The LinkedIn
Inbound Team
Thirty small agents, each one a single instruction file you run inside Claude. You paste something you already have, a draft, a thread, a profile section, and the agent hands back one line no other agent in the pack hands back.
Nothing in here logs into your account, reads your inbox, or sends anything for you. You paste, it answers, you decide what goes out. Every folder is text you can open and read, and the count is checkable: count the directories.
Five stages, six agents in each
The stages sit in the order a stranger meets you: they read your profile, they read a post, they say something under it, they ask for the thing you made, and then they write to you. Here is every agent and the one line it returns.
- 01Headline Rewriter profile-01-headline-rewriter Returnsthree headline candidates with the character count printed against the 220 limit
- 02About Opener Writer profile-02-about-opener Returnsthe two lines that sit above the fold, cut from sentences you already wrote
- 03Experience Block Writer profile-03-experience-rewriter Returnsthe description block rewritten as what you do for a buyer, five lines at most
- 04Featured Section Picker profile-04-featured-picker Returnswhich three go in Featured, in which order, with a caption line each
- 05Search Term Picker profile-05-search-term-picker Returnsten search phrases a buyer types, each mapped to the profile field it has to appear in
- 06Rival Claim Mapper profile-06-rival-claim-mapper Returnsthe two-column claim ledger, the claim only they make and the claim only you make
- 01First Line Writer content-01-first-line-writer Returnsfive opening lines, each under twelve words and each traceable to a numbered sentence of your own draft
- 02Specific Detail Finder content-02-detail-finder Returnsone question per claim the draft never grounds, with the claim quoted verbatim above it
- 03Post Structure Cutter content-03-post-structure-cutter Returnsit reformatted, with the see-more fold marked at a printed character count and the line-break map printed beside it
- 04Comment Question Miner content-04-comment-question-miner Returnsthe topics ranked by how many times the question appeared, with the commenters' words quoted
- 05Offer Mention Placer content-05-offer-mention-placer Returnseither one closing line with the line number it belongs after, or a printed decline saying the link is not there
- 06Carousel Slide Cutter content-06-carousel-slide-cutter Returnsslide-by-slide text with the hand-off sentence printed at every cut, plus the boundaries it rejected
- 01Comment Angle Writer engage-01-comment-angle-writer Returnsthe comment text with the exact line of their post it answers
- 02Own Post Reply Writer engage-02-own-post-reply-writer Returnsone reply per commenter, each ending in a question only that commenter could answer
- 03Audience Overlap Counter engage-03-audience-overlap-counter Returnsthe buyer share of the sampled commenters with the division printed
- 04Repeat Engager Tracker engage-04-repeat-engager-tracker Returnsthe repeat-engager roster with each name's appearance count across the posts supplied
- 05Connection List Ranker engage-05-connection-list-ranker Returnsthe speak-to order with the headline phrase that ranked each connection
- 06Role Change Note Writer engage-06-role-change-note-writer Returnsthe two-line note tied to the specific change, naming the old seat and the new one
- 01Magnet Demand Counter capture-01-magnet-demand-counter Returnsthe magnet concept people asked for most, with every ask counted and quoted
- 02Lead Magnet Namer capture-02-magnet-namer Returnsfive title candidates and one subtitle for the strongest of the five
- 03Magnet Cover Writer capture-03-magnet-cover-writer Returnsthe cover lines with the true, backed count
- 04Gate Keyword Picker capture-04-gate-keyword-picker Returnsthe new keyword and the arming sentence, placed in the draft
- 05Gate Reply Writer capture-05-gate-reply-writer Returnsthe delivery message, nothing else attached
- 06Lead List Builder capture-06-lead-list-builder Returnsone row per captured name, unknown fields marked unknown
- 01Hand-Raiser Sorter convert-01-hand-raiser-sorter Returnsthe two-bucket split of who gets a conversation and who gets the file and nothing else
- 02DM Intent Reader convert-02-dm-intent-reader Returnsthe intent verdict for the thread: asking for the file, asking for help, or selling, with the line that decided it quoted
- 03Inbound Reply Writer convert-03-inbound-reply-writer Returnsthe first reply to a warm inbound message: one question, and no pitch
- 04Call Ask Writer convert-04-call-ask-writer Returnsthe call-ask sentence, attached to the sentence of theirs that earns it
- 05Stalled Thread Reviver convert-05-stalled-thread-reviver Returnsthe single revival message, two sentences, quoting their last concrete detail
- 06Pitch Filter convert-06-pitch-filter Returnsthe decline message for a thread that is selling to you, two lines
You need a Claude account and a place to keep the folder
There is no install, no key, no signup inside the pack. Download it, unzip it, then work down the steps below.
Download the pack and unzip it
Start here. The zip carries all thirty agent folders, their fixtures, and the manifest. Unzip it wherever you keep documents, then follow the folder steps below.
Once it is unzipped you have everything. The rest of this page is about running one of the thirty, and nothing below needs another download.
Running an agent in claude.ai
- Sign in at claude.ai.
- Create a Project and give it the pack's name. In the project knowledge panel, upload the SKILL.md of the one agent you want to run.
- If Projects are not available on your plan, skip step 2: open a new chat and paste the whole SKILL.md file into the message box instead.
- Under it, paste your own input, the draft or the thread or the list, and send.
- Drop the pack folder inside the project folder you already work in.
- Ask for the agent by name in the session: run the gate reply writer on this, then paste your input.
- Claude Code opens the SKILL.md from disk itself, so nothing needs pasting twice.
One agent per conversation. Two SKILL.md files loaded at once and the second set of rules argues with the first, which shows up as an answer in neither format.
Where the folder goes, and how it is laid out
Nothing reads its own path, so Desktop, Documents, or a Drive folder all behave the same. Inside the unzipped folder, five stage folders, six agent folders in each, and one SKILL.md per agent next to its fixtures.
the-linkedin-inbound-team/
MANIFEST.md
profile/
profile-01-headline-rewriter/
SKILL.md
fixtures/
headline-with-offer.md
profile-02-about-opener/
...
content/
engage/
capture/
convert/
MANIFEST.md at the top lists all thirty with the line each one returns. It is the page you are reading now, in plain text, so you can search it without opening a browser.
One run from each stage
Five real shapes, one per stage, so you can see what goes in and what comes back before you spend anything of your own on it. Inputs below are made up.
Offer sentence: I work with operations leads at regional freight firms to move dispatch off spreadsheets. 1. "The shift handover checklist" | PDF | example.site/handover | a visitor gets the steps we run at shift change 2. "Rebuilding a dispatch board, week by week" | post | example.site/rebuild | what the rebuild looked like week by week 3. "Work with me" | page | example.site/start | what the engagement covers and how to begin oneWhat comes back
## Tile 1 The shift handover checklist class: TEACH (a visitor can use it on their own) caption: The steps we run at shift change, and who signs off each one.
Tiles 2 and 3 print the same way, PROOF then ASK, and a census at the bottom counts every asset you supplied against the three tiles.
New hires stop asking questions after week two. In February I sat with a new dispatcher for three afternoons and she asked me nine things. By March she asked me nothing.What comes back
S1 subject noun: hires accounts naming it: none verdict: UNGROUNDED > "New hires stop asking questions after week two." S1 asks for: a named person or company Q: Which new hire had stopped asking you questions by week two?
The two sentences about the dispatcher never say "hires", so the claim above them stays ungrounded and the question asks you to name who you meant.
Their post: "We doubled the support team in one quarter and our first-response time got worse." My first-hand fact: we added two dispatchers in one month and handover went from 12 minutes to 20 for about three weeks.What comes back
matched line: "We doubled the support team in one quarter and our first-response time got worse." comment: "Same shape at nine people. We added two dispatchers in one month and handover went from 12 minutes to 20, and it stayed there about three weeks before it came back." words: 31 / 40
The comment is a neutral draft holding your fact. Put it in your own words before you post it.
Magnet name: The Shift Handover Checklist Link: example.site/handover Commenter's first name: DanaWhat comes back
# Delivery message "Hey Dana, here's The Shift Handover Checklist: example.site/handover. That's everything, no follow-up needed from you." question marks in message: 0 second ask present: no
The two check lines are the point of the agent. A question mark in a delivery message turns a handover into a second ask, so it counts them and prints the count.
Me: "Hey Dana, here's The Shift Handover Checklist: example.site/handover." Dana: "Thanks. Ours lives in three notebooks and nobody agrees which one is current. Does a checklist fix that or do you need something behind it?" Her headline: Operations lead at a regional freight companyWhat comes back
situation stated: yes "ours lives in three notebooks and nobody agrees which one is current" question to you: yes "does a checklist fix that or do you need something behind it" the reply: "A checklist covers the steps at handover. The notebook question sits upstream of it. When two notebooks disagree today, who decides which one is current?" pitch check: clean
One question, no offer sentence anywhere in it. Asking for a call is a different agent, and it waits until there is a stated problem to ask a call about.
When an agent refuses
Sooner or later one of these will decline to answer you. That is the pack working. Each agent is written to stop rather than invent the thing you did not give it, and to say which thing that was. A refusal is a short instruction, and the run it blocks is the one that would have put a made-up detail into your own words.
THE ANSWERS ARE NOT IN THE INPUT A detail supplied here would be one I made up, printed in your voice, in a post with your name on it. Answer the questions from memory and paste the draft again.verbatim from content-02-detail-finder, Refusal 3
The other shape is a bracket in the middle of an otherwise finished answer. The agent writes everything it can and marks the gap where a value should sit, with the value named:
"Hey [NEEDS: the commenter's first name], here's the Shift Handover Checklist: example.site/handover. That's everything, no follow-up needed from you."
That is what capture-05-gate-reply-writer prints when the comment carries no first name. The message is still usable, and the one word it could not know is sitting in brackets waiting for you rather than filled with a guess.
Start with the fixtures
Every agent folder carries a fixtures folder holding at least one practice input. Run an agent on its own fixture first: you get a full output on material you did not have to prepare, and you see the shape before your own draft is in the chat.
content/content-02-detail-finder/fixtures/draft-mixed-claims.md a draft with claims and accounts mixed capture/capture-05-gate-reply-writer/fixtures/no-link-yet.md the input that fires the refusal convert/convert-03-inbound-reply-writer/fixtures/inbound-reply-thanks-only.md
Several fixtures exist to make an agent refuse. Those are worth running twice: once to read the refusal, once with the missing field added, so you know what each agent asks for before you need it.
Before you open the folder
Yes. Download it above and the folder is yours to keep. There is nothing to buy inside it and no account to make beyond the Claude account you run it in.
No. Every agent is a text file of plain English: what to paste, what it does with it, what it will not do. You open a chat, paste, and read the answer.
Run it again with more input. These agents read only what you paste into the chat, so a thin answer usually means a thin paste: the whole draft rather than the paragraph, both sides of the thread rather than the last message.
Open its SKILL.md and edit it. Keep the refusal sections if you want the same behaviour, since those are the lines that stop an agent from filling a gap with something plausible.
I build and run the LinkedIn inbound engine behind a founder's feed: what gets posted, where the comments land, how the asks get captured, and who gets written back to. The thirty agents in this pack are pieces of that work, cut small enough for one person to run in a chat.
If you would rather have it run as one operation, book a call and I will go through what your feed is doing now.
Book a call