| Version | Supported |
|---|---|
| 1.0.x | ✅ Yes |
We take security seriously! 🐝 If you discover a security vulnerability in this project, please report it responsibly.
- DO NOT open a public GitHub issue for security vulnerabilities
- Instead, email us at: security@dubsopenhub.com
- Or use GitHub's private vulnerability reporting
Please provide as much of the following as possible:
- 📝 Description of the vulnerability
- 🔄 Steps to reproduce
- 💥 Potential impact
- 💡 Suggested fix (if you have one)
- ⏱️ Acknowledgment within 48 hours
- 🔍 Assessment within 1 week
- 🛠️ Fix or mitigation as quickly as possible
- 🎉 Credit in the release notes (unless you prefer anonymity)
This repository has the following GitHub security features configured:
| Feature | Status | Notes |
|---|---|---|
| ✅ Dependabot Alerts | Enabled | Monitors dependencies for known vulnerabilities |
| ✅ Dependabot Security Updates | Enabled | Auto-creates PRs to fix vulnerable dependencies |
| ✅ Secret Scanning | Enabled | Detects accidentally committed secrets |
| ✅ Secret Scanning Push Protection | Enabled | Blocks pushes containing secrets |
| ✅ Code Scanning (CodeQL) | Available | Static analysis for security bugs |
Since this is a Copilot CLI skill (no runtime code, only markdown instructions), the primary security considerations are:
- 🔑 No secrets in skill files — SKILL.md and agent.md should never contain API keys, tokens, or credentials
- 📜 Safe instructions — Skill instructions should never instruct the agent to bypass security controls
- 🔍 Dependency awareness — If dependencies are added in the future, keep them updated
Since this skill orchestrates hundreds of AI agents and processes user-provided task descriptions, prompt injection is a relevant concern:
- 🔒 Depth Guard — 3-layer enforcement prevents runaway spawning: prompt-level, contract-level, and config-level
- 🧹 Input sanitization — Task descriptions are compressed through 4 layers of context reduction (128K → 128 tokens), stripping potential injection payloads
- 🚫 No credential passthrough — User input is used as task descriptions only; it is never interpolated into system-level commands or used to access external services
- ⚖️ Consensus scoring — Even if one agent is influenced by injected content, the median-of-3 consensus mechanism and cross-family review limits the impact on final scores
- 👻 Shadow scoring — Hidden criteria that agents never see provide an independent quality audit, catching outputs that look good but contain errors
This project is licensed under the MIT License.