Contact discovery (spend last)

# Contact discovery (spend last)

You have the "lead-scorer" MCP server connected (Lead Scorer CRM — endpoint https://mcp.lead-scorer.com/mcp, Bearer key created in Lead Scorer → Settings). Use its tools for every read and write. Never invent data: if a tool result is empty, say so.

> **Context first.** If I have an ICP & offer context pack (see the "ICP & offer context pack" skill), read it before anything else and use it instead of guessing. If I do not, ask me the three questions you actually need answered, then continue.

## Goal
Find contact details for list <LIST_ID> — and only for the leads that survived scoring and enrichment and are going into a campaign now.

## Why this one runs last
`find_lead_contact_info` is the most expensive call in the platform: **3 credits per email, 15 per phone**. Scoring is free, enrichment is about 1 credit a lead. So the order is not a style preference, it is where the money goes.

The failure mode is well documented: a run that finds emails for a whole list before deciding who is worth contacting spends roughly 8 credits per usable contact. Running score → enrich → contact, in that order, brings it close to 1. Same leads, same outcome, a fraction of the spend.

An email found for a lead nobody writes to is not an asset. It is 3 credits, gone.

## Steps
1. **Confirm the shortlist.** `get_leads_from_list` filtered to the leads that scored well AND are enriched AND are going into this campaign. If you cannot name the campaign, stop — you are not ready to spend.
2. **Estimate first.** `find_lead_contact_info` with `dry_run: true`. It returns the cost and your balance, spends nothing, runs nothing. Show me the figure before proceeding.
3. **Ask before a large spend.** Above the account threshold the call refuses without `confirm: true`, and says what it would have cost. Do not pass `confirm` to make an error go away — come back to me with the number.
4. **Email before phone.** A phone number costs five times an email. Only look for one if the sequence actually calls.
5. **Collect asynchronously.** Results land minutes later. Re-read with `get_leads_from_list` rather than assuming a failure and re-running — a re-run is a second charge.
6. **Report.** Leads targeted, credits estimated, credits actually spent, emails found, and the ratio of credits per contact obtained.

## Hard rules
- Never on a whole list. Only on a named shortlist headed for a named campaign.
- Never without a `dry_run` first.
- Never re-run a lookup because results have not arrived yet.
- A refusal from the spend guardrail is information, not an obstacle: report it, do not route around it.