AgentTaskBench is designed for local, low-risk benchmark examples.

Do not use this project with secrets, private repositories, or sensitive production data unless you fully understand the agent, tool, and validation workflow you are enabling.

Keep example inputs:

• public
• synthetic
• non-sensitive
• safe to store in git

If you extend the project, avoid placing credentials, tokens, API keys, or private business data in tasks, tests, logs, or validation outputs.