Skip to content

Security: CognicellAI/a2a-workbench

Security

SECURITY.md

Security Policy

Supported Versions

Security fixes are accepted for the latest commit on main until the project publishes versioned releases.

Reporting A Vulnerability

Please report suspected vulnerabilities privately to the project maintainers. If a public GitHub repository is configured, prefer GitHub's private vulnerability reporting flow.

Do not open a public issue containing credentials, exploit details, private endpoint URLs, or live tokens.

Deployment Warning

This project is intended as a local developer workbench. The server route can forward operator-supplied prompts and credentials to configured A2A endpoints.

Strict URL policy runs before discovery, authentication, each operation, and every redirect. Responses and individual SSE events are size bounded and validated before entering the domain model. Credential headers, OAuth tokens, cookies, and secret-shaped fields are recursively redacted from evidence.

Unsigned Agent Cards are permitted by the protocol. If a card contains signatures, the strict client requires at least one signature accepted by the configured trust store; the host trust store is responsible for rejecting expired or revoked keys.

Before hosting this app for other users, configure:

  • Authentication for the app itself.
  • A2A_UPSTREAM_ALLOWLIST for approved upstream hosts.
  • Appropriate A2A_UPSTREAM_TIMEOUT_MS and A2A_UPSTREAM_MAX_BYTES limits.
  • Network controls around private address ranges.

Keep A2A_ALLOW_PRIVATE_NETWORKS=false unless the instance is protected and intentionally scoped to private endpoint testing.

There aren't any published security advisories