Security Policy

Reporting a Vulnerability

Please do NOT open a public issue for security vulnerabilities.

If you discover a security vulnerability in Edge Agents, please report it responsibly:

1. Use GitHub's private vulnerability reporting, or
2. Email us at root@foresthub.ai.

We take security reports seriously and will respond as soon as we can.

Supported Versions

Version	Supported
Latest release	Yes
Older versions	No

Security Update Process

• Security fixes are released as patch versions.
• Advisories are published via GitHub Security Advisories.

Scope

Edge Agents is a workflow runtime engine plus an LLM proxy. Examples of issues we consider security-relevant:

• Credential leakage — LLM provider keys (OpenAI / Anthropic / Gemini / …) or control-plane credentials exposed through logs, error messages, or serialization.
• Untrusted workflow config — a deployed workflow graph or expression that escapes its intended sandbox (file, network, or host access beyond a node's contract).
• Request forgery (SSRF) — e.g. via the web-fetch or other HTTP-calling nodes.
• Local builder server — the HTTP server fh-workflow open runs on localhost: file access beyond the served workflow's directory, or reads/writes triggered by another origin (CSRF, DNS rebinding). The engine itself serves no inbound HTTP; its only control-plane surface is outbound (FH_BACKEND_URL / ENGINE_SECRET), covered by the credential-leakage item above.
• Injection — in config parsing, expression evaluation, or request construction.
• Denial of service — unbounded resource consumption in the runner.

Issues in upstream LLM providers (OpenAI, Anthropic, Google, …) should be reported to those providers directly.