Bidding Agent Data Interface

The Problem It Solves

Integrating external data sources into AI Agents often involves complex API interactions. Bidding and procurement data is valuable but presents specific challenges: converting natural language queries into API parameters, routing to the correct endpoint, and ensuring traceability. Traditional approaches require manual coding for parameter mapping, error handling, and data lineage, increasing development complexity and maintenance overhead. For instance, when a user asks "find IT procurement opportunities for hospitals in Guangdong," the Agent must parse the intent, auto-fill default parameters (e.g., time range, announcement stage), and select appropriate APIs; otherwise, irrelevant results or missing fields may occur.

How the Skill Works

Core Capabilities

The skill (bid-agent-data-interface-v2) encapsulates a complete toolchain for bidding data processing, offering the following capabilities for Agents:

• Natural Language Rewriting: Using the aiSearchSubmitPolling endpoint, it converts user natural language questions into structured search conditions, auto-filling default parameters like region, industry, and time range.
• Intelligent Interface Routing: Automatically selects endpoints based on query intent. For example, SearchProjectForAI (Agent-specific) or searchProjectApi for search lists, getZTBStructreDetail for details, and getCollectUrl for traceability.
• Structured Output Standards: Returns data in a tabular format with key fields such as 项目名称 (project name), 金额 (amount), and 链接 (link), marking missing information (e.g., "公告未披露" for undisclosed announcements).
• Traceability Assurance: Fetches original announcement links and attachments to verify data provenance, supporting RAG or analysis workflows.

Key Steps

A typical workflow using this skill involves:

1. Query Parsing: The Agent first calls aiSearchSubmitPolling to generate search conditions from the user query, outputting identified conditions and auto-completed parameters.
2. Search Invocation: Based on conditions, it calls a search endpoint (e.g., searchProjectApi) to retrieve project listings, returning results in a table format.
3. Detail Retrieval: For deeper information, it invokes getZTBStructreDetail to get structured details, including procurement requirements and risk notes.
4. Result Integration: Merges search and detail data, outputs them per specifications, and suggests next-step calls (e.g., attachment traceability).
5. Error Handling: Stops calls and provides clear guidance for issues like missing API keys or insufficient quota, avoiding automatic retries.

Applicability and Limitations

• Data Dependency: Relies entirely on the Shibobo Technology API, requiring a valid BBIAO_API_KEY. Data coverage is national across industries, but field completeness depends on original announcement disclosures.
• Interface Constraints: The skill uses only a whitelist of six endpoints (e.g., SearchProjectForAI, getCollectUrl), maintaining a clear capability boundary.
• Output Standards: Must state auto-completed conditions; missing fields are marked with specific text (e.g., "不详"), with no data fabrication.
• Error Scenarios: For quota exhaustion (error code 0100590006), it provides a purchase link then stops; non-bidding content (e.g., recruitment, exams) is excluded.
• Use Cases: Suitable for integrating bidding query functions into Agents for market analysis, opportunity mining, or data ingestion, but not for real-time monitoring or high-frequency batch collection.