A team reviews whether the local agent session and repository setup provide a safe execution context. The repository is cloned to ~/work/payments-api, and the developer wants Copilot CLI to make only repository-scoped changes. Evaluate the following statements. 1. Starting from the repository path is safer than starting from the home directory. 2. The untracked customer export should be removed or isolated before the agent session. 3. The prompt fully prevents Copilot CLI from accessing unwanted local files.

Statement 1 is correct because starting the agent session from the repository path (~/work/payments-api) restricts the agent's initial scope to that directory, making it less likely to accidentally access files outside the intended project. Statement 2 is correct because the tmp/customer-export.csv file is untracked and contains sensitive data. Removing or isolating it prevents accidental inspection or modification by the agent, enhancing security. Statement 3 is incorrect. While the prompt includes "Do not inspect customer data files," natural language prompts are not a foolproof security mechanism. An agent might still access or process such files if its underlying models or tools interpret the instruction imperfectly or if other directives override it. Relying solely on a prompt for file access control is insufficient.