Qianlima Bidding Intelligence Assistant

Problem

Bidding, procurement, contract expiry, and enterprise data are scattered across public notices, internal signals, and business registers. Teams often need to verify whether a project appeared, whether a company won, whether a result set continues, and whether links, contacts, or amounts are real. A model that fills gaps can invent missing fields.

How It Works

The skill routes natural-language requests to fixed chains: search-projects, search-purchase-intentions, search-expiring-projects, and enterprise-intel, with helpers such as area, next, refine, detail, and clarify. Complex queries read references/capabilities/query-strategy.md first and confirm sources, conditions, and call count before execution. Facts come from scripts/qianlima_skill.py. Follow-up actions must use saved session_id and result_set_id instead of silently starting new searches.

Boundaries

Project, purchase-intention, and expiring-project queries require at least one non-empty keyword. Vague requests like “projects in Beijing today” should trigger clarification. Sensitive fields such as contacts, source text, and phone numbers are only called when explicitly requested. The skill distinguishes purchaser, supplier, winner, bidder, and target item. Account access uses https://ai.qianlima.com, and web information must not replace interface facts.