Weisheng WeCom SCRM

Problem

In WeCom customer operations, data for customers, tags, groups, assets, live codes, broadcasts, follow-up records, chat archives, and opportunities is often scattered across pages and tables. Teams need a way to answer operational questions without manually switching backends and reconstructing filters. This skill turns business questions into controlled queries and guarded operations.

How it works

• Business-language prompts: ask for recent new customers, customer tags, follow-up records, chat context, groups, or broadcast scope.
• Runbook-first execution: after triggering, it reads references/agent-runbook.md and follows check-env then check-identity, without reordering steps.
• Read vs. write separation: query requests can return conclusions directly; create, edit, delete, or send actions require explicit user confirmation.
• Document-bound API calls: the flow reads the relevant API document via fetch-raw-doc first, with a cache validity window around 2 hours; call-api uses --doc-url as the document binding, while service_name, uri, and method are completed by the script.
• Script-based execution: normal commands use python3 scripts/scrm.py; use scripts/scrm_sdk.py only when continuing to parse JSON output.
• Low-noise output: for end users, responses lead with business conclusions and avoid exposing internal fields such as service_name, api_path, or doc_url unless requested.

Boundaries

• Requires a Weisheng WeCom SCRM APP KEY; the material says it is stored on the local device and not uploaded. Authorization can be revoked by resetting the key.
• In WeCom bot conversations, if the key is not configured, the skill can use sender_id and bot_id from context to request and persist it automatically; normal CLI or IDE workflows still require manual setup.
• If the current mode cannot run commands or scripts, switch to an executable mode first.
• For image or file uploads/downloads, consult references/file-utils.md to convert local files into usable URLs.