AgentGuard is an early-stage project. Security fixes are generally applied to the main branch and included in future releases.

At this stage, we recommend using the latest version from the default branch or the most recent tagged release, if available.

Please do not report security vulnerabilities through public GitHub issues, discussions, or pull requests.

Instead, use GitHub’s private vulnerability reporting feature in this repository’s Security tab:

https://github.com/glaborie/agentguard/security

This is the preferred channel for reporting vulnerabilities in AgentGuard.

When submitting a report, please include as much of the following as possible:

• a clear description of the vulnerability
• the affected component, service, or file(s)
• steps to reproduce
• proof of concept, if available
• impact assessment
• any suggested remediation

We will aim to:

• acknowledge receipt of a vulnerability report within a reasonable time
• investigate and validate the report
• assess severity and impact
• prepare and release a fix when appropriate
• coordinate disclosure responsibly

Response times may vary depending on project availability and issue complexity.

This policy applies to security issues in the AgentGuard codebase and maintained deployment/configuration assets.

Examples may include:

• authentication or authorization weaknesses
• remote code execution risks
• prompt injection protection bypasses
• sensitive data exposure
• secrets handling issues
• dependency-related vulnerabilities with practical impact
• insecure default configurations
• tenant or data isolation weaknesses, if introduced in the future

The following are generally out of scope unless they lead to a concrete, practical security impact in this repository:

• theoretical issues without a demonstrable exploit path
• best-practice suggestions without a specific vulnerability
• vulnerabilities only present in third-party services, unless caused by this project’s configuration or integration choices
• denial of service requiring unrealistic resources or privileged access in a local-only setup

Please allow time for investigation and remediation before public disclosure.

We ask reporters to avoid public disclosure until:

• the issue has been confirmed
• a fix or mitigation is available, when feasible
• affected users have had a reasonable opportunity to update

If you deploy AgentGuard, we recommend that you:

• keep dependencies and container images up to date
• avoid committing secrets into the repository
• rotate exposed credentials immediately
• review default passwords and change them before non-local deployments
• restrict network exposure of internal services
• use HTTPS and proper access controls in any shared or production environment
• treat prompts, traces, datasets, and model outputs as potentially sensitive data
• review third-party model, telemetry, and storage integrations for compliance requirements

AgentGuard handles observability data, traces, prompts, retrieval context, and model outputs. Depending on deployment and usage, these may contain sensitive or business-critical information.

Operators are responsible for:

• choosing an appropriate deployment environment
• controlling access to dashboards, APIs, and storage systems
• configuring retention, logging, and data handling in line with their security and compliance requirements