The resolve-before-asking skill provides a focused capability for AI agents within the AI Finance Hub: it prevents an agent from asking the user "who is X?" when the brain already contains the answer. This is not a general question-answering system, but rather a mechanism to enforce an exhaustive lookup chain before any identity question reaches a human operator. Its core purpose is to ensure the agent's knowledge base is thoroughly checked and synthesized before requesting clarification on an identity.\n\nresolve-before-asking is particularly useful in environments where an agent frequently encounters new entities or processes bulk information. It ensures that internal knowledge is prioritized and used effectively, reducing friction and unnecessary human prompts.\n\n

The Exhaustive Lookup Chain\n\nWhen faced with an unknown identity, resolve-before-asking orchestrates a detailed, six-step lookup process within the agent's environment. This chain ensures no stone is left unturned before escalation:\n\n1. Cross-Brain Synthesis via think: The agent first attempts to synthesize existing, potentially disparate pieces of information across the brain using the think tool to form a coherent understanding of the entity.\n2. Direct Page Search and Reading: It then performs direct searches across existing pages and reads relevant content within the agent's knowledge base. The search tool is critical here.\n3. Mounted Source Queries: If internal pages don't yield results, queries are extended to mounted external sources accessible to the agent. This involves the query tool to retrieve structured or semi-structured data.\n4. Timeline and Graph Traversal: The agent analyzes historical data and relationships using timeline and graph traversal, looking for mentions, connections, or past interactions involving the entity. backlinks can also be used to find references.\n5. Web Search: As a penultimate step, the agent conducts a targeted web search to find public information about the identity.\n6. Escalation with a Hypothesis: Only after exhausting all previous steps will the agent escalate the question to a human. Crucially, it never presents a bare "unknown." Instead, it formulates a hypothesis based on its findings, even if partial, providing context to the human.\n\nFurthermore, resolve-before-asking mandates that pages created during ingestion must have relationships resolved immediately. No placeholder text is left behind for identities, ensuring data integrity from the outset, often using put_page to update or create complete entries.\n\n

When resolve-before-asking Activates\n\nThis skill is designed to trigger under specific conditions related to identity resolution:\n\n* Identity, Role, or Relationship Questions: Any internal prompt or user query that asks "who is X?", "what is Y's role?", or "how is A related to B?" will activate this skill.\n* Placeholder Text on Pages: If the agent identifies pages with unresolved placeholder text indicating an unknown entity, resolve-before-asking intervenes to fill in the gaps.\n* Presented 'Unknown' Entities: When processing information where an entity is presented as unknown or lacks sufficient context, the skill initiates a lookup.\n* Bulk Ingestion Without Resolved Identities: During large-scale data ingestion, if entities are introduced without their identities or relationships fully resolved, resolve-before-asking steps in to perform the necessary lookups.\n\nConsider an AI agent processing a new financial report mentioning "Dr. Evelyn Reed." Without resolve-before-asking, the agent might immediately ask the user, "Who is Dr. Evelyn Reed?" if her name isn't in a directly linked internal document. With resolve-before-asking active, the agent first synthesizes information using think, then searches internal pages, queries mounted sources, traverses timelines and graphs for past interactions or related entities. If still unknown, it would perform a targeted web search. Only after exhausting these steps, and with a formulated hypothesis like "Dr. Evelyn Reed, potentially a lead researcher at PharmaCorp mentioned in a 2022 acquisition document," would it escalate to a human, never just a bare 'unknown'.\n\n

Distinguishing Skill Responsibilities\n\nIt is important to understand the precise boundaries of resolve-before-asking to appreciate its specific utility. This skill owns whether asking a human is justified regarding an identity, not the broader mechanics of interaction or information retrieval:\n\n* Not General Question Answering: General question answering, such as providing market insights or summarizing documents, falls under the domain of the query skill. resolve-before-asking is narrowly focused on identity resolution.\n* Not Page-Creation Decisions: The decisions around creating or enriching pages, or general brain operations (brain-ops), are separate responsibilities, even though resolve-before-asking might trigger the need for new or updated page data (using put_page).\n* Not How to Format a Question: Once a decision is made to ask a human, the formatting or delivery of that question is handled by other skills, such as ask-user. resolve-before-asking solely dictates the necessity of the question.\n\n

FAQ\n\nQ: How does resolve-before-asking prevent an agent from asking "who is X?" immediately?\nA: It enforces a strict, six-step internal lookup process using tools like think, search, query, timeline, backlinks, graph, and put_page before the agent is permitted to escalate an identity question to a human.\n\nQ: What happens if the agent's brain genuinely doesn't contain the answer after all lookup steps?\nA: The skill ensures that the agent will escalate to a human, but it will always do so with a formulated hypothesis based on its findings, rather than simply stating it doesn't know.\n\nQ: Is resolve-before-asking a tool for general information retrieval or conversation?\nA: No, its scope is specifically limited to resolving identities, roles, or relationships. General question answering or conversation management are handled by other specialized skills.\n\nImplementing resolve-before-asking helps maintain a high bar for agent autonomy in identity resolution. It ensures that human attention is reserved for truly novel or complex identity queries, making agent operations more efficient.