|
| 1 | +# Security Policy |
| 2 | + |
| 3 | +## Overview |
| 4 | + |
| 5 | +HolyClaude runs AI coding agents inside a Docker container with elevated capabilities. This document explains the security model, what the container can access, and how to report vulnerabilities. |
| 6 | + |
| 7 | +## Container Capabilities |
| 8 | + |
| 9 | +HolyClaude requires the following Docker capabilities: |
| 10 | + |
| 11 | +| Capability | Why | Risk | |
| 12 | +|-----------|-----|------| |
| 13 | +| `SYS_ADMIN` | Chromium sandboxing (Linux namespaces) | Standard for any Chromium-in-Docker setup | |
| 14 | +| `SYS_PTRACE` | Debugging tools (strace, lsof) | Allows process inspection within the container | |
| 15 | +| `seccomp=unconfined` | Chromium syscall requirements | Removes syscall filtering for the container | |
| 16 | + |
| 17 | +These are required for Chromium to function and are standard across Playwright, Puppeteer, and CI/CD browser testing setups. They do **not** grant the container access to the host system beyond what Docker normally allows. |
| 18 | + |
| 19 | +## Permission Modes |
| 20 | + |
| 21 | +| Mode | Default? | What it means | |
| 22 | +|------|----------|--------------| |
| 23 | +| `allowEdits` | **Yes** | Claude can edit files freely, asks before running shell commands | |
| 24 | +| `bypassPermissions` | No | Claude runs any command without confirmation | |
| 25 | + |
| 26 | +The default `allowEdits` mode is safe for most users. `bypassPermissions` is documented for power users who understand the implications. |
| 27 | + |
| 28 | +## Credential Storage |
| 29 | + |
| 30 | +- API keys and authentication tokens are stored in `./data/claude/` on the host (bind-mounted to `~/.claude/` in the container) |
| 31 | +- Credentials never leave the container — HolyClaude does not proxy, intercept, or transmit credentials to any third party |
| 32 | +- The container communicates directly with AI provider APIs (Anthropic, Google, OpenAI) using your credentials |
| 33 | + |
| 34 | +## Network Access |
| 35 | + |
| 36 | +The container has unrestricted outbound network access. This is required for: |
| 37 | +- AI provider API calls (Anthropic, Google, OpenAI) |
| 38 | +- npm/pip package installations |
| 39 | +- Git operations (clone, push, pull) |
| 40 | +- Any web requests Claude Code makes during development tasks |
| 41 | + |
| 42 | +## Reporting a Vulnerability |
| 43 | + |
| 44 | +If you discover a security vulnerability in HolyClaude: |
| 45 | + |
| 46 | +1. **Do not** open a public GitHub issue |
| 47 | +2. Use [GitHub Security Advisories](https://github.com/CoderLuii/HolyClaude/security/advisories/new) to report privately |
| 48 | +3. Include: description, steps to reproduce, and potential impact |
| 49 | +4. You will receive a response within 48 hours |
| 50 | + |
| 51 | +## Supported Versions |
| 52 | + |
| 53 | +| Version | Supported | |
| 54 | +|---------|-----------| |
| 55 | +| latest | Yes | |
| 56 | +| < 1.0.0 | No | |
0 commit comments