Copilot Guardian is designed with security-first principles:

All logs and context sent to AI models are automatically sanitized:

• GitHub tokens (ghp_*, ghs_*, gho_*)
• Bearer tokens
• API keys
• Passwords in error messages

• Log collection, redaction, and artifact persistence happen locally.
• Copilot requests are sent to GitHub Copilot via authenticated GitHub sessions, using @github/copilot-sdk by default.
• Optional gh copilot CLI flows are supported for terminal-first reproducible runs.
• Full audit trail maintained in .copilot-guardian/ directory

• All raw inputs and outputs are saved (.raw.txt files)
• Users can inspect exactly what data was sent to AI models
• No hidden third-party telemetry or data collection

If you discover a security vulnerability in Copilot Guardian, please:

1. DO NOT open a public issue
2. Email: info@flamehaven.space
3. Include:
   • Description of the vulnerability
   • Steps to reproduce
   • Potential impact
   • Suggested fix (if any)

• 24 hours: Initial acknowledgment
• 7 days: Detailed assessment and action plan
• 30 days: Fix implementation and disclosure

# Ensure GitHub CLI is authenticated
gh auth status

# Use token with minimal required scopes
# Required: repo, workflow

# Never commit .env files
# Use GitHub Secrets for CI/CD environments

# Always review patches before applying
copilot-guardian fix --interactive

# Use Conservative mode for production
# Review the "Quality Verdict" before accepting

• Guardian inherits gh CLI's authentication model
• Ensure gh is up to date: gh version

• LLMs can hallucinate - always review patches
• Use the Anti-Slop quality checks
• Test patches in non-production environments first

• GitHub API rate limits apply
• Copilot API rate limits apply
• Guardian implements exponential backoff

Every Guardian run creates:

.copilot-guardian/
├── [timestamp]-context.raw.txt    # What was sent to AI
├── [timestamp]-analysis.json      # AI response (structured)
└── [timestamp]-patches.json       # Generated patches

This enables:

• Post-incident forensics
• Compliance audits
• Privacy verification

• GitHub Actions logs (for analysis)
• Repository metadata (via gh CLI)
• Source code context (when using MCP)

• User credentials
• Unrelated source code
• Personal information
• Telemetry or usage statistics

• All data is stored locally in .copilot-guardian/
• User controls retention (can delete directory)
• No cloud storage or external databases

• All processing is local
• No data transfer to third parties (except GitHub Copilot API via the user's authenticated SDK/CLI session)
• User has full control and right to erasure

• Compatible with GitHub Enterprise
• Works within corporate firewalls
• No external dependencies beyond GitHub APIs

Subscribe to security advisories:

gh repo subscribe flamehaven01/copilot-guardian --alerts

• Security Issues: info@flamehaven.space
• General Issues: GitHub Issues
• Documentation: docs/

Last Updated: 2026-02-13 Security Policy Version: 0.2.7