The Friday morning agent that tells you what your LinkedIn campaigns actually did this week. Variant-aware, narrative-first, no dashboards.
Any of these natural language phrases activates the skill inside Claude Code.
The Campaign Intelligence skill runs every Friday at 11:00 CET. It pulls the week's data from the Unipile Campaigns DB and Variants DB, computes per-variant performance, identifies the winning angle, surfaces under-performers, and writes a narrative digest to Slack and a structured page to Notion.
Where Looker Studio dashboards make you go look at numbers, this skill brings the answers to you. The narrative format means the team gets aligned on the week's takeaways without a meeting. For Earleads agencies running 3 to 10 LinkedIn campaigns simultaneously, this is the weekly attention focus.
The Campaign Intelligence skill sits at the **listen** node for active LinkedIn campaigns. It reads the previous week's variant-level data, computes performance, and produces a narrative interpretation pushed to Slack.
The cadence (every Friday at 11:00 CET) is intentional. Friday mid-morning is when operators have time to act on the week's data before the weekend. Slower cadence (monthly) misses the chance to kill underperforming variants early. Faster cadence (daily) is noise.
The weekly campaign analysis. Yalc runs this on schedule via launchd. Output is a Slack post plus a Notion summary. No dashboards required.
NOTION_API_KEYSLACK_BOT_TOKENSchedule via launchd plist `com.earleads.campaign-intelligence`. Runs every Friday at 11:00 CET. Reads the per-account Slack channel ID from the Yalc workspace config. Falls back to the default ops channel if no per-account override exists.
Friday mid-morning gives operators time to act on the week's takeaways. Earlier in the week and the previous Friday's data isn't a full week. Later and the weekend swallows the urgency.
It groups variants by hypothesis (the angle each variant tests), computes weighted reply quality per variant, and flags the top-scoring as winners. Bottom 25 percent get flagged for kill.
Implicitly. The narrative often says "the curiosity angle is dead, try a value-first angle next". Explicit variant drafting happens in unipile-campaign with this report as input.
Paused campaigns are noted but not analyzed for performance. Resumed campaigns count from the resumption date.
Yes. It reads the prior week's report from Notion and produces delta analysis (acceptance rate change, reply quality shift, new variants introduced).
A Claude classification of each reply into positive intent, objection, not-now, not-interested, out-of-office, automated. Positive intent and meaningful objections count as quality; auto-replies and not-now don't.
Clone the Yalc skill set, drop in your env, run from your next Claude Code session.
gh repo clone Othmane-Khadri/YALC-the-GTM-operating-system && cp -r YALC-the-GTM-operating-system/.claude/skills/campaign-intelligence ./.claude/skills/