Invitations LinkedIn pertinentes

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

# LinkedIn connection requests that get accepted

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.

> **Context first.** Call `get_my_memory`, then `compile_context_pack` when a lead or campaign is in scope. Treat personal memory as user-owned context, not verified public CRM data. If my memory is empty, ask me the three questions you actually need answered, then continue.

## Goal
Prepare today's connection requests for list <LIST_ID>: a short, signal-based note per person, inside safe volume limits. I review the batch and send.

## Steps
1. **Pick the batch.** `get_leads_from_list`, keep leads with a verified LinkedIn URL. Cap the batch (see limits below) and prioritize by score and signal freshness.
2. **Find the signal** per lead, best available first: `get_lead_posts` / `fetch_profile_posts` (their own recent post — strongest), then a mutual context or shared background, then a company event (`get_company_info`), then their role reality. **If none of these exists, skip the lead and say so.** A generic note is worse than no note.
3. **Write the note.** Target 150-200 characters, structured as: one hook sentence (the specific signal) + a blank line + the CTA.
   - Reference the signal in the first clause, concretely.
   - The CTA earns the connection, it does not ask for a conversation: "Connectons-nous pour [concrete gain for them]" — name the value in their terms ("l'IA appliquée au recrutement", "aux RH", "au conseil").
   - No pitch, no link, no meeting request. The ask is the connection itself.
   - **Banned closings**, they read as a template: "Je serais ravi d'en échanger.", "On reste en contact ?", "Au plaisir de vous compter dans mon réseau."
   - No em dash (— –), no Markdown, French typographic apostrophes (').
4. **Tag** what you used with `add_tags_to_lead` (e.g. `signal:post`, `signal:hiring`) so the post-accept follow-up knows what to build on.
5. **Report** the batch as a table: lead · signal level · note · character count · skip reason if skipped.

## Limits that protect the account (check yours before scaling)
- Personalized note length: **300 characters on Premium/Sales Navigator, 200 on Free/Basic** — and Free/Basic accounts can only add a note to a handful of invitations per month. Write for 200 unless I tell you I'm on Premium.
- Volume: LinkedIn throttles around 20-25 a day and ~100 per rolling 7 days, but **Lead Scorer caps invitations at 15 per account per day**, all campaigns combined, and lower while an account is warming up. Plan for the cap the engine enforces, not the one LinkedIn tolerates: anything above it is simply deferred to the next day.
- Stop immediately on a CAPTCHA or a "weekly invitation limit reached" warning, and tell me.

## Benchmarks (industry figures, not guarantees)
Personalized requests land roughly 35-55% acceptance; generic or blank ones 15-25%. Above 40% is healthy; sustained below 20% means the targeting or the note is wrong — and a low accept rate is itself an account risk. Diagnose the signal level before rewriting the copy.

## Hard rules
- Never guess a LinkedIn profile URL. Verified or skipped.
- Never claim a shared connection, event or read you have not verified.
- Draft only. I send the batch myself.