Trade Workflow Router

Trade Workflow Router

Overview

Act as the entry point for foreign-trade workflows. The core rule is to ask the user to choose a clear task when their request is broad, then route the selected work to the relevant skill or executable tool.

When to Use

Use this skill when the user only mentions foreign trade, export sales, customer development, or asks what the agent can do.

When the user already asks for a specific task, route directly to the matching skill without showing the full menu.

Inputs

• User request
• Existing profile configuration, if available
• Existing customer list, website URL, reply text, or quotation details, if provided

Outputs

{
  "intent": "",
  "selected_workflow": "",
  "required_inputs": [],
  "skills_to_use": [],
  "tools_to_run": [],
  "next_step": ""
}

Procedure

1. If the user request is broad, ask the user to choose from this 外贸工作菜单:
2. Find overseas prospects and collect lead candidates
3. Process a CSV or Excel customer list in batch
4. Research a company website
5. Score and prioritize prospects
6. Find decision-maker clues
7. Write personalized outreach emails
8. Classify buyer replies
9. Plan follow-up tasks
10. Create a quotation draft
11. Export quotation HTML, PDF, or Excel files
12. Ask the user to choose one item or describe the desired result in their own words.
13. When file output is needed, ask for the preferred output format once and use the recommended default if the user does not care:

• prospect discovery: csv + json
• quotation export: html + excel
• pass the chosen formats to tools with --formats

1. Route prospect collection to prospect-discovery; use tools/collect_prospects.py when file output is needed. If the agent has source pages for the current task, pass them as runtime --source-url values. If DISCOVERY.yaml sets discovery_mode: "native_scrapling_spider" or scrapling_spider.enabled: true, collect_prospects.py uses tools/scrapling_spider_runner.py and writes prospects.raw.csv, prospects.raw.json, and crawl_report.json. If MCP is configured, the same workflow is exposed by tools/scrapling_mcp_server.py.
2. Route batch list processing to prospect-list-enrichment; use tools/batch_prospect_pipeline.py when one-step file output is needed.
3. Route website background research to company-research.
4. Route ranking and qualification to prospect-scoring.
5. Route decision-maker work to decision-maker-finder; use tools/decision_maker_finder.py when JSON output is needed.
6. Route personalized email drafting to email-crafting.
7. Route inbound buyer messages to reply-classification.
8. Route reminders and next-touch timing to follow-up-planner.
9. Route quotations to quotation-generator; use tools/render_quotation.py for HTML, PDF, or Excel file export.
10. When the user provides a website, CSV, Excel file, or discovery configuration, run the tool for the selected workflow before summarizing results.
11. Do not output company facts, scores, buyer claims, or personalized email lines unless they are backed by fetched evidence from the downstream tool output.
12. After completing one workflow, recommend the most relevant next workflow and ask whether the user wants to continue.

Verification

• Broad foreign-trade requests receive the menu before work begins.
• Specific requests route directly to the matching skill.
• The selected workflow lists missing inputs before execution.
• File-output workflows ask for an output format once, then use the recommended default when the user has no preference.
• The router does not claim completion before the selected downstream workflow finishes.
• Website and list workflows are based on fetched evidence, not generic examples.
• Prospect discovery with configured public source URLs produces crawl_report.json; if no rows are found, report that status instead of inventing prospects.
• The router offers a relevant next step after each completed workflow.

Common Mistakes

| Mistake | Fix | | --- | --- | | Showing the menu after a specific request | Route directly to the requested skill | | Running every skill at once | Use only the selected workflow and required dependencies | | Asking vague follow-up questions | List the exact missing files or fields | | Stopping after one task without context | Offer the most relevant next workflow |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Xuxchloris
• Source: Xuxchloris/hermes-export-skills
• License: MIT

Install and usage instructions live in the source repository linked above.