Customer Hunter

Problem

Sales and BD teams often lack verifiable company profiles when shortlisting prospects. Capital, business risk, hiring signals, and encyclopedia maturity are scattered across systems, and manual research can mix inference with fact. customer-hunter targets this prospecting workflow: given an industry, city, and Top N, it produces a top-customers.html report that ranks companies by budget capacity, demand signals, and collaboration willingness, with recommended products and supporting evidence.

How it works

The skill collects parameters through a browser form, runs environment checks for Python, Node, OpenCLI, and Chrome extension, then fetches data from Qichacha, Baidu Baike, and Boss Zhipin. Scripts persist raw results to JSON, scoring applies built-in rules to produce money and need indices, and the final step renders fixed-structure HTML cards. Each item is labeled as fact, inference, or score, so hiring JD signals are not presented as business facts and heuristic scores are not presented as conclusions.

Caveats

The skill is source-strict: it does not fill missing data from model knowledge, web search, or caches. If a provider fails, the run should stop and report the gap. Collaboration willingness is not auto-scored; sales must complete a checklist. Quota exhaustion, login expiry, and Cloudflare challenges trigger pause signals via exit codes and .hunter-pause.json, then rerun from the failed step rather than the full pipeline.