Analyse des performances de campagne

Les consignes originales en anglais ci-dessous sont aussi celles du skill copié ou téléchargé.

# Campaign performance teardown

You have the "lead-scorer" MCP server connected (Lead Scorer CRM — endpoint https://mcp.lead-scorer.com/mcp, authenticated with Lead Scorer OAuth). Use its tools for every read and write. Discover resource IDs with the available list/search tools; never guess or probe sequential IDs, and ask me when no discovery tool exists. Never invent data: if a tool result is empty, say so. An API key is only a manual fallback for clients without OAuth support.

## Goal
Turn campaign <CAMPAIGN_ID>'s numbers into one named cause and one applied fix. "Improve the copy" is not a diagnosis.

## Steps
1. **Pull the state.** Resolve the campaign with `list_campaigns` when needed, then call `get_campaign` and `list_campaign_actions` — sent, opened, replied, bounced, still queued, per touch. `get_leads_from_list` for the underlying list quality.
2. **Walk the ladder in order and stop at the first broken rung.** Fixing a later rung while an earlier one is broken wastes the campaign:
   - **Bounces > 2%** → list quality. Stop; go back to sourcing and enrichment.
   - **Open rate < 25%** with clean delivery → subject lines, or placement. Test subjects before touching the body.
   - **Reply rate < 3%** with healthy opens → the message. Almost always the personalization level, not the wording: check whether drafts used level 1-2 signals or fell back to segment truths.
   - **Replies but no meetings** → the ask. Too much friction, or the wrong next step for that seniority.
   - **Mostly "not now"** → timing and trigger choice, not copy. Change which signal you source on.
3. **Compare the top and bottom deciles.** Read the 5 best-performing messages and the 5 worst. What is structurally different? Name it in one sentence.
4. **Apply the fix to what has not gone out yet** with `update_campaign_action_draft`. A teardown that only produces advice changes nothing.
5. **Tag the outcome** on the leads (`add_tags_to_lead`) so segments stay honest across campaigns.
6. **Report.** Metric table, the one named cause, what you changed, and the single thing to do differently on the next campaign.

## Hard rules
- One cause per teardown. If you list five, you have not diagnosed anything.
- Small numbers lie: below ~50 sends, say the sample is too small instead of inventing a trend.
- Never change more than one variable between two campaigns, or the next teardown is unreadable.