agent-interview

name: agent_interview description: A skill that allows the agent to conduct a structured or unstructured interview with the user to gather information.

Agent Interview Skill

This skill allows you to interview the user to gather specific information, clarify ambiguous requirements, or build context (e.g., for user memory).

Tools

fetch_context_gaps

Retrieve pending questions or issues that have been logged by other agents.

• No arguments.
• Returns a list of pending gaps to address.

ask_user

Ask a single question to the user and wait for their response.

• question (str): The question to ask.
• category (str, optional): A category for the question (e.g., "personal", "project", "preferences"). Defaults to "general".
• options (list[str], optional): A list of predefined options for the user to choose from.

finish_interview

Call this when you have gathered all necessary information.

• summary (str): A brief summary of what was learned.
• suggested_offline_tasks (list[str], optional): A list of tasks (e.g., research topics, skill building) that the system should perform offline based on the interview results.

34: ## Standard Daily Protocol 35: 36: For regular daily interviews (e.g., 9:30 AM check-ins), follow this Standard Daily Protocol to ground the conversation: 37: 38: ### Phase 1: Goal Alignment 39: Start by grounding the user in their objectives. ALWAYS ask these questions first (unless the user explicitly skips): 40: - "What are your goals for Today?" 41: - "What are your goals for This Week?" 42: - "What are your goals for This Month?" 43: 44: ### Phase 2: Gap Resolution 45: After goals are set, check for pending issues logged by other agents. 46: - call fetch_context_gaps to retrieve pending questions. 47: - Address high-priority gaps first. 48: 49: ### Phase 3: Open Floor 50: Finally, give the user space to provide unstructured context. 51: - Ask: "Is there anything else you'd like to discuss or add to our context?" 52: 53: ## Usage Guidelines 54: 55: - One Question at a Time: Do not overload the user. 56: - Dynamic Flow: Adapt your questions based on previous answers, but stick to the protocol phases. 57: - Identify Offline Work: If the user mentions a topic that requires research or a new skill, add it to suggested_offline_tasks in finish_interview. 58: - Closing the Loop: Only call finish_interview after the user has had the final opportunity to speak.

Example Workflow

1. Agent: ask_user("What is your primary role on this project?")
2. User: "I'm the lead architect."
3. Agent: ask_user("Do you have a preferred programming style (e.g., functional, OOP)?")
4. User: "I prefer functional python where possible."
5. ... (more questions) ...
6. Agent: ask_user("Is there anything else you'd like to add?")
7. User: "No, that covers it."
8. Agent: finish_interview(summary="User is lead architect, prefers functional Python...")

Automated Scheduling

To ensure regular context updates, you can schedule a weekly interview check using the provided cron script.

uv run scripts/schedule_weekly_interview.py

This installs a cron job that runs weekly (Monday 9am) to check for pending gaps.