trade-mail-group

Use When

• The user has 5-10 prospect companies and wants to avoid one generic cold email.
• The input includes product, target market, company names, industry, country, source, and short customer-intel notes.
• The next step is strategy and draft preparation, not automated sending.

Do Not Use When

• The user needs SMTP setup, bulk sending, warm-up, tracking, or reply automation.
• The user only has a broad market with no company list.
• The task requires confirming procurement intent, budget, decision makers, or reply probability without evidence.

Minimum Input

Required:

• our_product
• target_market
• tone
• customers[]
  • company_name
  • country
  • industry
  • source
  • intel_summary
  • known_signal

Optional but useful:

• source URLs or source notes
• forbidden claims
• brand voice
• offer constraints
• previous follow-up history

See:

• examples/minimal-input.json
• examples/public-germany-input.json
• examples/meta-leads-visible-redacted-input.json

Workflow

1. Read the customer list and extract only stated facts.
2. Group customers into 2-4 mail groups using shared industry, use case, source signal, or risk boundary.
3. For each group, separate:
   • facts from the input
   • conservative business inferences
   • missing information that needs human review
4. Write one outreach angle per group.
5. Write 2-3 English subject line candidates per group.
6. Draft a short first email per group.
7. Add a first follow-up cadence.
8. Add handoff notes for 客户背调skill and 开发信skill.

Output Contract

Return Markdown by default and keep the section names stable. Include:

• one-line conclusion
• input quality check
• grouping table
• per-group rationale
• per-group outreach angle
• English subject candidates
• first email draft
• follow-up cadence
• facts / inference / review notes
• explicit non-goals

If JSON is requested, use these top-level keys:

• sample_type if provided by the input
• one_line_conclusion
• input_quality
• groups[]
  • group_id
  • group_name
  • companies
  • facts
  • inferences
  • outreach_angle
  • subject_candidates
  • first_email_draft
  • follow_up_cadence
  • human_review_points
• handoff
  • to_customer_intel_skill
  • to_outreach_email_skill
  • non_goals

If the input is too weak to group safely, return blocking_recommendation instead of groups[]. Include reason, missing_fields, recommended_next_action, safe_use_of_current_data, and do_not_do.

Quality Gates

• Every customer must appear in exactly one group unless explicitly excluded with a reason.
• Every group must include facts, conservative inferences, and human review points.
• Every first email must include uncertainty language when the source signal is indirect.
• Direct category overlap, competitor, supplier, or marketplace profiles must be marked as risk / benchmark / collaboration candidates, not ordinary buyers.
• If source_url is present, do not claim the source was verified unless you actually read it in this run.
• If the input is too weak to group safely, return a "needs customer intel first" recommendation instead of forcing groups.

Rules

• Do not send email.
• Do not configure SMTP, mailbox warm-up, tracking pixels, or sequence automation.
• Do not invent customer demand, budget, supplier status, decision maker, or reply probability.
• Do not write "they need our product" unless the input explicitly proves it.
• Prefer "may be relevant because..." over hard claims when the signal is indirect.
• If a company fits more than one group, choose the safer group and mark the overlap.
• If source evidence is weak, make the email more exploratory.

Recommended Handoff

• Send uncertain company facts to 客户背调skill.
• Send approved group angle and selected subject line to 开发信skill.
• Keep final sending manual or in the user's approved sending system.

Example Outputs

• examples/minimal-output.md
• examples/minimal-output.json
• examples/public-germany-output.md
• examples/public-germany-output.json
• examples/meta-leads-visible-redacted-output.md
• examples/meta-leads-visible-redacted-output.json

Validation

Run python3 scripts/validate_examples.py after editing examples or the JSON output contract.