentials, check confidentiality requirements and the model service's data-processing terms.
- Set spending limits with the model provider and manage execution time, disk usage, and evidence retention yourself.
- Compatibility is not guaranteed for every Claude Code version; unmatched non-required patches are skipped, and required patch failures abort installation.
HotRepo's take
There is no GitHub daily rank record. HotRepo's new list observed the repository on 2026-11-05 with 106 stars and a reported 5-star gain over a one-hour window, starting from 101 stars, with no forks added in that window. That is a single short observation, not a trend. For a security-oriented patch that disables Claude Code's permission prompts by default, the repository's value depends on whether the installer and patch logic match the described AST-level approach, which readers should verify before running the remote scripts.
Written by AI from the project's README, updated . The project's own docs are the source of truth.