Skip to content

Security: LCV-Ideas-Software/deepseek-cli

SECURITY.md

Security Policy

Reporting a Vulnerability

If you discover a security vulnerability in this repository, please do not open a public issue. Report it privately to the repository maintainer.

Contact: alert@lcvmail.com

Please include:

  • Description of the vulnerability
  • Steps to reproduce
  • Potential impact
  • Suggested fix, if available

We will acknowledge your report within 24 hours and work to resolve it promptly.

Supported Versions

Version Supported
Latest yes
Previous releases security updates only

Threat Model

deepseek-cli is designed for a single-user trusted workstation. It can read files, write files after confirmation, execute shell commands after confirmation, and connect to configured MCP servers.

Important boundaries:

  • Do not expose this CLI as a network service without an authenticating wrapper.
  • Do not run it in a directory that contains secrets you do not want the model to read.
  • Do not commit .deepseek/, .env, API keys, MCP bearer tokens, or node_modules/.
  • MCP server credentials are process environment data. Treat .deepseek/settings.json as sensitive when it contains literal headers; prefer ${VAR} placeholders.

Hardening

  • HTTP/SSE/Streamable HTTP MCP connections use official MCP SDK transports.
  • xAI streaming tool calls are emitted only after JSON arguments are complete.
  • Tool outputs continue the Responses API conversation as function_call_output.
  • Local user settings writes are atomic.
  • Top-level crash handlers redact common secret-bearing strings.

Dependency and Supply Chain

  • CI runs npm ci and npm test on main and pull requests.
  • Dependabot monitors npm and GitHub Actions daily.
  • Release publishing uses npm Trusted Publishing with provenance; npmjs.com publication does not require an NPM_TOKEN.

There aren't any published security advisories