We take security seriously. If you discover a security vulnerability, please report it responsibly.

Do not open a public issue. Instead, please:

1. Email the maintainer directly (see package.json for contact info)
2. Include a description of the vulnerability
3. Include steps to reproduce if possible
4. Include potential impact

You should expect a response within 48 hours. We will:

• Confirm receipt of your report
• Investigate and verify the vulnerability
• Develop and test a fix
• Release a patch
• Credit you in the release notes (unless you prefer to remain anonymous)

smol-agent has the ability to execute shell commands via the run_command tool. This is a core feature but comes with inherent risks:

• Only run smol-agent in environments you trust
• Review the agent's proposed changes before confirming destructive operations
• Be cautious when processing untrusted input

The agent can read and write files within your project directory. Path traversal protections are in place, but:

• Run smol-agent in a dedicated project directory
• Don't run as root/administrator
• Review file operations before confirming

If you use the web_search and web_fetch tools:

• These require an Ollama API key
• Data is sent to Ollama's servers for processing
• Review Ollama's privacy policy at https://ollama.com/privacy

1. Review before executing: The agent will show you what it plans to do
2. Use version control: Commit your changes before running smol-agent
3. Limit scope: Run in a dedicated project directory
4. Monitor token usage: Be aware of context window limits
5. Keep updated: Use the latest version of smol-agent and Ollama