If you discover a security vulnerability in AgentSentry, please report it responsibly.

Do NOT open a public GitHub issue for security vulnerabilities.

1. GitHub Security Advisories (preferred): Use GitHub's private vulnerability reporting to submit a report directly through GitHub.

2. Email: Send details to the repository maintainers via GitHub.

• Description of the vulnerability
• Steps to reproduce
• Potential impact
• Suggested fix (if any)

• Acknowledgment: Within 48 hours
• Initial assessment: Within 1 week
• Fix timeline: Depends on severity
  • Critical: Within 72 hours
  • High: Within 1 week
  • Medium/Low: Next release cycle

AgentSentry includes built-in security capabilities:

• Secret Detection: Scans for 15 types of hardcoded credentials (API keys, tokens, connection strings, JWTs)
• PII Scanner: Detects 15 categories of personally identifiable information in logging statements
• Dashboard Authentication: Token-based access control for the monitoring dashboard
• Hash-Chained Audit Log: Tamper-evident event storage with SHA-256 chain verification
• Permission Enforcement: File-level and command-level allowlist/denylist

As of v0.6.0, the MCP server requires authentication by default and rejects all requests unless AGENT_SENTRY_ACCESS_KEY is set.

• Production / network-exposed: set AGENT_SENTRY_ACCESS_KEY to a strong random value (e.g., openssl rand -hex 32).
• Local development only: set AGENT_SENTRY_NO_AUTH=true to disable authentication. This emits a stderr warning on every startup and should never be used in production.
• The deprecated AGENT_SENTRY_REQUIRE_AUTH variable has been removed.

• The Supabase provider is experimental and should not be used in production environments.
• Vector search data is stored unencrypted in local SQLite. Use filesystem-level encryption for sensitive environments.