linkedin-personas
Skill6 starsChanged 12 days agoMITView on GitHub
Builds 2 to 4 persona cards with the user and puts each lead in one. Use before writing to a new list, or when messages sound generic or repeat the job title.
In this repository5 other files it gives its agents.
- bereach-setupSkill
- cold-emailSkill
- linkedin-messagesSkill
- linkedin-outreach-methodSkill
- linkedin-voiceSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 38
Before drafting, read references/persona-card.md (Core, Rules) and references/frameworks.md (Which card field comes from which framework).
- Very specific: line 64
Before sorting, read references/persona-card.md (ICP, plan groups and cards.
- Very specific: line 81
Only when the user asks for a score: bereach_check_fit with the campaignId, icp as the plan's ICPs in their words, personas as …
- Very specific: line 96
After 7 days linkedin-outreach-method draws Results per persona (tag persona:<short-name>, bereach_inbox for replies).
- Very specific: line 107
After a rename or merge: one bereach_contact_update per persona, tags: { add: [new], remove: [old] }, then save again (Step 3).
- Distinctive: line 116
Use age, gender, hobbies, school, personality type or a photo to sort or to write.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Personas
The ICP (in the search plan) says which companies. A persona says which person inside them and what to say to them: a role in a situation, never a person with a life. The title finds the card; the card writes the message, never "I saw you are Head of X".
Blocks: Persona cards v1 and Who is who on linkedin-outreach-method's Plan and personas
page (its "3 premiers messages" band comes from the v1 cards at once, via linkedin-messages), Who is who
again on Invitations to send; each its own page only when this skill runs alone. Before drawing a block,
read references/artifacts.md (its section).
Step 1. Read what exists
bereach_memory_list,fullContent: true, forofferinganduser-profile; the offering's "Personas" section is the starting point.- Their website, CRM (if a connector is there) and documents (
references/sources.md); inside linkedin-outreach-method, after its question round. - The search plan, if
bereach_plan_searchesran here: each group (a role in one segment) goes into the Found by of its card; groups that would get the same first message share one card, even across segments; roles the plan guessed end in(guess). - The account's untouched people, as linkedin-outreach-method split them (a few hundred per persona, never every row): their real titles shape Found by.
- What got replies, from linkedin-outreach-method (only with the user's yes): its "Their words" list seeds Pain, Objection and Their words. Never read the same conversations twice.
Step 2. Draft the cards with the user
Asked in linkedin-outreach-method's one question round, only what Step 1 left empty or (guess):
- Your best 3 customers: who signed, their title, what happened just before the yes?
- The "no" you hear most, and your answer to it?
- What have you already offered a prospect (a sheet, a plan, a sample of your work), and from what price?
Before drafting, read references/persona-card.md (Core, Rules) and references/frameworks.md (Which card
field comes from which framework); the band names only a framework the cards visibly apply.
Draft 2 to 4 cards. Cards are short: a seller writes a message from them in one read. A line that is not the
user's words or a reply ends in (guess). Get right:
- Proof: a real case in the user's words. Unnamed, it needs no consent; a named client waits for the yes. None given: "None yet", never a case Claude wrote, never a site or deck figure.
- Offer: only one the user really made (a message, their own words) or a prospect asked for. None: empty, the user is asked once; never a study, estimate, audit or map they have not offered.
- The objection answer and price: the user's words only (no service, speed or result they never stated), else an empty text field.
Deliver: Persona cards v1, provisional by design; an answer given later is saved and the block republished in place.
Step 3. Save the cards
bereach_memory_save,type: "offering", only on a change. It replaces the whole entry: the full existing text (or what they sell, from the question round) plus a## Personassection with every card and any validated templates.- New titles or roles go to the next
bereach_plan_searches; a Why now, Signal or Proof picks its door in linkedin-outreach-method'sreferences/search-doors.md.
Step 4. Put each lead in one persona
Before sorting, read references/persona-card.md (ICP, plan groups and cards; More: Not this persona).
Untouched people first, then each search's, from the row (title, company, whyNow); no profile data:
amber pas de profil : N. Keep on the row any fact settling the persona, dated (month and year).
- One persona per lead. Two fit: the one whose Why now window the dated
whyNowfalls in, or whose ICP the company is in, elseCHECK. A batch rule for titles outside a card's Found by adds them to it on the same rule row; else they stayCHECK. sure: the title matches exactly one card's Found by and nothing puts the company outside its ICP. Enough to tag.CHECK: two cards fit with nothing to split them, or the company is outside the ICP. No tag, no message until the user picks; the others go on.- Out of scope (student, intern, side project, the sender's own profile, team or company, too big) stays
at the search step, never reappears. One added later:
HOLD, no message; on the user's yes,bereach_contact_updatewith the list'scampaignIdandlifecycleStage: "rejected". - Tag:
bereach_contact_update, allcontactIdsof one persona,tags: { add: ["persona:<short-name>"] }, one short name per persona everywhere.
Only when the user asks for a score: bereach_check_fit with the campaignId, icp as the plan's ICPs in
their words, personas as "<short-name> (title, title, title)" (3 titles at most). Tag from
fit.persona; it hides and blocks nobody; say its cost before and what it spent after.
Deliver: Who is who. Sure rows need no yes; open rows are held and asked on Invitations to send, never a stop of their own. Its total is the count every later page reads.
Step 5. Hand over to the messages
linkedin-messages writes one template per persona from Why now, Pain, Proof, Objection and Offer, line 1 from the lead's brief (every person briefed before every message), else the moment line, never their title or headline. A reply asking for something is evidence for the Offer.
Step 6. Learn from results
After 7 days linkedin-outreach-method draws Results per persona (tag persona:<short-name>,
bereach_inbox for replies). Then, with the user, never on a partial read:
- A line changes only when 2 or more replies say the same, quoted; every change names its source (a reply, or "votre idée"). A new offer with no source is a question, never filled in.
- Confirmed lines become
seen in N replies; contradicted ones are rewritten; an answer is asked again only when replies contradict it, and a new objection's answer is the user's. - Nothing confirmed, fixed or merged under 10 invited.
- A change to Pain, Offer or the objection answer is queued as that persona's next test (linkedin-outreach-method's measuring order), never edited live.
- 0 replies on 8 or more first messages: the Why now check, then one plain change or a
pause; the user picks. - After a rename or merge: one
bereach_contact_updateper persona,tags: { add: [new], remove: [old] }, then save again (Step 3).
Deliver: Persona cards v2, only the changed lines, one decision.
Never
- Invent a pain, quote, proof, offer, tool, date or answer the user did not give and no reply or brief shows.
- Quote a named customer before the user says they may be cited.
- Use age, gender, hobbies, school, personality type or a photo to sort or to write.
- Say where the data comes from or how people were found.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

