Mingyi Atlas includes tooling for authorized security assessment. Security reports and security-tool contributions must be handled carefully.

Do not open a public issue for vulnerabilities that could put users at risk.

Report security issues privately to the maintainers. If no private channel is listed for your deployment or fork, create a minimal public issue asking for a security contact without including exploit details, secrets, target data, logs, or proof-of-concept payloads.

Please include:

• Affected version or commit.
• A concise description of the issue.
• Impact and affected component.
• Reproduction steps using a local or synthetic target when possible.
• Any logs or artifacts with secrets removed.

Security fixes target the latest published version and the current main branch unless a maintainer announces otherwise.

Pentest mode is for authorized testing only. Contributions must preserve these boundaries:

• No destructive actions by default.
• No brute force, credential theft, credential dumping, persistence, malware behavior, denial of service, or unauthorized exploitation.
• No hash cracking, key recovery, or wordlist attacks in built-in tools.
• No SSRF probes against metadata services, localhost, or internal addresses unless explicitly authorized and represented in scope.
• No active request-smuggling payloads in default tooling. Passive assessment is acceptable.
• Network tools must be scope-gated, timeout-limited, and output-bounded.
• Browser/container tools must capture artifacts without leaking private data into the repository.

Never commit:

• API keys, OAuth tokens, session cookies, private keys, or passwords.
• Real customer data, target data, or pentest artifacts.
• Screenshots, browser traces, or container outputs from private systems.
• .mingyi-atlas/ runtime state unless a fixture is intentionally sanitized and documented.

When updating dependencies:

• Prefer minimal updates with a clear reason.
• Run the relevant tests and pnpm check.
• Note security implications in the PR description when the update affects auth, MCP, browser/container execution, permissions, or pentest tooling.