The Claude Code Playbook for LinkedIn: Outreach and Content

This playbook covers running LinkedIn outreach and LinkedIn content production entirely inside Claude Code: skills installed once and triggered by natural language, a shared safety and dedup layer so nothing runs unattended recklessly, and a clear handoff point to Prosp once a relationship goes from cold to warm.

Two halves, one shared foundation:

1. Outreach: ICP definition, list building, connection requests, feed engagement, enrichment
2. Content: post writing, hooks, carousels, the content calendar, and the anti-AI-slop pass

Every skill below is a real, installable Claude Code skill pulled from the existing GTM and LinkedIn libraries in this workspace, consolidated here so LinkedIn specifically has one home rather than being scattered across separate lead mags.

Why Claude Code specifically, not Cowork or chat

Claude Code gives skills persistent local files, version control, and scheduled execution, which matters for LinkedIn work because both outreach and content are recurring, not one-off. A skill installed once in a project folder runs the same way every time it is triggered, with campaign logs and enrichment logs building up in the same folder over weeks rather than living inside separate chat threads that eventually scroll out of reach.

Cowork is the right choice if you want zero terminal interaction and are comfortable with a slightly less customisable setup. This playbook assumes Claude Code specifically because the skill and safety-cap architecture below depends on persistent local files and campaign logs that Claude Code manages more reliably for ongoing work.

Project setup

mkdir linkedin-system && cd linkedin-system
mkdir -p .claude/skills
mkdir -p campaigns
mkdir -p enrichments
mkdir -p content/drafts
mkdir -p content/published
claude

Write a CLAUDE.md in the project root before installing any skill. This is your standing context file: ICP definition, brand voice rules, banned phrases, and the daily caps every outreach skill will check. Every skill below reads from this file so you do not re-explain your business every session.

The one file every skill in this playbook reads before doing anything. Get this right once and every outreach and content skill downstream inherits it automatically.

The CLAUDE.md template

# CLAUDE.md - LinkedIn System

## ICP
Title(s): [exact titles]
Company type: [B2B SaaS / agency / other]
Company size: [employee range]
Geography: [regions]
The specific problem they have right now: [in their words, not marketing language]
What they are already doing about it: [current workaround or tool]
Where they spend time on LinkedIn: [5-10 real creator names they follow]
Disqualifiers: [what rules someone out immediately]

## Offer
[What you sell, in one sentence]

## Brand voice
WE ARE: [3-5 adjectives]
WE ARE NOT: [3-5 anti-adjectives]
Banned: no em dashes, no hashtags in post bodies, no: leverage, synergy, bandwidth,
genuinely, honestly, straightforward
British English. One idea per line. Three lines maximum per paragraph.

## Daily caps (outreach)
Connection requests: 20/day
Profile enrichment lookups: 100/day
Searches: 15/day
Feed comments: 15/day, 30-60 seconds apart

## Content rules
No AI-slop phrasing: no "it's not just X, it's Y", no rhetorical question openers,
no triple-adjective stacks, no generic superlatives without a number attached.
Every post needs one specific detail (a name, a number, a real example) that
only the author could have included.

4 skills covering the full outreach motion, from finding the right people to warming them up before ever sending a connection request. All 4 share the daily caps and cross-skill dedup defined in CLAUDE.md.

linkedin-outreach: full B2B outreach campaign

Triggered by: "run an outreach campaign", "find leads and connect for [ICP]"

The 8-stage flow: intent-first ICP discovery, a draft ICP proposal, filter editing before anything runs, company search, decision-maker extraction, false-positive filtering, multi-account selection with dedup, and connection sends with mandatory campaign log persistence.

mkdir -p .claude/skills/linkedin-outreach

Run a LinkedIn outreach campaign.
Load CLAUDE.md for ICP and offer.
Connection note: reference something specific about their company or role.
Under 200 characters. No pitch.
Check daily cap before starting. Report remaining budget.
Log every send to ./campaigns/{today}-outreach.md

linkedin-high-intent: outreach to post commenters

Triggered by: "find leads from this post URL"

Scrapes all commenters and reactors on a given post, filters on visible signal first (no enrichment yet, to conserve credits), enriches only the matches that pass, then launches outreach on confirmed matches only. Post-engagement outreach outperforms cold outreach meaningfully because the prospect already signalled interest in the exact topic the connection note references.

Run a high-intent campaign from a post.
Post URL: [paste]
Filter visible signal first before enriching.
Connection note references the post they engaged with. Sound human, no pitch.
Log to ./campaigns/{today}-high-intent.md

linkedin-feed-engage: auto-comment on ICP feed posts

Triggered by: "comment on my feed", "warm up these prospects"

Scrolls the feed, identifies ICP-matching posts, generates a 2-3 sentence value-adding comment for each (never a generic compliment, never mentioning your product), and posts with strict pacing. This is the automated version of the comment-first list-building strategy: commenting on a creator's posts for a few days before ever sending a connection request produces meaningfully higher accept rates than a cold connection request alone.

Run a feed engagement session.
Comment mode: HITL-approved.
2-3 sentences, must add something the post did not say.
No product mentions, no "great post", no em dashes.
Pacing: 30-60 seconds apart. Stop at 15 comments.
Log to ./campaigns/{today}-feed-engage.md

linkedin-enrich: bulk profile enrichment

Triggered by: "enrich these profiles", "get emails for this list"

Turns a list of LinkedIn URLs into structured data: role, company, size, location, and a professional email when findable. Checks the last 30 days of enrichment logs before running any lookup so you never pay twice for the same profile.

Enrich these LinkedIn profiles: [paste URLs]
Output: CSV with role, company, location, email if findable.
Check ./enrichments/ from the last 30 days before any lookup.
Log to ./enrichments/{today}-enrich.md

The warmup sequence, chained

Day 1-3: linkedin-feed-engage comments on 3 to 5 posts from each ICP-matching target.

Day 4-5: linkedin-outreach sends the connection request, referencing that engagement history in the note.

Day 6+ (after acceptance): linkedin-enrich pulls full profile data, then hand off to Prosp for the DM sequence.

This is the slowest sequence to run but produces the highest acceptance and reply rates, since the prospect has seen your name several times before the connection request arrives.

6 more outreach-adjacent skills, pulled from the wider GTM skill libraries in this workspace, for the parts of outreach that sit before or around the core 4.

icp-prompt-builder

Produces the exact search filter string and qualification criteria from an ICP definition, the direct input any list-building skill needs.

Load CLAUDE.md. Produce the ICP search prompt:
titles, seniority, company size range, industry, geography, technologies,
the one disqualifier that rules someone out immediately, and the exact
Prospeo or Sales Navigator filter string to use.

disco-like: lookalike company discovery

Finds companies similar to a seed list of your best-fit customers, for expanding a target list once the obvious first-pass targets are exhausted.

Seed: [5-10 domains of best-fit customers]
Find [N] similar companies. For each: name, domain, why it matches,
size estimate, whether already contacted.
Output as lookalike-companies.csv

competitor-engagers: the strongest buying-window signal

Finds people actively engaging with competitor LinkedIn content. Someone commenting on a competitor's product post is actively researching your exact category right now.

Competitor LinkedIn profiles to monitor: [URLs]
Scrape commenters and reactors on recent product/pricing/feature posts.
Filter for ICP matches. Find verified email for each match.
Output: name, title, company, email, LinkedIn URL, which post they engaged with.

buying-signal-monitor

Watches public LinkedIn activity for buying intent and returns warm leads, each anchored to the specific post that proves the signal. Preps evidence-backed outreach; does not send it, the same read-only design principle used across the outreach research layer.

list-quality-scorecard

Grades a finished lead list across 8 dimensions (email validity, title match, size fit, geography, duplicates, missing fields, seniority, domain health) before it goes near a live sending account. Only a B grade or above should move to outreach; anything lower goes back through filtering first.

warm-signal-tracker

Monitors repeat commenters, profile viewers, and post-savers against the ICP and flags anyone who has crossed the threshold to be worth a proactive DM, rather than waiting passively for them to message first.

The writing side. 6 skills covering hook selection, drafting, the humanizer pass, and the write-hook-last workflow.

linkedin-post-writer

Drafts a full post using a chosen hook mechanism (specific-number, contrarian, mistake-confession, before-and-after, direct-question) and one of four core body structures (story-to-lesson, list-with-context, problem-agitate-solve, framework-reveal).

Write a LinkedIn post about [TOPIC].
Select the hook type based on the idea, name it in the output.
Apply CLAUDE.md brand voice throughout.
Return: hook type used, the post, one CTA suggestion, one A/B hook variant.

hook-last-editor

The write-the-hook-last method: writes the full post body first, then finds the single sharpest line already inside it and pulls it to the top as the hook, rather than trying to write the hook before the substance exists.

Here is a full draft: [paste]
Find the sharpest, most specific line already in this post.
Move it to the top as the hook. Do not invent a new hook from scratch.
If nothing in the draft is strong enough to pull, say so: the problem
is the post, not the hook, and it needs a rewrite.

hook-extractor

Reverse-engineers the hook formula from any viral post pasted in and returns a blank, fillable template using that exact structure for your own topic.

humanizer

Strips the specific phrasing patterns that read as AI-generated: em dashes, the "it's not just X, it's Y" construction, rhetorical question openers, triple-adjective stacks, and generic superlatives with no number attached. Run as a separate pass after the first draft, not as a constraint during initial generation.

List every phrase in this draft that reads as generic or AI-generated.
Rewrite each flagged section in plainer, more specific language.
Keep everything else in the draft unchanged.

content-bucket-planner

Sorts a week or month of planned topics into a 40% growth / 30% authority / 20% conversion / 10% personal split, flagging any week overweighted toward one bucket. Posting 100% authority content to an audience that does not know you yet is handing out business cards in an empty room; posting 100% growth content with no conversion layer builds an audience that never buys.

point-of-view-extractor

Interviews you about what you believe that contradicts your industry's consensus, and turns that into a bank of contrarian angles. A genuine point of view is the one thing running underneath every content bucket that a competitor cannot copy and that AI cannot generate on its own, because it is a stance built from real experience, not a fact.

5 skills for turning content into visuals and running it as an ongoing calendar rather than a one-off post.

carousel-generator

Takes source material (a blog post, transcript, or outline) and produces a slide-by-slide carousel structure first, separately from visual design. One idea per slide is the rule; cramming two ideas onto a slide is the single most common carousel mistake.

Source: [paste blog post, transcript, or outline]
Step 1: draft the slide-by-slide structure. One idea per slide.
Step 2: write the slide text and a consistent visual template description.
Step 3: write the caption. Tease the carousel content, do not restate it,
so there is still a reason to swipe through.

total-addressable-content-finder

Given the ICP in CLAUDE.md, returns what that audience is currently watching, reading, and reacting to on LinkedIn right now: the specific input for building content around conversations already happening rather than starting from zero attention.

borrowed-reach-formatter

Takes a topic and drafts it in five distinct formats (brand-jacking, newsjacking, referencing a known figure, hot take, trend-riding) so you can compare angles rather than committing to one blind.

content-planner

Builds a 7-day plan with topics, formats, hooks, and posting times in one pass, so content ideation happens once a week rather than daily under time pressure.

repurposer

Converts content from another platform (a tweet, a video transcript, a blog post) into a native LinkedIn post: re-hooks the opening for LinkedIn's fold, expands to the 900-1,300 character range, moves any links to the first comment, and runs the humanizer pass automatically before returning the draft.

The outreach half and the content half are not separate businesses. Content builds the audience outreach draws from, and outreach converts the engagement content generates. This is where the two halves connect, and where Prosp sits.

The full loop

1. content-planner and linkedin-post-writer produce and publish content on a schedule
2. linkedin-feed-engage runs the daily comment routine on ICP-matching creators independent of your own posting
3. Comments and engagement on your own published content get monitored by warm-signal-tracker and buying-signal-monitor
4. linkedin-high-intent runs against any of your own posts that got strong engagement, or against a relevant creator's post, to find and score commenters
5. linkedin-outreach sends connection requests to confirmed matches, referencing the specific engagement
6. Once accepted, linkedin-enrich pulls full profile data
7. The enriched, warm contact hands off to Prosp for the structured DM sequence

Why the handoff to Prosp happens at step 7, not earlier

Everything before step 7 requires judgement: deciding who actually matches the ICP, writing a comment that adds real value instead of generic praise, and drafting a connection note that references something specific rather than a template. That is exactly the work Claude is suited to and where the skills above earn their place.

Step 7 onward requires reliable, scheduled execution across a growing number of accepted connections: sending the first DM, tracking who replied and who has not, following up on schedule, and moving people through pipeline stages without anyone falling through a gap as volume increases. That is what Prosp is built for. Handing the sequencing off at exactly this point, once Claude has already done the judgement work of finding the right person and earning the connection, is what lets this whole system run without someone manually babysitting a spreadsheet of accepted connections and follow-up dates.

All 21 skills, the daily setup routine, and the weekly rhythm in one place.

All 21 skills, by category

Core outreach (4): linkedin-outreach, linkedin-high-intent, linkedin-feed-engage, linkedin-enrich

List building and signals (6): icp-prompt-builder, disco-like, competitor-engagers, buying-signal-monitor, list-quality-scorecard, warm-signal-tracker

Writing (6): linkedin-post-writer, hook-last-editor, hook-extractor, humanizer, content-bucket-planner, point-of-view-extractor

Content production (5): carousel-generator, total-addressable-content-finder, borrowed-reach-formatter, content-planner, repurposer

Install everything

mkdir -p .claude/skills/{linkedin-outreach,linkedin-high-intent,linkedin-feed-engage,linkedin-enrich,icp-prompt-builder,disco-like,competitor-engagers,buying-signal-monitor,list-quality-scorecard,warm-signal-tracker,linkedin-post-writer,hook-last-editor,hook-extractor,humanizer,content-bucket-planner,point-of-view-extractor,carousel-generator,total-addressable-content-finder,borrowed-reach-formatter,content-planner,repurposer}

mkdir -p campaigns enrichments content/drafts content/published

Write CLAUDE.md before installing any skill content. Every skill above reads from it.

The weekly rhythm

Monday: content-planner builds the week's topics and formats. point-of-view-extractor and total-addressable-content-finder feed the ideas that need a fresh angle.

Daily: linkedin-post-writer drafts the day's post, hook-last-editor and humanizer run as the pre-publish pass, linkedin-feed-engage runs the daily comment routine.

As engagement comes in: warm-signal-tracker and linkedin-high-intent identify who is worth a connection request.

Weekly: linkedin-outreach runs a fresh campaign batch against the current ICP list. list-quality-scorecard grades any new list before it gets used.

Ongoing: every accepted connection gets enriched, then handed to Prosp for the DM sequence, so the volume of warm relationships never outpaces the ability to actually follow up on them.