Przejdź do treści
Biblioteka skilli
Sprzedaż i obsługa klienta · Automatyzacja

agentify-outbound

/agentify-outbound

Prowadzi wysyłkę sprzedażową na LinkedIn i e-mailem przez agenta AI w narzędziu Zevari, od podłączenia po tryb autopilota.

Repozytorium
ai-sdr-kit ↗
Licencja
MIT
Ostatnia zmiana repozytorium
08.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
1
Adresy URL w SKILL.md
2
Przestarzałość

Brak sygnałów. Repozytorium zmienione 08.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co sprawdzić przed instalacją
  • SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.

Co robi i kiedy

Prowadzi wysyłkę sprzedażową na LinkedIn i e-mailem przez agenta AI w narzędziu Zevari, od podłączenia po tryb autopilota.

Wyzwalacze z opisu

get my agent doing sales
Oryginalny opis (EN)

The operating procedure for an AI agent running a person's outbound on LinkedIn and email with Zevari. Use when the user asks to set up, run, automate or agentify their outbound, act as their AI SDR, start LinkedIn or cold email outreach, or "get my agent doing sales". Covers standing orders, the trust ladder and eight phases from connecting Zevari to autopilot.

Jak działa

  1. For the human: how to use this file
  2. For the agent: what this file is
  3. 1. Where everything lives
  4. 2. The trust ladder
  5. 3. Phase 1 - Connect
  6. 4. Phase 2 - Business brain
  7. 5. Phase 3 - Voice
  8. 6. Phase 4 - ICP, list and scoring
  9. 7. Phase 5 - Design the campaigns
  10. 8. Phase 6 - The feedback loop
Sekcje z nagłówków SKILL.md, w oryginale · 491 linii, 5 854 słów
Podgląd SKILL.md
---
name: agentify-outbound
description: The operating procedure for an AI agent running a person's outbound on LinkedIn and email with Zevari. Use when the user asks to set up, run, automate or agentify their outbound, act as their AI SDR, start LinkedIn or cold email outreach, or "get my agent doing sales". Covers standing orders, the trust ladder and eight phases from connecting Zevari to autopilot.
---

> Synced copy of https://zevari.ai/guides/agentify-outbound-sop.md. When Zevari is connected, `skills_get` with slug `agentify-outbound` returns the latest version: prefer it.

<!-- guide below: scripts/sync.sh replaces everything after this line -->

# Agentify Your Outbound with Zevari

**Operating procedure for your persistent AI.** Works with any persistent agent (Grok bot, Meta Muse, Claude Duet, etc.) or any other AI agent that can connect to the Zevari MCP.

---

## For the human: how to use this file

1. Drag this file into your AI agent (in a Grok bot, give it to your Chief of Staff).
2. Paste the **Start** prompt at the bottom.
3. Your agent takes it from there. It walks you through every phase, one step at a time, and asks before it does anything.

You don't need to read the rest. Your agent will. If you want the short version: your agent starts by doing nothing, learns your business, voice and buyer from you, drafts everything for your approval, and only runs on its own once you've approved its work unchanged for about a week.

---

## For the agent: what this file is

You are the operator of this person's outbound. This file is your standing order for outbound. Follow it step by step. Save it in your files as `OUTBOUND-SOP.md`. Keep one copy. When the person changes a rule, update this file and the rule files below, never make a second copy.

The person is your manager. They decide. You prepare, draft, verify and report.

### Standing orders (always on)

1. **Default state is nothing.** Until the person approves it, you do not connect, message, comment, like, react, email, enroll leads, start a campaign, spend credits, publish an automation or change a setting. Reading, searching, scoring and drafting are allowed.
2. **Ask before every write, at the current trust level.** Section 2 says which jobs need approval. When in doubt, ask.
3. **Gap rule.** If your files don't say it, write `[GAP]` and ask. Never invent a number, a client name, a result, a credential or a testimonial. A believable invention is worse than a blank, because nobody catches it.
4. **Confirm the context every session.** At the start of every chat, call `zevari_context_get`, tell the person which organization and LinkedIn sender you see, and have them confirm before you call `zevari_context_confirm`. Restate the organization and sender before staging any approval. In a scheduled run, use only the context lock named in the task instructions. If none is named, stop and report.
5. **Approval phrases are exact.** When Zevari returns a chat approval phrase, show it word for word: "To approve in this chat, reply exactly: <phrase>." Give the browser link only as a fallback.
6. **One canonical file per thing.** `BUSINESS.md`, `OFFER.md`, `ICP.md`, `VOICE.md`, `OUTBOUND-SOP.md`. Update them. Never create a second voice, a second ICP or a "v2" file.
7. **Every correction becomes a rule.** When the person corrects a draft or a score, fix the item, then update the rule file so it doesn't happen again, and show the change.
8. **Verify what you did.** After any write, check it in Zevari (pending, progress, the lead's status) before you report it as done. Report what you checked.
9. **Never work around safety.** If Zevari denies an action for safety, limits or working hours, stop and report. Never raise caps, disable the warm-up or change working hours without asking.
10. **Keys stay out of chat.** Tokens and API keys go in your secure input or your private environment file. Never print them back.
11. **A reply stops everything.** When a person replies on any channel, stop every sequence to them and bring the reply to the human with a drafted answer.
12. **Be short.** The person is busy. Lead with what you need from them.

---

## 1. Where everything lives

| What | In your files | In Zevari |
|---|---|---|
| Business, offer, proof | `BUSINESS.md`, `OFFER.md` | Settings → Business profile (profile, positioning, stories) |
| Buyer | `ICP.md` | Settings → Business profile (target fields) and AI team → ICP Scoring → Rules |
| Voice | `VOICE.md` (master) | AI team → Voice Writer → Your voice (Never say, Not your voice, Signature lines) |
| Lists | `LISTS.md` (each search, its URL, how it was built) | AI team → Prospecting → Searches, and Leads |
| Campaigns | `CAMPAIGNS.md` (name, segment, steps, status) | Campaigns |
| Approvals | the trust table in this file | Approvals, and Settings → AI approvals |
| Replies | the daily brief | Conversations |
| Safety | the caps in this file | Safety Center, Settings → LinkedIn senders, Settings → Working hours |
| Stats | `TRAINING-LOG.md` (approval rate per job, per week) | Campaign progress |

Your files win over chat history. Zevari is the system of record for leads, campaigns and conversations. Don't copy those into your files. Read them from Zevari when you need them.

---

## 2. The trust ladder

Every outbound job starts at level 0 and climbs on its own evidence. The jobs are:

- **J1 Sourcing:** run searches, build lists
- **J2 Scoring:** score leads against the ICP
- **J3 Connection requests**
- **J4 LinkedIn DMs** to people who accepted
- **J5 Cold email**
- **J6 Replies** to people who answered
- **J7 Top-tier leads** (score 5): every channel, highest care

| Level | Name | What you do | What the person does |
|---|---|---|---|
| 0 | Off | Nothing | Decides to start the job |
| 1 | Read | Read, search, report what you see | Checks your report against reality |
| 2 | Draft | Draft each item, they approve or correct each one | Approves item by item in chat |
| 3 | Approve the batch | They approve a whole campaign or list up front, you run it inside that box and report daily | Approves batches, reads the daily brief |
| 4 | Autopilot | You run it within the written caps and only escalate exceptions | Spot-checks weekly |

**Move a job up one level when:**
- Level 1 → 2: your reports matched what they saw.
- Level 2 → 3: they approved 80 to 90% of your items unchanged for about 5 working days, and at least 25 items.
- Level 3 → 4: 2 to 4 weeks of batches with no surprises in the brief, and they say yes.

**Move a job down one level on any surprise** (wrong person contacted, a claim that isn't true, a tone they hate, a safety warning). Fix the rule, log it, and climb back.

**J6 Replies and J7 Top-tier stay at level 2 or 3 the longest.** Email (J5) usually earns autopilot first.

Record each job's level and its weekly approval rate in `TRAINING-LOG.md`. That log is the evidence for every promotion.

### How the ladder maps to Zevari's approval setting

Zevari has one approval setting for the workspace (Settings → AI approvals). It is not per job. So there are two gates:

- **Outer gate, Zevari:** what Zevari lets through without a click.
- **Inner gate, you:** which jobs you may run without asking the person. You enforce this from the trust table above, even when Zevari would let more through.

| Zevari setting | What it means | Use it when |
|---|---|---|
| **Strict** | Browser approval for every write | The person wants to approve everything in the Zevari app |
| **Fast** (Zevari's default) | Chat approval for low and medium risk actions, including campaigns up to 20 leads and 5 steps. High-risk actions stay browser-only | Levels 0 to 3. This is where you start |
| **Autopilot** | Eligible DMs, connection requests, comments and reactions run with no click. Optional switch "Include campaign launches" | At least one LinkedIn job is at level 4 |
| **Autopilot Unattended** | Adds paid work (like email lookups), inbound reply content and bigger bundles. Requires "I accept responsibility for unattended sending" | You run scheduled jobs while the person is away, and they say yes |

Rules for the two gates:
- Only an owner or admin can change the Zevari setting. The person changes it, or tells you to and approves it. Never change it on your own.
- The strictest setting wins (workspace, user, sender).
- When Zevari is on Autopilot but a job is still level 2 or 3 in your table, you still ask the person first. Your table is the inner gate.
- Some actions are never auto-approved, even on Unattended: new LinkedIn posts, InMails, and one-off emails sent outside a campaign. Plan for a human approval on those.
- Inside Zevari, the AI team cards show their own unlocks: **ICP Scoring** autopilot unlocks at 10 reviews with 80% correct, and the **Follow-up / Voice Writer** send switch unlocks after 10 approved sends in 30 days. Treat those as the minimum, not the bar. Your bar is section 2.

---

## 3. Phase 1 - Connect

**Goal:** you can see their Zevari workspace, their LinkedIn sender and their safety status. Nothing else changes.

Ask the person for, or confirm:
1. **A Zevari account** with a paid plan (Settings → Billing & plan).
2. **Their LinkedIn connected as a sender** (Settings → LinkedIn senders). Outbound goes from a person, not a company page. People reply to people.
3. **Sales Navigator** (recommended, not required). It gives better search filters and higher weekly limits.
4. **The MCP connection.**
   - Claude, ChatGPT or Codex: Settings → MCP shows the steps for each. No key needed.
   - Grok bot or any agent that can't do a browser login: Settings → Developers → MCP access token. Label it with your name (for example "Grokbot"), pick the workspace and sender, set expiry to the longest option (1 year). The person pastes it into your secure input, never into chat. Put a calendar reminder for them 2 weeks before it expires. When it expires, everything stops with no warning.
5. **Email (optional, for email campaigns):** Settings → Integrations → Email provider. Resend for a small volume from their own domain (under about 200 cold emails a week). Instantly or Smartlead past about 500 a week.

Then test, read-only:
- `zevari_context_get` → confirm org and sender with the person → `zevari_context_confirm`.
- Read the safety status and connection limits (`linkedin_get_safety_status`, `linkedin_get_connection_limits`). Tell the person their weekly invite limit and whether the sender is still in the warm-up period (new accounts ramp up over about 14 days).
- Read working hours (`settings_get_working_hours`). Default is 07:00 to 19:00 in their time zone. Ask if that's right.
- Read the approval setting (`approval_policy_get`). It should be Fast. If it's Autopilot or Unattended, tell them and ask if they want it set back to Fast while you train.
- List what's already there: searches, lists, campaigns, connected mailboxes. Flag anything stale (old campaigns running, mailboxes they no longer use).

**Report:** "Connected to [org], sender [name]. Approval: [mode]. Invite limit: [n] a week, warm-up [yes/no]. Working hours: [hours]. Already running: [list]. Stale: [list]. Nothing changed."

**Gotcha:** if the bot can't find something it should see, the token may be for the wrong workspace. Check before regenerating anything. Test first, regenerate only if the test fails.

---

## 4. Phase 2 - Business brain

**Goal:** `BUSINESS.md` and `OFFER.md` exist, the person has corrected your readback, and Zevari's Business profile matches them.

1. Ask what they already have: website, sales deck, proposals, call recordings, sent emails, case studies, pricing, past wins. Real material beats summaries. If they use Claude or ChatGPT heavily, suggest they ask it for a full brain dump of the business from their past chats and upload the result. It's faster than typing.
2. **One business at a time.** If they sell several things, start with one offer. A second offer later gets its own campaigns.
3. Write `BUSINESS.md` (who they are, what they sell, to whom, how deals happen) and `OFFER.md` (the offer in one line, price or range, what the buyer gets, why now, proof they can use, the call to action). Every factual line ends with its source. Anything missing is `[GAP]`.
4. **Readback.** Show the files as short numbered lines. Ask them to correct any line ("line 3 is wrong"). Apply corrections to the files.
5. **Selling points list.** Ask for the 5 to 10 reasons a buyer says yes, in their words. Save them in `OFFER.md`. You'll weave these into copy.
6. **Objections.** Ask for the top 3 to 5 objections and how they answer each. Save them in `OFFER.md`.
7. Fill Zevari Settings → Business profile from the files: company, what you sell, target industry, role, size, location, default outreach goal and resource URL (booking link or landing page). Use `profile_get` first, then `profile_update`. Show the person what you'll change before you change it.

**Before any campaign, check the funnel.** Ask: when someone says yes, what happens next? Is the booking link tested? Does the landing page work? Who follows up? If a step after the reply is broken, say so and fix it first. More leads into a broken step makes things worse.

---

## 5. Phase 3 - Voice

**Goal:** your drafts sound like them, and Zevari's Voice Writer agrees with `VOICE.md`.

1. **Zevari already has a start.** AI team → Voice Writer analyzes their LinkedIn writing automatically. Read it (`voice_profiles_get`) and show them a summary.
2. **Collect real writing, in order of purity:**
   1. Messages and emails they wrote themselves, with no AI.
   2. Drafts they edited and actually sent.
   3. Their best LinkedIn posts.
   Aim for about 15 samples.
   Little real writing of their own? Offer a ready-made style. Call `skills_list` with `category: "voice"`. Each description says what the style is, who it's for, who inspired it and gives an example. Show them the 3 that best fit their ICP and let them pick. Apply it to one campaign: `voice_profiles_init` with `scope: "campaign"` and the `campaign_id`, then `voice_profiles_patch` on that scope with the returned `revision`, filling `writing_style` (`voice_summary`, `tone`, `sentence_patterns`, `formatting_rules`, `preferred_vocabulary`, `avoided_vocabulary`, `distinctive_phrases`) from the style's instructions (`skills_get`). Their own writing replaces it once they have enough.
3. **Do not train on:** warm-up emails, AI drafts they didn't edit, templates, anything that got zero replies, other people's writing.
4. Write `VOICE.md`: how they open, how long their messages are, words they use, words they never use, sign-off, punctuation habits, tone. Add a short **Never say** list and a **Signature lines** list.
5. Mirror it into Zevari: save samples with `voice_dna_save`, and update Voice Writer's Never say, Not your voice and Signature lines with `voice_profiles_patch`. Show the person the diff first.
6. **Test.** Write 3 sample messages to a made-up lead from their ICP. Ask: "Would you send this as is?" Correct, update `VOICE.md`, repeat until 2 of 3 pass.

`VOICE.md` is the master. If Zevari's voice and `VOICE.md` disagree, `VOICE.md` wins, and you update Zevari to match.

Don't over-tune here. Get drafts flowing, then improve them during Phase 6 with real leads.

---

## 6. Phase 4 - ICP, list and scoring

**Goal:** a list of the right people, and you score them the way the person would.

### 6a. Find the real ICP

1. **Start from real buyers, not assumptions.** Ask for their customer list (CSV, CRM export, or read the CRM if you're connected to it). Group customers by role, industry, company size and what they paid. Show which group is biggest and which pays most. That's the first ICP.
2. Ask: of the deals you've closed, which 3 would you clone? What could you see about them 30 days before they'd take your call? Who looks right on paper but never buys?
3. Write `ICP.md` in three blocks: **Lead** (who to target, must-haves), **Not a fit** (disqualifiers), **Not a lead at all**. Add a 1 to 5 scoring rubric with an example person for each score.
4. Put the rubric into Zevari: AI team → ICP Scoring → Rules (targeting preferences and scoring rubric). Show the person first.

### 6b. Build the list (Sales Navigator)

1. Translate the ICP into a Sales Navigator search: title keywords, industry, geography, headcount. LinkedIn can't filter on everything (for example "owns the practice"). Say which criteria the search can't check, because you'll check those during scoring.
2. The person runs the search and pages through results. **Tune until about half of a page of results looks qualified.** If fewer, change the search, don't filter by hand. Fix the list at the source.
3. **Size:** about 1,000 to a few thousand people is healthy. Under about 700 is too small, widen it. Over about 10,000 is too broad, narrow it.
4. **Make two searches:** one with Sales Navigator's "Posted on LinkedIn" filter (active in the last 30 days), one without. Active people get LinkedIn. Quiet people mostly get email.
5. **Name every search by how it was built**, for example "Family medicine NP - US - 1 to 50 - active". You'll always know where a lead came from. Record it in `LISTS.md`.
6. The person (or you, with approval) adds it in Zevari: AI team → Prospecting → Searches → Add search, paste the Sales Navigator URL, name it, Save and start. Searches refresh daily.

### 6c. Train your scoring (batches of 25)

1. **Remove exclusions before scoring:** current customers, competitors, people already in a campaign, anyone on their skip list. Use Zevari's Exclusion rules. Don't spend credits on people you'll never contact.
2. Score 25 leads. For each one show: name, title, company, score 1 to 5, and one line of why.
3. The person marks each right or wrong, with a reason for the wrong ones.
4. Update the rubric in `ICP.md` and Zevari's Rules from the reasons. Log the batch accuracy in `TRAINING-LOG.md`.
5. Repeat until 2 batches in a row are 80 to 90% right.
6. **Check Zevari's own qualifier before trusting it.** Take 20 leads Zevari marked qualified and score them yourself. If you'd keep fewer than 80%, tell the person, and keep scoring yourself until the Rules are fixed. Zevari's ICP Scoring card unlocks its own autopilot at 10 reviews and 80% correct. Use the Review tab to give it the same corrections.

**Your scoring is level 2 for J2 until this bar is met.** After that, J2 can move to level 3: you score daily and show the counts plus a sample of 5 in the brief.

---

## 7. Phase 5 - Design the campaigns

**Goal:** campaigns that match the lead's score and activity, with copy the person approved, all built and waiting. Nothing enrolled yet.

### 7a. Route by score and activity

The sequences come from the **Outbound Sequence Playbook**. Load it with `skills_get` slug `outbound-sequence-playbook` (plain text: https://zevari.ai/guides/outbound-sequence-playbook.md). It lists every sequence step by step with the exact `campaigns_v2_create` program. The numbers below are its sequence numbers. When you talk to the person about scores, use Zevari's words: 4 to 5 is Strong fit, 3 is Moderate fit, 1 to 2 is Weak fit.

| ICP score | LinkedIn active (last 30 days) | Quiet on LinkedIn |
|---|---|---|
| 5 (Strong fit) | #12 5/5 MVP. 1 to 2 new a day | #13 5/5 quiet: InMail + email |
| 4 (Strong fit) | #11 LinkedIn-active core (or #8 Connect and four follow-ups) | #9 Quiet on LinkedIn: email-led |
| 3 (Moderate fit) | #3 Fit check by email. No LinkedIn invites | #3 Fit check by email |
| 1 to 2 (Weak fit) | No contact | No contact |

InMail steps need a Premium or Sales Navigator sender. Without one, use #7 Connect, email if they don't accept for score 5. For the person's very first campaign, or a free LinkedIn account, start with #1 Connect and say hi on about 10 people and prove the first message gets replies before adding steps.

One persona (segment) per campaign. Don't mix. Name campaigns by segment, channel and start week, for example "NP family medicine - LinkedIn - Strong fit".

### 7b. Sequence rules

Use the playbook's sequences unless the person's files say otherwise. Don't invent new timings.
- Connection requests go out with **no note** by default. A note only when it's specific to that person and couldn't go to 1,000 people.
- Every touch brings a new reason. Never "checking in" or "circling back".
- A reply on any channel stops every remaining step for that lead and moves the conversation to the Inbox. Confirm this in the campaign before you enroll anyone.

### 7c. Copy rules (check every draft before showing it)

- Written in `VOICE.md`. Uses the selling points and proof in `OFFER.md` only. No invented numbers or names.
- LinkedIn DMs under about 50 words. Emails 40 to 80 words.
- One idea and one ask per message. The first message earns a reply, not a meeting.
- Every message has a clear next step (a question or a call to action).
- Say why now (urgency) and what they get (clear value).
- Ask yourself: "If I got this, would I reply?" If not, rewrite.
- Lead with their likely objection in a later touch. Take the fear off the table.
- Only use Zevari's allowed fields: `{{first_name}}`, `{{last_name}}`, `{{full_name}}`, `{{company_name}}`, with a fallback like `{{first_name|there}}`. Anything else is written per lead or left out.
- **Templates first, then personalization.** Run approved templates for the first week to get a baseline reply rate, then test AI personalization on the active group and compare.

### 7d. Build and review

1. Draft each campaign's sequence. Show it as a table: step, timing, message.
2. Show the first 5 leads per campaign with the research reason for each.
3. The person gives feedback (Phase 6 format). Revise until they say the sequence is approved.
4. Build the campaign in Zevari in draft: `campaigns_v2_create` with the playbook's program for that sequence, every `<write: …>` placeholder replaced with approved copy, and `draft: true`. Set `daily_new_leads` and the working hours. Save the campaign in `CAMPAIGNS.md` with its approval date.
5. Do not enroll anyone yet.

**Don't stall here.** If the person is still polishing after two sessions, propose shipping the current version on approval and improving it with real data. A placeholder link is fine while a landing page is being built. A campaign that sends beats a perfect one that doesn't.

---

## 8. Phase 6 - The feedback loop

**Goal:** every correction makes you permanently better.

### How the person should give feedback (teach them this once)

> Name it, quote it, fix it, say why.
> "Variant B, the connection follow-up. You wrote [X]. It should be [Y]. Because [reason]."

Talking is better than typing. If they use voice, take the whole thing and sort it out yourself. If they highlight a passage and send it to you, fix only that passage.

### What you do with feedback

1. Fix the item.
2. Decide where the rule belongs:

| Feedback about | Update |
|---|---|
| Tone, words, length, style | `VOICE.md` and Zevari Voice Writer (Improve your voice, Never say) |
| Who to target or skip | `ICP.md` and ICP Scoring Rules |
| Claims, proof, offer, CTA | `OFFER.md` |
| Timing, steps, channels | `OUTBOUND-SOP.md` and the campaign |
| One step's copy, campaign-wide | the campaign step (`campaigns_v2_update_step_copy`) |
| One lead only | that lead's message (`campaigns_v2_update_lead_message`) or lead notes |

3. Show the change: "Updated VOICE.md: added 'never open with a compliment'."
4. Log it in `TRAINING-LOG.md`: date, job, approved unchanged or corrected, the rule changed.

Editing a step or lead message in Zevari creates a new approval. Old approval phrases stop working. Tell the person.

### The daily brief (every working day, before they start work)

```
NEEDS YOU NOW
- Replies, with a drafted answer for each
- Approvals waiting, with the exact approval phrase
- Anything blocked (expired token, safety warning, missing info)

TRAINING
- Today's drafts or scores for jobs at level 2: good or bad, and why?

WHAT HAPPENED (yesterday, per campaign)
- New leads added, requests sent, accepted, DMs, emails, replies
- Approval rate this week per job, and any job ready to move up
```

Keep it short enough to answer in 10 minutes. One place to approve: this chat.

---

## 9. Phase 7 - Go live on approval

**Goal:** real campaigns running, every send approved by the person, inside safe limits.

1. Confirm the context (org and sender) and restate it.
2. Enroll the first batch: up to 20 leads into one campaign. Under Zevari's Fast setting, campaigns up to 20 leads and 5 steps can be approved in chat. Show the approval phrase exactly.
3. After the person approves, schedule it (`campaigns_v2_schedule`) only when they say go.
4. **Verify:** check `campaigns_v2_pending` and `campaigns_v2_progress`, and report what you saw.
5. Repeat daily, one batch per campaign per day.

### Starting pace (per sender, per day)

| Channel | Start at | Grow to |
|---|---|---|
| LinkedIn-led campaigns | about 10 new leads a day | about 20 a day |
| Email campaigns | about 15 a day | 30 to 40 a day |
| Score 5 (all channels) | 1 a day | 1 to 2 a day |

Start at half pace and increase weekly if acceptance and reply rates hold. Zevari enforces its own weekly limits by LinkedIn plan, plus a warm-up ramp for new senders. Read the live limits (`linkedin_get_connection_limits`) instead of assuming. When a limit is reached, Zevari defers the work, it doesn't fail. Never change caps without asking.

### First checkpoint, one week in

Report acceptance rate and reply rate per campaign. A good LinkedIn benchmark is 35 to 50% acceptance and 15 to 25% replies to messages. Below that, change the copy or the list, not the volume. Test one change at a time, and judge only after 100 to 200 leads.

---

## 10. Phase 8 - Graduate to autopilot

**Goal:** jobs that have earned it run without daily approvals, inside written limits, with a way back down.

1. Each week, read `TRAINING-LOG.md` and tell the person which jobs meet the bar in section 2.
2. Promote one job at a time. Ask the person: "J5 cold email has been approved unchanged 9 of 10 times for 6 days, 47 items. Move it to level 3?"
3. **Write the box** before you promote: daily caps, segments in scope, what stops the job, what you escalate. Add it to the job's line in this file.
4. **When the first LinkedIn job reaches level 4,** the person changes Zevari Settings → AI approvals to **Autopilot** (they may turn on "Include campaign launches" so daily enrollment doesn't wait on them). Jobs still below level 4 in your table keep going through the person in chat.
5. **For scheduled runs while they're away,** the person may choose **Autopilot Unattended** and accept the terms. Explain what it adds first: paid lookups, inbound reply content, larger bundles. Only with their yes.
6. **Scheduled runs need a context lock.** Ask the person to confirm this chat's Zevari context lock with a long duration (365 days) and put its id into the scheduled task's instructions. A scheduled run never picks a context on its own.
7. **Kill switch.** Teach the person one line: "Pause all outbound." You then pause all LinkedIn sending in Zevari (`linkedin_pause_all`) and pause every active campaign, and confirm. Test it once with them. Know the narrower pauses too: one campaign, one lead.
8. **Stays manual:** new LinkedIn posts, InMails and one-off emails always need a human approval in Zevari. Replies (J6) stay at level 2 or 3 unless the person explicitly promotes them.

On any surprise at level 4: pause that job, drop it a level, tell the person what happened and what rule you changed.

---

## 11. Replies

Every reply is the most important thing in the brief.

1. Check Conversations daily. Zevari's Reply Qualifier marks buyer fit, buying interest and conversation depth. Use it, but read the thread yourself.
2. Confirm the sequence to that person stopped on every channel.
3. Draft an answer in their voice, using `reply_context_get` for proof and objection handling:
   - **Interested:** offer two specific times or the booking link, plus one piece of proof.
   - **Not now:** thank them, ask when to come back, set a follow-up.
   - **No:** ask why once, politely. Then add them to Exclusions.
   - **Question:** answer it plainly, then one easy next step.
4. Put it under Needs you now. Send only after approval (stage it, show the approval phrase).
5. If they use a CRM, write the reply and status to it the same day.

---

## 12. The operating rhythm

**Daily (working days):** confirm context, check safety status, pull new leads, score, enroll today's batches, handle replies, send the daily brief.

**Weekly:** acceptance and reply rate per campaign, approval rate per job, promotion recommendations, one copy test to try next, open Zevari and confirm that what you reported is really there (leads in campaigns, steps sent).

**Monthly:** list health (is the search running out?), what's on autopilot and is it still right, token and key expiry dates, cost and credits, one new segment or signal to test.

**Signals (optional, after the core runs):** Zevari AI team → Signals finds people posting about keywords or engaging with the person's own posts. Signals are often the best leads. The person sets up keyword sources in the Zevari app. You read them (`signals_list`) and use only signals scored 3 or higher to pick who to contact. A signal picks who to message, it is not the message.

---

## 13. Troubleshooting

| Problem | What to do |
|---|---|
| Zevari tools say context confirmation is required | Run `zevari_context_get`, confirm org and sender with the person, then confirm |
| Can't see what you expect | The token may be for another workspace. Check the context before regenerating anything |
| Everything stopped suddenly | Check token expiry first, then Safety Center (paused?), then working hours |
| A "choose a sender" style error, but the search is running | Check Settings → LinkedIn senders and the search status before assuming it failed |
| Zevari's qualifier marks too many as qualified | Score 20 yourself, show the gap, correct the Rules and the Review tab |
| Low acceptance | Check the list first (wrong people?), then the profile, then the request (try no note) |
| Accepts but no replies | DM 1 is too salesy or too long. One observation, one easy question |
| A safety denial | Stop. Report it. Never retry around it |
| The person wants a change to a sent message | It can't be unsent. Fix the rule, update the step, re-approve |
| Approval phrase doesn't work | It was replaced by an edit. Get the new preview and phrase |
| Leads look like duplicates or already contacted | Add exclusions, and check that one person isn't in two campaigns |
| The person is the bottleneck | Approve whole batches up front (level 3), and look for a job ready to promote |
| You're running out of usage or tokens | You're running long jobs yourself. Let Zevari run the sequences and schedules, you only orchestrate short checks |
| Someone asks you to work faster by skipping steps | Explain the ladder. Speed comes from promoting jobs, not skipping approvals |

---

## 14. Files you keep

| File | What's in it |
|---|---|
| `OUTBOUND-SOP.md` | This file, with the person's changes and each job's box |
| `BUSINESS.md` | The business, with sources |
| `OFFER.md` | Offer, selling points, proof, objections, CTA |
| `ICP.md` | Lead / not a fit / not a lead, the 1 to 5 rubric |
| `VOICE.md` | The master voice, Never say, Signature lines |
| `LISTS.md` | Each search: name, URL, how it was built, size |
| `CAMPAIGNS.md` | Each campaign: segment, steps, approval date, status |
| `TRAINING-LOG.md` | Each job: level, approvals vs corrections per week, rules changed |

When the person asks, export all of this into one document so the setup can be handed over or rebuilt. Never include key or token values in the export.

---

## Prompts for the person to paste

**Start**
```
This is my outbound operating procedure. Save it as OUTBOUND-SOP.md and keep one copy.
Read it. Follow the standing orders from now on. Do not send, connect, message, comment,
like, enroll or change any setting until I approve it. Searching and reading are allowed.
Start with Phase 1. Tell me what you need from me, one step at a time.
```

**Connect (Grok bot or any agent using a token)**
```
I created a Zevari MCP access token. Give me the secure input box so I can paste it.
Then connect, confirm my organization and LinkedIn sender with me, and tell me what you
see: approval setting, invite limits, warm-up, working hours, searches, campaigns.
Change nothing.
```

**Business brain**
```
Run Phase 2. Ask me for my materials, then write BUSINESS.md and OFFER.md with a source
on every line and [GAP] for anything missing. Read them back to me as numbered lines.
```

**Voice**
```
Run Phase 3. Read my Zevari voice, then ask me for my best real writing. Build VOICE.md,
show me the Zevari voice changes before saving them, then write me 3 test messages.
```

**ICP and scoring**
```
Run Phase 4. Start from my real customers. Build ICP.md and a 1 to 5 rubric. Then score
25 leads from my list and show me name, title, company, score and one line of why.
```

**Campaigns**
```
Run Phase 5. Draft the campaigns by score and LinkedIn activity. Show me each sequence
and the first 5 leads per campaign with your reason. Build them in Zevari in draft.
Enroll nobody until I say go.
```

**Daily brief**
```
Every working day before [time], send me the daily brief from Phase 6: Needs you now,
Training, What happened. Show me the schedule before you save it.
```

**When a draft is wrong**
```
This is wrong: [which draft, which step]. You wrote [X]. It should be [Y] because [why].
Fix it, update the right rule file so it doesn't happen again, and show me the change.
Don't make a second copy of any file.
```

**Promote a job**
```
Show me TRAINING-LOG.md for [job]. If it meets the bar in section 2, write the box for it
(caps, scope, what stops it, what you escalate) and move it up one level. Tell me if
anything needs to change in Zevari's approval setting, and what.
```

**Kill switch**
```
Pause all outbound now. Pause LinkedIn sending and every active campaign, then confirm
what you paused.
```

Wersje

W bazie nie ma innych skilli o nazwie agentify-outbound.

Migawka bazy z 13.11.2026 · wersja paczki 086123eafa15 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 91outbound-sequence-playbook
    Opisuje krok po kroku sekwencje wysyłek sprzedażowych i kampanie Zevari, w tym trasowanie leadów i kontakt przez LinkedIn.
  • 100cold-email
    Prowadzi B2B cold outbound, który zbiera odpowiedzi: od maili i follow-upów po konfigurację wysyłki, LinkedIn, kadencje wielokanałowe i obsługę odpowiedzi.
  • 100gx-outbound-cadence-expert
    Projektuje wieloetapowe sekwencje outbound w e-mailu, na LinkedIn, przez telefon i wideo.
  • 100gtm-outbound
    Tworzy sekwencję outbound i teksty: kadencję, kanały i przekaz dla profilu docelowego, w tym cold maile i outreach na LinkedIn.
  • 90b2b-outbound
    Buduje outbound B2B: zimne e-maile trafiające do skrzynki, prospecting na LinkedIn, dostarczalność i personalizację na dużą skalę.
  • 100cold-outreach
    Tworzy i ulepsza outbound B2B: prospecting, zimne e-maile, outreach na LinkedIn, research kont, sekwencje sprzedażowe i bezpośredni kontakt prowadzony przez założyciela.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.