Book a call

A Fable InLead playbook

AI sales automation, with a person in charge.

A Claude Code setup that researches each company, writes one message and checks it. A person approves before anything goes out. Nine steps, a worked example and twelve templates you can copy. For B2B founders and heads of sales.

By Saiful Islam Saif, Founder, Fable InLead. . 37-minute read.

The setup ideas credit Charlie Hills (MarTech AI) and Anthropic’s Claude Code documentation. Every outside fact links to its source at the back.

Fable InLead flame mark
9

Steps from a company name to an approved draft. A person approves at step 7.

69%

Of B2B buyers prefer to check AI-generated insights with sales reps. S3

7%

Of 300 B2B companies call AI SDR campaigns core. Respondents rated the odds of a return very poor. S2

40%+

Of agentic AI projects will be cancelled by the end of 2027, Gartner forecasts. S4

Get the PDF.

The whole playbook as a 32-page PDF, free. Share your work email and download it at once.

We use your details to give you the download. See our privacy policy.

What this playbook gives you.

A Claude Code setup that researches each company, writes one message and checks it. A person approves before anything goes out.

9

Steps from a company name to an approved draft. A person approves at step 7.

69%

Of B2B buyers prefer to check AI-generated insights with sales reps (S3).

7%

Of 300 B2B companies call AI SDR campaigns core. Respondents rated the odds of a return very poor (S2).

40%+

Of agentic AI projects will be cancelled by the end of 2027, Gartner forecasts (S4).

Where it comes from.

Written up in November 2026 from the sales system template we offer as a two-week build. The setup ideas credit Charlie Hills (MarTech AI) and Anthropic’s Claude Code docs.

What you get.

The folder layout, nine steps, one worked example on a made-up company, costs, our mistakes and twelve templates.

How to read the tags.

S1 means source 1. Sources are listed at the back. Times are our estimates.

Contents.

Read Part 1 for the why. Build Part 2 once. Run the nine steps in Part 3 every day, and use Part 4 to plan and budget.

  1. Part 1. Why a person stays in charge
  2. 01What this builds, in one pictureoverview4
  3. 02Who it is forfit4
  4. 03What the data says about AI alonedata5
  5. Part 2. Build the setup
  6. 04The harness: rules, memory, helpersbuild7
  7. 05The loop: one batch, then stopbuild8
  8. 06Connect your tools by jobbuild9
  9. Part 3. The nine steps
  10. 07Step 1. Fit checkdaily11
  11. 08Steps 2 and 3. Research, then a second sourcedaily11
  12. 09Step 4. No reason, no messagedaily13
  13. 10Steps 5 and 6. Draft, then check twicedaily13
  14. 11Step 7. A person sends itdaily15
  15. 12Step 8. Replies and stuck dealsas needed16
  16. 13Step 9. Keep in touch, report on Fridayweekly17
  17. Part 4. Run it
  18. 14What it costsmoney19
  19. 15Mistakes we made and fixedours20
  20. 16Your two-week setup planplan21
  21. 17Twelve templatescopy21
  22. Reference
  23. 18Sources and credits31

Where it comes from.

Written up in November 2026 from the sales system template we offer as a two-week build. The setup ideas credit Charlie Hills (MarTech AI) and Anthropic’s Claude Code docs.

What you get.

The folder layout, nine steps, one worked example on a made-up company, costs, our mistakes and twelve templates.

How to read the tags.

S1 means source 1. Sources are listed at the back. Times are our estimates.

Part 1

Why a person stays in charge.

What the setup does, who it suits, and what the data says. Read this before you build anything.

  1. 01What this builds, in one pictureoverview
  2. 02Who it is forfit
  3. 03What the data says about AI alonedata
Chapter 01

What this builds, in one picture.

You give it a company name. It gives you back a checked first message and the research behind it. A person on your team decides if it goes out.

From a company name to an approved message.
Nine steps. The AI prepares. A person approves at step 7 and sends every reply.
The AI prepares and checks You decide After 1 Fit check 2 Research 3 Second source 4 Reason to write? 5 Draft one message 6 Two checks: script, gate 7 A person approves 8 Replies 9 Nurture, report No dated reason?Wait, keep in touch. Set up to waitfor a person's yes. 1 Fit check 2 Research, 2 years of signals 3 Second source for each fact 4 Reason to write? If not, wait 5 Draft one message 6 Script check, then a checker 7 A person approves and sends 8 Replies and stuck deals 9 Keep in touch, Friday report
Our own diagram of the setup described in this playbook.

What you end up with

A folder on your computer that you open in Claude Code. It holds a short rules file, four files about your business, your contact list, a state file and a few helpers. Each run takes the next five contacts, works them and stops.

A queue file each day. Every entry shows the person, the reason to write with its date and source, the draft and its check score. You approve, edit or reject each one. The setup is built so this takes 10 to 15 minutes a day.

A Friday report. Replies and meetings first, then what was sent, then one to three changes for next week.

Chapter 02

Who it is for.

It suits teams that care more about the quality of each first message than the count. Check yourself against both lists before you start.

A good fit

You already sell, and you know who buys.

B2B founders and heads of sales with a clear buyer, a list or CRM of a few hundred to a few thousand accounts, and deals worth a researched message. You have 10 to 15 minutes a day to approve drafts.

Not a fit yet

You need volume, or you have no buyer profile.

If you want thousands of sends a week, this is the wrong tool. If you cannot name who buys and why, start with your last 20 wins. Our 2027 GTM playbook, Play 1, shows how.

What you need on day one

  • A Claude plan that includes Claude Code. See chapter 14 for prices.
  • A computer you can install it on. Mac, Windows or Linux all work.
  • An export of your contacts as a CSV file. Name, company and a website or LinkedIn link are the minimum.
  • Your own past writing: ten messages or emails you sent that got replies. The setup learns your voice from them.
  • About two hours across the first week to answer setup questions. Chapter 16 has the plan.

You do not need to code. You type plain requests into Claude Code, and it works with the files.

Chapter 03

What the data says about AI alone.

Two surveys point one way. In the GTM report, respondents rated the odds of a return from AI SDRs very poor at every company size. And most B2B buyers prefer to check AI-generated insights with a sales rep. Both point to a person who checks the work.

300 B2B companies
7%

Call AI SDR campaigns a core channel. Teams rated their odds of a return very poor at every company size. S2 p.16, p.23

B2B buyers
69%

Turn to sales reps to check AI-generated insights. S3

Forecast
40%+

Of agentic AI projects will be cancelled by the end of 2027, Gartner predicts. S4

What sales teams report

The 2026 State of B2B GTM Report asked 300 B2B software and AI companies about their channels. Agentic campaigns, often sold as AI SDRs, sat in the “figuring it out” group. Teams rated their odds of a return very poor at every size. Only 7% called them core, down 4 points on the year. S2 p.16, p.17, p.19, p.23

What buyers report

Gartner surveyed 645 B2B buyers. 69% said they prefer to check AI-generated insights with sales reps. S3 A message from a person who has done the reading helps with that.

What analysts expect

Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027. It published that forecast in June 2025. S4

The rule this playbook follows

So the AI does the reading, the checking and the first draft. A person decides who gets a message and what it says.

Part 2

Build the setup.

Three parts you build once. The folder and its rules, the loop that runs each batch, and the tools it may use.

  1. 04The harness: rules, memory, helpersbuild
  2. 05The loop: one batch, then stopbuild
  3. 06Connect your tools by jobbuild
Chapter 04

The harness: rules, memory, helpers.

Charlie Hills puts it this way: “The model is the engine. The harness is the car.” The harness is the folder around the model: its rules, memory and helpers. Switch models later and the folder stays. S10

The folder

This is the layout of the template. Every name below is a plain file or folder you can open and read.

sales-system/
  CLAUDE.md            the rules Claude reads first
  brain/
    company.md         who you are, who you turn away
    customer.md        who buys, their words, signs they need you now
    offer.md           what you sell, proof you may quote
    voice.md           how you write, words you never use
  data/
    contacts.csv       your contact list
    do-not-contact.csv
  research/accounts/   one short file per company
  messaging/queue/     drafts waiting for a person
  state/               loop-state.json and a short run log
  deliverables/reports/ the Friday report
  scripts/             free checks that run before any AI check
  .claude/agents/      the four helpers
  .claude/skills/      the five jobs
1. Rules

One short CLAUDE.md.

Claude Code reads it at the start of every session. S6 Keep it under a page, with the hard rules at the top. Claude reads it as guidance, so it can still slip. S1 The rule that nothing sends also goes into permissions and a hook. Template I in chapter 17 shows how.

2. Memory

Four brain files and two state files.

The brain files hold facts about your business, each with its source. state/loop-state.json says where the last run stopped, so the next run picks up from there. state/LOOP-STATE.md is the short run log.

3. Helpers

One helper per job.

Each helper is a subagent with its own context window and its own model. S7 A small, fast model does the simple jobs. A stronger one writes and checks.

The six hard rules in the template

  1. Nothing is sent, posted, connected or booked without a named person’s OK on the exact text.
  2. Never invent a fact, a price, a result, a client name or a quote.
  3. Facts about your business come only from the brain files. If a fact is missing, ask.
  4. The AI is set up so it never opens LinkedIn. If a LinkedIn page, captcha or login prompt ever appears, stop and write it in state/LOOP-STATE.md.
  5. Never type or store passwords or two-factor codes.
  6. Never contact anyone on the do-not-contact list, a current client or a competitor.

Build the brain from evidence

Charlie Hills has published a set of four prompts, one for each brain file. S14 The template uses the same four files. Charlie’s guide uses your about page, recent posts and past proposals. Our version: feed them your website, three call notes, your proposal and ten messages that got replies. Every line in a brain file should point to where it came from. If a line has no source, delete it.

HelperJobModel in the templateWhy
ResearcherFit check, find signals, fetch pagesSmall and fast (Haiku)Reading and sorting need speed more than style
WriterDrafts, replies, nurture commentsStrong (Sonnet; Opus for warm replies)Writing in a real person’s voice is hard
GateScores each draft coldStrong (Sonnet), separate contextIt never sees how the draft was written
ReporterCounts and the Friday reportSmall and fast (Haiku)Counting needs no judgement

Why a separate gate? A writer checking its own draft knows what it meant to say, so it reads past the gaps. The gate reads only the draft, the research file, your voice file and your offer file. It never saw the drafting, so it judges the draft cold.

Chapter 05

The loop: one batch, then stop.

Gather, act, check, record, stop. Each run does one small batch in a fresh session and writes down where it stopped. The frame credits Charlie Hills’ loop guides. S11

The loop

What one run does.

You type /sales-loop. It works five contacts by default, then stops and tells you where the drafts are.

  1. Ready check. Brain files built? Contact list there? Required fields filled? Blocked by a warning? It stops at the first gap and says what is missing.
  2. Pick the batch. A script picks the next five rows marked new or due for nurture. It skips the do-not-contact list.
  3. Fit check. The researcher marks each row fit, maybe or no fit. No-fit rows get a reason and are never researched.
  4. Research. Fit and maybe rows only. Skip any company researched in the last 30 days.
  5. Draft and check. The writer drafts one message per person. A script checks it, then the gate scores it.
  6. Nurture plan. For people already in touch, from posts a person pastes in: posts to like, one comment worth leaving, pages to follow.
  7. Record and stop. Write the queue file, update each row, update loop-state.json, add to the run log. One line back to you.
Tool
Claude Code, the folder, two small scripts
Time
A few minutes of yours per run, to read the queue
Done when
The queue file has a draft or a skip reason for every row in the batch

The run log (state/LOOP-STATE.md)

Three lines per run, newest at the bottom. The next run reads this first. That is how it remembers without loading every past chat.

2026-11-12 run 7: 5 rows. 3 drafts PASSED, 1 NEEDS A PERSON, 1 skip (competitor).
Waiting: messaging/queue/2026-11-12.md
Next: rows 36 to 40. Two nurture comments due on Friday.

Example lines on made-up data.

Rules that keep it cheap

  • Scripts first, models second. Counting rows and checking banned words cost nothing in a script.
  • Small batches. Five contacts a run. A fresh session each time, so old chat does not pile up.
  • Read only what a step needs. Never load the whole contact list into the chat.
  • Cache research for 30 days. Do not pay to read the same company twice in a month.
  • Keep long procedures in skills. A skill’s full text loads only when it is used, unlike CLAUDE.md. S9
Chapter 06

Connect your tools by job.

Pick tools by the job they do, then connect as few as you can. Claude Code connects to outside tools through MCP servers. S8 Charlie Hills’ rule is three to five live connectors, used weekly. S12

JobExamples (pick one)Start with
ResearchClaude Code’s built-in web search and fetch, Exa, FirecrawlBuilt-in search. Add a search connector when volume grows
Contact dataApollo, Prospeo, Clay, HunterOne tool on its smallest plan, read-only
CRMHubSpot, Attio, PipedriveYour own CRM. Or a CSV export
EmailGmail, Smartlead, InstantlyDrafts only for the first two weeks
LinkedInNo connector. A person reads Sales NavigatorA person reads and sends on LinkedIn. No bulk-send tools
CalendarCalendly, Cal.comA booking link
Call notesFathom, Granola, GongPaste notes into a meetings folder
ReportingA file report, your CRM’s reports, NotionA file report

Claude Code loads most tool details only when a tool is needed, so extra servers cost little context. S8 We still keep the list short, for fewer logins to protect.

Four safety lines. Start every connector read-only, and add write access one tool at a time. Give each tool the narrowest scope it offers. Never hand your LinkedIn login to a scraper or a cloud tool. It is built so that nothing sends without a person (chapter 11).

Three chains in the template. Research to queue: list, fit check, research, draft, checks, approval queue. Reply to meeting: the writer drafts, a person sends. Friday report: one short report.

Part 3

The nine steps.

One made-up company runs through the steps. Northwind Advisory is an IT consultancy in Lisbon. The seller is Example Training Co, which trains new sales reps. Neither company is real.

  1. 07Fit checkstep 1
  2. 08Research, then a second sourcesteps 2, 3
  3. 09No reason, no messagestep 4
  4. 10Draft, then check twicesteps 5, 6
  5. 11A person sends itstep 7
  6. 12Replies and stuck dealsstep 8
  7. 13Keep in touch, report on Fridaystep 9
Chapter 07

Step 1. Fit check.

Decide who not to research before you spend anything. Fit, maybe or no fit, against your customer file. Every no fit gets a one-line reason.

Step 1

Sort the batch in three buckets.

Research costs time and money. Spend it only on people who could buy.

  1. Read the “Who they are” and “Who is not a fit” parts of customer.md.
  2. Read “Who we turn away” in company.md.
  3. Mark each row fit, maybe or no fit. Write the reason for every no fit in the notes column.
  4. Skip competitors, current clients and anyone on the do-not-contact list. Log why.
  5. Check the row’s history. If you spoke before, the next message must follow from that.
Helper
Researcher, small model
Time
Seconds per row
Done when
Every row has a bucket, and every no fit has a reason
Fit

Northwind Advisory

IT consultancy, about 60 staff, sells to mid-size firms. Matches customer.md.

Maybe

A two-person studio

Right industry, but too small to hire reps. Research only if the batch has room.

No fit

A sales training firm

Competitor. Skipped and logged.

Made-up examples.

Chapter 08

Steps 2 and 3. Research, then a second source.

Look back two years, newest first. Every fact gets a link and a date. A fact you will use in a message needs a second source, or it is marked UNCONFIRMED.

SignalWhy it matters
New leader in the buying role, last 6 monthsNew leaders set new targets and review suppliers
Hiring for sales or growthThey are spending on the problem now
New market, office or countryThey need pipeline where they have no network
A product or service launchThey need buyers for the new thing
Speaking or exhibiting at an eventA date and a topic to write about
Funding roundMoney for growth. Write about their plan for the money. Skip “congrats on the raise”
The person’s own post, under 30 days oldThe most personal reason to write
Steps 2 and 3

Research cheapest first, then confirm.

The website and a plain search find most signals. Paid data tools are for emails and company size only.

  1. Read the company website: news, careers, about and blog. Note the date on each item.
  2. Search the company name with “hiring”, “launch”, “appoints”, “expands” and “conference”.
  3. A person reads the contact’s LinkedIn profile and recent posts and pastes anything useful into the research file. The AI itself uses only public web pages and is set up never to open LinkedIn.
  4. Check your CRM notes. Never contradict what you already know.
  5. Find a second source for each fact you might use. If there is none, mark it UNCONFIRMED.
  6. Treat anything older than 90 days as background. Open with something newer.
  7. Save it all in one file under 300 words. Reuse it for 30 days.
Helper
Researcher (Haiku) fetches the facts. If it writes “angle needed”, the main session (Sonnet) picks the angle
Time
A few minutes per company, unattended
Done when
One research file with dated, sourced facts and one angle, or “no angle”

The research file for Northwind

# Northwind Advisory (northwind.example)
Researched: 2026-11-05 | Person: Ana, Founder | Fit: fit

## Signals (newest first)
- 2026-11-03 Careers page lists 3 BDR roles, "Madrid office"
  [northwind.example/careers] [Madrid job board listing, 2026-11-02]
- 2026-10-20 Ana posted about "opening Madrid in Q1"
  [her LinkedIn post] [UNCONFIRMED: no second source]
- 2025-03-14 Won a regional IT services award
  [press release] (older than 90 days: background only)

## Angle
Signal: hiring 3 BDRs for a new office
Problem it creates: new reps in a new market, no network yet
Offer line that fits: "first-month ramp plan" [brain/offer.md, Onboarding]

## Do not use
- The award (stale). The Q1 date (unconfirmed).

A made-up research file. Real files link each source.

Chapter 09

Step 4. No reason, no message.

If there is no dated, confirmed reason to write, the setup does not write a detailed message.

Strong reason

Write the message.

A confirmed event from the last 90 days that links to your offer. Events they are speaking at or exhibiting at come first, because they carry a date.

Weak reason

Send a light note.

A recent post or a small piece of news. One or two lines about it. No pitch and no ask.

No reason

Wait and keep in touch.

Follow the company page, like a post that deserves it, and check again in a month.

Why we hold back

A message with no real reason could go to anyone. Swap the name and nothing else changes. In our experience, buyers spot that in a line. So when there is no reason, the setup waits a month and looks again.

Northwind passes. The BDR hiring is dated 3 November and has two sources.

Chapter 10

Steps 5 and 6. Draft, then check twice.

One fact, what it likely means for them, one easy question. Then a free script checks the draft, and a separate checker scores it.

Step 5

Write the first message.

Three to five short lines. No pitch, no price, no link and no meeting ask in a first message.

  1. Line 1: one real, dated fact about them, from the research file.
  2. Line 2: what that fact likely means for them, in their words from customer.md.
  3. Line 3: one easy question about their situation that they can answer in a line.
  4. Match your voice file: line length, greeting and sign-off.
Helper
Writer, strong model
Time
Under a minute per draft
Done when
The draft uses only facts from the research file

Before and after, with the real script output

We ran both drafts through the free script in our template on 17 November 2026. This is what it printed.

Before
Hi Ana, I hope this finds you well! I noticed that you are expanding to Madrid (long dash here) congratulations!
At Example Training Co we help companies like yours leverage cutting-edge sales training to unlock growth.
Would you be open to a quick call next week to explore synergies? Let me know if you're interested!
FAIL
 - long dash
 - banned phrase: i hope this finds you well
 - banned phrase: i noticed that you
 - banned phrase: leverage
 - banned phrase: unlock
 - banned phrase: cutting-edge
 - banned phrase: let me know if you're interested
 - more than one exclamation mark
After
Hi Ana, I saw Northwind is hiring three BDRs for the new Madrid office.
New reps in a new market usually take a while to book their first meetings.
How are you planning their first month?
PASS

The before draft really used a long dash. We wrote “(long dash here)” because this playbook never prints one.

Step 6

Check twice: a script, then a cold reader.

Run the script first, then the gate. The script finds banned words and long sentences in seconds. The gate reads for sense, facts and tone.

  1. The script. It checks length, sentence length, question count, links in a first message, your never-list words, leftover brackets like [Name] or {{first_name}}, long dashes and stock AI phrases. The starter version in template H also fails activity numbers used as proof. Any FAIL goes back to the writer.
  2. The gate. A separate helper reads only the draft, the research file, voice.md and offer.md. It scores 0 to 10, two points each:
    • Every fact is in the research file with a date and a source.
    • It is written for this person. If the name could be swapped, this scores 0.
    • It sounds like the voice file.
    • It asks one thing a busy buyer can answer in a line, or nothing.
    • It promises nothing the offer file does not say.
  3. 8 or more passes to the queue. Below 8, the writer gets one rewrite with the gate’s notes.
  4. A second fail goes to the queue marked NEEDS A PERSON, with both drafts.
Helpers
A Python script, then the gate on a strong model
Time
Seconds
Done when
Every draft in the queue shows PASSED with a score, or NEEDS A PERSON

The 10 human checks we add on our own drafts

On top of the script and the gate, we read every draft against ten plain checks. They come from a “Humanizer” prompt shared on LinkedIn by Noam Nisand, which we read in October 2026. The wording here is ours. S15

  1. Cut sentences about importance, impact or broader trends.
  2. Cut AI words such as pivotal, crucial, landscape, foster and delve.
  3. Cut contrast lines that set up one idea only to knock it down.
  4. Cut vague authority like “experts say”. Name the source or drop the line.
  5. Cut neat summary endings and one-line morals.
  6. Swap general claims for a specific how, number or name.
  7. Split any sentence over 16 words.
  8. If a sentence could fit a thousand other topics, delete it.
  9. No motivational or preaching tone.
  10. It should sound like a message to one person. Read it aloud as the sender.
Chapter 11

Step 7. A person sends it.

The setup prepares and queues. It is set up never to send. A person reads each draft, edits or rejects it, and then sends it.

The queue entry a person sees

## Ana, Founder, Northwind Advisory | LinkedIn DM | PASSED 10/10
Why now: hiring 3 BDRs for Madrid, 2026-11-03 [research/accounts/northwind.md]
Draft:
Hi Ana, I saw Northwind is hiring three BDRs for the new Madrid office.
New reps in a new market usually take a while to book their first meetings.
How are you planning their first month?
Approve / edit / reject:

How sending works

LinkedIn’s User Agreement bans bots and unauthorized automated tools for adding contacts, sending messages or fake engagement. It also bans browser add-ons that scrape profiles. S5 Every message is written for one person and approved by a person before it goes. We never run bulk-send tools on a founder’s profile.

Back the send rule with a setting. Anthropic’s docs say Claude reads CLAUDE.md as context, and that a hook can block an action whatever Claude decides. S1 So deny the send and post tools in Claude Code’s permissions, and add a hook that blocks them too. Template I in chapter 17 shows both. On LinkedIn, a person reads and sends. The AI works only from public web pages.

Email

Send through your own tool or mailbox, no more than 20 a day at first. Keep drafts-only mode for the first two weeks, until the drafts need few edits. Our 2027 GTM playbook, Play 13, covers sending domains and inbox rules.

Chapter 12

Step 8. Replies and stuck deals.

Answer exactly what they asked, first. Then take one step forward. For a deal that has stalled, write a short read before you write the message.

Step 8

The reply mode for a stuck deal.

Adapted from Charlie Hills’ free Machiavelli skill. The method and its steps are his. We changed the names of some steps and shortened them. S13

  1. The read. One line: what the silence most likely means.
  2. Their position. His power map, in our short form: what they want, what worries them, what else they could do, how soon they must decide and who signs off.
  3. The mode. Patient (give them room and add value) or direct (name the decision and the date). One line on why. Patient or direct is our version of his fox or lion choice.
  4. The moves. Rank two or three options, and note the price of each for you.
  5. The message. Short, ready to paste. It ends on a question they can answer easily. Our addition: leave numbers for the call.
Helper
Writer, strongest model for warm leads
Time
A few minutes, plus your read
Done when
A person has sent the reply, or decided to wait
Northwind, three weeks after a callThe read
What they wantThree new reps booking meetings in Madrid before Q2
What worries themThe founder’s time, already stretched by the new office
What else they could doTrain the reps themselves with what they have
Their deadlineReps start in January
Who signs offTheir co-founder, who controls the budget
ModePatient. Nothing is lost by waiting a week, and the start date is fixed
Hi Ana, thinking about your January start.
I put the first-month plan for new reps on one page, so your
co-founder can read it in two minutes.
Want me to send it over?

A made-up deal and reply.

Chapter 13

Step 9. Keep in touch, report on Friday.

Most people are not ready the first time you write. Stay in view without asking for anything, and look at the numbers once a week.

Keep in touch

Keep it light and specific.

Follow the company page. Like a post that deserves it. Leave one specific comment when a post is worth it, with no pitch. After a “not now”, check in once, four to six weeks later, with something new for them.

Friday report

Results first.

Replies and meetings at the top. Then what was sent. Then one to three changes for next week, each tied to a number. Keep it to one screen.

What not to report as proof

Invites sent, acceptance rates and emails sent measure effort. Use them to spot problems. Judge the week on replies, meetings and deals.

The Friday report for the example week

# Week of 2026-11-09 (made-up example)
Results: 3 replies, 1 meeting booked (Northwind, Thu 14:00)
Sent: 12 first messages, 4 follow-ups, 7 skipped (no reason yet)
What worked: messages that opened on a hiring signal got 2 of 3 replies
What did not: event openers older than 60 days got no reply
Changes for next week:
1. Open on hiring or new-leader signals first, because 2 of 3 replies
2. Drop event openers older than 60 days, because 0 replies
Waiting on you: 5 drafts in messaging/queue/2026-11-13.md

A made-up report. Your numbers will differ.

Part 4

Run it.

What it costs, what went wrong for us, a two-week plan and twelve templates. Everything you need to start on Monday.

  1. 14What it costsmoney
  2. 15Mistakes we made and fixedours
  3. 16Your two-week setup planplan
  4. 17Twelve templatescopy
Chapter 14

What it costs.

You pay your tool vendors directly. List prices in USD, monthly billing, read on 17 November 2026. Data tools change their prices often, so we leave them out of the totals. Check each price at checkout. S16

LeanStandardFull
Who it suitsA founder who wants research and drafts, and sends from LinkedIn with no paid toolsLinkedIn and email for one personSeveral channels, with data enrichment
ClaudePro, $20Max ($100, 5x Pro usage)Max ($100, 5x Pro usage), plus API use, about $130 (our estimate)
LinkedInNone, $0Sales Navigator Core, about $120Sales Navigator Core, about $120
Contact dataFree tiers, $0One data tool on its smallest plan, priced separatelyA data tool plus an enrichment tool, priced separately
EmailNone, $0Smartlead Base, $39, plus 3 Google Workspace mailboxes at $8.40, about $25Instantly Growth, $47, plus 5 mailboxes at $8.40, $42
CRMHubSpot Free, $0HubSpot Free, $0Attio Plus, $50 (1 user)
About per month$20About $285, plus your data toolAbout $490, plus data tools

Why Full pays for both: the Max plan covers the hours you work with Claude Code yourself, and the API covers helpers that run on a schedule when nobody is at the keyboard.

Costs that are not on any price page

  • Your approval time. Plan for 10 to 15 minutes a day.
  • New email domains. Plan two to four weeks of mailbox warm-up before cold sending. In our experience, month one books little.
  • LinkedIn account risk. LinkedIn’s rules do not allow tools that automate activity on its site, and accounts can be restricted. S5 In this setup a person reads and sends everything on LinkedIn. The AI works only from public web pages, so it is set up never to touch the account.
  • Data credits. In our experience, most data tools reset credits each month with no rollover. Start on the smallest plan.
  • Email checks. A one-off ZeroBounce pack costs about $39 for 2,000 checks. Buy one before each new list goes out.
Chapter 15

Mistakes we made and fixed.

Four rules here came from our own mistakes.

October 2026

We wrote about old news.

Invite notes opened on news two to four months old. Fix: the opening line uses the newest item, ideally from the last four weeks.

October 2026

We drafted from one post.

First messages came from a single LinkedIn post, with no fit check. Fix: fit check, two years of signals and a second source come before any draft.

October 2026

We used effort as proof.

A follow-up quoted invites sent as if they were results, to 19 people. Fix: proof is the buyer’s result only, and the starter copy check in template H now fails activity numbers.

November 2026

Our inbox check covered too little time.

It covered only the last 24 hours and missed a labelled email. Fix: every check covers each open thread and the last seven days, in every inbox.

Chapter 16

Your two-week setup plan.

Week one builds the folder and tests the research. Week two adds drafting and the checks.

DayWhat you do (days 1 to 5 are week one)Done when
1Install Claude Code. Create the folder. Write CLAUDE.md with the six hard rulesClaude Code opens the folder and repeats the rules back
2Build voice.md and customer.md from your own messages and call notesEvery line has a source
3Build offer.md and company.md. Add the proof you may quoteFour brain files, no line without a source
4Export contacts to contacts.csv. Fill the do-not-contact listThe check script reports no missing required fields
5Run research on 10 accounts you know wellYou agree with 8 of the 10 angles
6Add the four helpers and the copy check scriptThe before draft in chapter 10 fails, the after draft passes
7Run the first full batch of five, drafts onlyFive entries in the queue, each with a score
8Connect your CRM and email, read-onlyA test run reads a deal and a thread without changing them
9Run two more batches. Note every edit you makeYour edits get smaller from batch to batch
10Write the first Friday report. Decide what to send next weekA one-screen report and a go or no-go

Prefer to build it with someone? We offer a two-week build of this system for B2B teams, handed over to you at the end. Message Saif on LinkedIn.

Chapter 17

Twelve templates.

Copy these into your own folder and edit them. The names in angle brackets are for you to fill in. Copy the code from the web version, fableinlead.com/playbooks/ai-sales-automation, because a PDF drops the spaces at the start of each line.

A. CLAUDE.md starter

# Sales system: rules Claude reads first

## HARD rules (Claude reads these first. Back rule 1
## with a permission and a hook, see template I.)
1. Nothing is sent, posted, connected or booked without a named
   person's OK on the exact text.
2. Never invent a fact, price, result, client name or quote.
   Every prospect fact needs a source link and a date.
3. Facts about our business come only from brain/. If missing, ask.
4. The AI is set up never to open LinkedIn. If a LinkedIn page,
   captcha or login prompt ever appears, stop and write it in
   state/LOOP-STATE.md.
5. Never type or store passwords or two-factor codes.
6. Never contact anyone on data/do-not-contact.csv, a current
   client or a competitor.

## How to run
/sales-loop runs one batch of 5, checks its work, writes the state
file and stops. Start each run in a fresh session.

B. Research file

# <Company> (<website>)
Researched: <date> | Person: <name>, <title> | Fit: fit / maybe
## Signals (newest first)
- <date> <fact> [source] [second source or UNCONFIRMED]
## Angle
Signal: ... | Problem it creates: ... | Offer line: ... [offer.md heading]
## Do not use
- anything stale, unconfirmed, private or awkward

C. Gate helper (.claude/agents/gate.md)

---
name: gate
description: Scores a finished draft 0 to 10 without seeing how it
  was written. Use after the copy check, before the queue.
model: sonnet
tools: Read
---
You did not write this draft. Judge it cold.
Read only: the draft, its research file, brain/voice.md, brain/offer.md.
Two points each:
1. Every fact is in the research file with a date and a source.
2. It is written for this person. Swappable name = 0 here.
3. It matches the voice file.
4. It asks one thing a busy buyer can answer in a line, or none.
5. It promises nothing offer.md does not state.
Reply only: SCORE n/10, FAILS, FIX (one line per fail).
Pass is 8 or more. Do not rewrite the draft.

D. Queue entry

## <Name>, <title>, <company> | <channel> | PASSED <n>/10
Why now: <signal, date> [research file]
Draft:
<message>
Approve / edit / reject:

E. Friday report

# Week of <date>
Results: <n> replies, <n> meetings booked, <n> deals moved
Sent: <n> first messages, <n> follow-ups, <n> skipped (reasons in data)
What worked: <one line, with a number>
What did not: <one line, with a number>
Changes for next week:
1. <change> because <number>
2. <change> because <number>

F. The sales loop skill (.claude/skills/sales-loop/SKILL.md)

Type /sales-loop in Claude Code to run it. This is a shortened version of the template file.

---
name: sales-loop
description: Runs one batch: ready check, pick 5 contacts, fit
  check, research, draft and gate, nurture plan, record, stop.
---
# One batch, then stop. Gather, act, check, record, stop.

0. Ready check. Read state/loop-state.json. Stop with one short
   message at the first gap: brain files not built, no
   data/contacts.csv, more than 20% of rows missing a required
   field, or "blocked": true.
1. Pick the batch. python3 scripts/check_contacts.py
   data/contacts.csv --next 5 returns
   rows with status new or nurture_due, minus do-not-contact.
2. Fit check (researcher). Fit, maybe or no fit against
   brain/customer.md and brain/company.md. No fit: one-line
   reason, status skip, no research.
3. Research (researcher). Fit and maybe rows only. Skip any
   company researched in the last 30 days.
4. Draft and gate (writer, then gate). One message per contact.
   Writer runs on sonnet; use opus for replies to warm leads.
   A gate fail goes back once. A second fail goes to the queue
   marked NEEDS A PERSON.
5. Nurture plan (writer). Posts to like, one comment draft if a
   post is worth it, pages to follow. Same queue file.
6. Record and stop. Write messaging/queue/<date>.md, update
   each row's status and next date, update state/loop-state.json,
   add 3 lines to state/LOOP-STATE.md, reply in one line.

Hard stop: if a LinkedIn page, captcha or login prompt appears,
set blocked: true and end the run. This skill is set up never to send.

G. The other three helpers (.claude/agents/)

The gate helper is template C. Each helper is a short file with its own model.

--- .claude/agents/researcher.md ---
---
name: researcher
description: Fit-checks contacts and gathers dated company signals.
model: haiku
tools: Read, Grep, Glob, WebSearch, WebFetch, Write
---
You check fit and gather facts. You do not write messages.
Every fact gets a date and a source link. A fact without a second
source is marked UNCONFIRMED. Never guess a date, name or number.
Write only to research/accounts/. If the angle needs judgement,
say "angle needed" and stop. The main session
(Sonnet) then picks the angle.
--- .claude/agents/writer.md ---
---
name: writer
description: Writes messages, follow-ups, replies and comments in
  the owner's voice from a research file.
model: sonnet
tools: Read, Write, Bash
---
Read brain/voice.md first. Use only facts in the research file and
claims in brain/offer.md. Run scripts/copy_gate.py on your draft
and fix every FAIL. Use Bash only for that one script. One draft per contact, one question at most.
You never send anything.
--- .claude/agents/reporter.md ---
---
name: reporter
description: Counts results from the sales files and writes the
  Friday report.
model: haiku
tools: Read, Grep, Glob, Write
---
Count from files only. No estimates, no research, no messages.
Write the report to deliverables/reports/ and reply in one line.

If you add a research connector such as Exa or Firecrawl, also add its tool names to the researcher’s tools line, for example mcp__exa__web_search_exa. A helper can only use the tools listed there.

HelperModelWrites toNever does
researcherhaikuresearch/accounts/Writes a message or guesses a date
writersonnetmessaging/queue/Sends, or runs any Bash command other than the copy check
gate (template C)sonnetIts verdict onlyEdits the draft it is judging
reporterhaikudeliverables/reports/Estimates a number it cannot count

Check each file against this table before the first run. If a helper can write somewhere the table does not list, take Write and Edit out of its tools line. The tools line works on whole tools, so it cannot limit a helper to one folder.

H. Copy check script (scripts/copy_gate.py)

A starter version that runs for free before any AI check. It prints shorter labels than the full script in chapter 10, and has fewer banned phrases. It also fails empty drafts, mail-merge placeholders such as {{first_name}}, and activity numbers used as proof. Add your own banned words from voice.md.

#!/usr/bin/env python3
"""Free first check for any draft. Runs before any AI check.
   python3 scripts/copy_gate.py draft.txt --channel dm"""
import argparse, re, sys
STOCK = ["i hope this finds you well", "just following up",
         "just checking in", "touching base", "quick question",
         "game-changer", "leverage", "synergy", "seamless"]
LIMITS = {"note": (4, 200, 0), "dm": (6, 600, 1),
          "email": (8, 900, 1), "comment": (3, 350, 1)}
# Effort counted as proof, e.g. "we sent 7,000 invites".
ACTIVITY = r"\b\d[\d,.]*\+?\s*(?:k\s+)?(?:messages|emails|invites|invitations|connections|calls|dials|touches|dms|follow-?ups|sequences|requests)\b"
ap = argparse.ArgumentParser()
ap.add_argument("draft"); ap.add_argument("--channel", default="dm", choices=list(LIMITS))
a = ap.parse_args()
text = open(a.draft, encoding="utf-8").read().strip()
max_lines, max_chars, max_q = LIMITS[a.channel]
fails = [] if text else ["empty draft"]
lines = [l for l in text.splitlines() if l.strip()]
if len(lines) > max_lines: fails.append(f"{len(lines)} lines")
if len(text) > max_chars: fails.append(f"{len(text)} characters")
if text.count("?") > max_q: fails.append("too many questions")
if re.search(r"https?://|www\.|\b[\w-]+\.(com|net|org|io|co|ai|app|example)\b", text, re.I): fails.append("link")
if "\u2014" in text or "\u2013" in text: fails.append("long dash")
if re.search(r"\{\{[^}]*\}\}|\{[^}]*\}|<[^>]*>|\[[^\]]*\]", text): fails.append("placeholder")
if re.search(ACTIVITY, text, re.I): fails.append("activity number")
for s in re.split(r"(?<=[.!?])\s+|\n", text):
    if len(s.split()) > 16: fails.append(f"long sentence: {s[:40]}")
low = text.lower()
fails += [f"banned: {w}" for w in STOCK if w in low]
print("PASS" if not fails else "FAIL")
for f in fails: print(" -", f)
sys.exit(1 if fails else 0)

I. The send lock (.claude/settings.json and a hook)

CLAUDE.md asks Claude not to send. This file makes sending much harder. Anthropic’s permissions page puts it plainly: “Permission rules are enforced by Claude Code, not by the model.” A PreToolUse hook can also block a call whatever Claude decides. S17 S18

Do this first. During the drafts-only weeks, deny your bulk-send tools in full, as the first two entries below do. Run /mcp to see the exact tool names your connectors use.

The hook works the other way round from a block list. A connector tool runs only if its first word is get, list, search, read or find, and no later word is on the write list. Any other name stops until you add it to EXTRA_ALLOW, so a new tool is blocked by default. Any call whose input names linkedin.com stops too, even for a tool on EXTRA_ALLOW, and so does an action name with a second __ inside it. Research tools often have other names, such as web_search_exa or firecrawl_scrape. Add the ones you use to EXTRA_ALLOW on day one, or the research step stops. Add the same names to the researcher’s tools line in template G, for example mcp__exa__web_search_exa, because a helper can only use the tools listed there. “onFailure”: “block” stops the call if the hook itself fails to run. It needs Claude Code v2.1.295 or later. S18 One limit stays: a Bash deny rule for curl does not catch the same program run by its full path or inside sh -c. Turn on the sandbox network allowlist when the block must hold. S17

{
  "permissions": {
    "deny": [
      "mcp__smartlead__*", "mcp__instantly__*",
      "mcp__claude-in-chrome__*",
      "WebFetch(domain:linkedin.com)", "WebFetch(domain:*.linkedin.com)",
      "Bash(curl *)", "Bash(wget *)"
    ],
    "allow": [ "Bash(python3 scripts/copy_gate.py *)" ]
  },
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "mcp__.*",
        "hooks": [
          { "type": "command",
            "command": "\"${CLAUDE_PROJECT_DIR}\"/.claude/hooks/block-send.sh",
            "onFailure": "block" }
        ]
      }
    ]
  }
}
#!/bin/bash
# .claude/hooks/block-send.sh  (chmod +x, needs jq)
# A connector tool runs only if its first word is a read verb
# (get, list, search, read, find) and no later word is on the
# write list below. Any call that names linkedin.com stops.
# Research tools often have other names. Add yours, for example:
# EXTRA_ALLOW="web_search_exa web_fetch_exa firecrawl_scrape
#   firecrawl_search people_search domain_search"
EXTRA_ALLOW=""
IN=$(cat) || exit 2
TOOL=$(printf '%s' "$IN" | jq -er '.tool_name') || exit 2
block() { echo "Blocked: $TOOL. A person does this step." >&2; exit 2; }
# Never let a tool reach LinkedIn, whatever its name.
printf '%s' "$IN" | jq -c '.tool_input // {}' | grep -qi 'linkedin\.com' && block
# Expect mcp__server__action, with no extra __ inside the action.
case "$TOOL" in mcp__?*__?*) ;; *) block ;; esac
ACTION=${TOOL#mcp__*__}
case "$ACTION" in *__*) block ;; esac
for t in $EXTRA_ALLOW; do [ "$ACTION" = "$t" ] && exit 0; done
# Split snake, camel, kebab and UPPER case into lower-case words.
W=$(echo "$ACTION" | sed -E 's/([a-z0-9])([A-Z])/\1_\2/g;
  s/([A-Z]+)([A-Z][a-z])/\1_\2/g; s/-/_/g' | tr 'A-Z' 'a-z')
case "${W%%_*}" in get|list|search|read|find) ;; *) block ;; esac
REST="_${W#*_}_"; [ "$W" = "${W%%_*}" ] && REST=""
# The write list: any of these words after the first one blocks.
for w in and then or send create update delete reply forward \
  start activate schedule book add upload publish invite connect \
  resume modify move archive write remove set put submit execute \
  trigger dispatch enroll approve merge upsert share like follow; do
  case "$REST" in *"_${w}_"*) block ;; esac
done
# These can name a thing (get_post, read_email). They block
# unless they are the last word.
for w in post message email call comment; do
  case "$REST" in *"_${w}_"?*) block ;; esac
done
exit 0

We tested it on 74 tool names and 11 calls with real inputs. It let through 18 plain read tools, such as get_post, read_email and searchContacts. It blocked the other 56: every send, post, campaign and data-change tool, mixed names such as get_or_create_contact, getAndSend and get_upsert, and names with a second __. Empty or broken input is blocked too. With six research tools on EXTRA_ALLOW, a firecrawl_scrape of a company website ran, and the same call on a linkedin.com page was blocked. It is careful by design: get_call_log, get_email_draft and create_draft are blocked as well, so add any you need to EXTRA_ALLOW.

The “allow” line above runs the copy check without asking. In the default permission mode, any other command that could change something waits for a person to approve it.

J. The data files

Put one export of your contacts in data/contacts.csv, with column names in the first row.

ColumnNeeded?What goes in it
name, companyYesFull name and company name
website or linkedin_urlAt least oneThe company website, or the person’s LinkedIn link
title, email, countryAdvisedJob title, work email if you send email, where they are
last_touch, notesIf you have themDate you last spoke (YYYY-MM-DD) and anything you already know
status, next_action_date, last_stepThe setup fills themnew, skip, researched, queued, messaged, connected, replied, replied_not_now, meeting, client or nurture_due
--- data/contacts.csv (first two lines) ---
name,company,website,linkedin_url,title,email,country,status,last_touch,notes,next_action_date,last_step
Ana Example,Northwind Advisory,northwind.example,,Founder,,Portugal,new,,,,
--- data/do-not-contact.csv (same folder; clients, partners, competitors, anyone who asked you to stop) ---
name,company,email,linkedin_url,domain,reason
--- state/loop-state.json ---
{"batch_size": 5, "last_run": null, "cursor": 0, "blocked": false,
 "blocked_reason": "", "counts": {"fit_checked": 0, "researched": 0,
 "drafted": 0, "passed_gate": 0, "needs_a_person": 0, "skipped": 0}}
#!/usr/bin/env python3
"""scripts/check_contacts.py: check the contact file and pick the next batch.
   python3 scripts/check_contacts.py data/contacts.csv            # field report
   python3 scripts/check_contacts.py data/contacts.csv --next 5   # next batch"""
import argparse, csv, datetime as dt, json, os, sys
REQUIRED = ["name", "company"]
ONE_OF = [("website", "linkedin_url")]
READY = {"new", "nurture_due", ""}
# A personal address blocks that address only, never the whole provider.
FREE = {"gmail.com", "googlemail.com", "outlook.com", "hotmail.com", "live.com",
        "yahoo.com", "icloud.com", "me.com", "proton.me", "protonmail.com"}

def load(path):
    with open(path, newline="", encoding="utf-8-sig") as f:
        return list(csv.DictReader(f))

def norm(s): return (s or "").strip().lower()

def dom(s):
    s = norm(s)
    if "@" in s: s = s.split("@")[-1]
    s = s.split("://")[-1].split("/")[0].split("?")[0].split("#")[0].split(":")[0]
    return s[4:] if s.startswith("www.") else s

def li(s):
    # One form for every LinkedIn link: linkedin.com/in/name.
    # Drops scheme, www, country (uk.), query, end slash and locale (/en).
    s = norm(s).split("?")[0].split("#")[0].split("://")[-1].rstrip("/")
    if "linkedin.com/" in s:
        return "linkedin.com/" + "/".join(s.split("linkedin.com/", 1)[1].split("/")[:2])
    return s

def blocked(path):
    # The do-not-contact file sits next to the contacts file.
    if not os.path.exists(path):
        print(json.dumps({"ok": False, "ask": "Create " + path + " first, even if it only has the header row."}))
        sys.exit(1)
    names, doms, links = set(), set(), set()
    for r in load(path):
        names |= {norm(r.get(k)) for k in ("name", "company", "email") if norm(r.get(k))}
        doms |= {dom(r.get(k)) for k in ("domain", "email") if dom(r.get(k))} - FREE
        if li(r.get("linkedin_url")): links.add(li(r.get("linkedin_url")))
    return names, doms, links

def hit(d, doms):
    # Blocks the domain itself and any subdomain, e.g. sales.northwind.example.
    return any(d == b or d.endswith("." + b) for b in doms)

ap = argparse.ArgumentParser()
ap.add_argument("csv"); ap.add_argument("--next", type=int, default=0)
a = ap.parse_args()
if not os.path.exists(a.csv):
    print(json.dumps({"ok": False, "ask": "Export your contacts to data/contacts.csv."})); sys.exit(1)
rows = load(a.csv)
gaps = {}
for i, r in enumerate(rows, start=2):
    for c in REQUIRED:
        if not norm(r.get(c)): gaps.setdefault(c, []).append(i)
    for p in ONE_OF:
        if not any(norm(r.get(c)) for c in p): gaps.setdefault("/".join(p), []).append(i)
report = {"rows": len(rows), "ok": bool(rows) and all(len(v) <= 0.2 * len(rows) for v in gaps.values()),
          "rows_missing": {k: len(v) for k, v in gaps.items()}}
if not rows: report["ask"] = "The contact file has no rows. Add your export first."
if a.next and rows:
    dnc = os.path.join(os.path.dirname(os.path.abspath(a.csv)), "do-not-contact.csv")
    (bl_names, bl_doms, bl_links), today, pick = blocked(dnc), dt.date.today().isoformat(), []
    for i, r in enumerate(rows, start=2):
        due = (r.get("next_action_date") or "").strip()
        if norm(r.get("status")) not in READY or (due and due > today): continue
        if not norm(r.get("name")) or not norm(r.get("company")): continue
        if not norm(r.get("website")) and not norm(r.get("linkedin_url")): continue
        keys = {norm(r.get(k)) for k in ("name", "company", "email")} - {""}
        doms = {dom(r.get(k)) for k in ("website", "email")} - {""}
        if keys & bl_names or li(r.get("linkedin_url")) in bl_links - {""}: continue
        if any(hit(d, bl_doms) for d in doms): continue
        pick.append({"row": i, **r})
        if len(pick) >= a.next: break
    report["batch"] = pick
print(json.dumps(report, indent=2))
sys.exit(0 if report["ok"] else 1)

K. Brain file starters (brain/)

Each file starts with these headings. Fill them from your own evidence: posts, emails, call notes, proposals. If a heading has no evidence, write NOT IN THE EVIDENCE under it, a device from Charlie Hills’ Company Brain prompts. S14 The other skills read these exact headings.

--- brain/voice.md ---
# How we write
## Numbers        (lines per message, words per sentence, sign-off)
## Words we use   (with the source of each)
## Never list     (phrases we never write)

--- brain/customer.md ---
# Who buys from us
## Who they are     (role, company size, market)
## Who is not a fit (who we skip, and why)
## What they are stuck on
## The words they use        (quoted from real messages)
## Signs they need us now    (events with a date)
## What I got wrong          (assumptions the evidence did not back)

--- brain/offer.md ---
# What we sell
## In and out of scope
## Price rules   (or "price only on a call")
## Proof we may quote        (results the client allowed us to share)

--- brain/company.md ---
# Who we are
## What we do, for whom
## Who we turn away
## How we work               (the steps a client sees)

L. The other four skills (.claude/skills/)

The sales loop calls these. Each is one SKILL.md file. Here is what each one reads, does and writes.

SkillReadsDoesWrites
signal-researchcustomer.md “Signs they need us now”, the company website, public newsFinds dated events from the last two years, checks each fact against a second source, picks one angle (the judgement step runs on Sonnet)research/accounts/<company>.md, under 300 words, reused for 30 days
message-draftThe research file, voice.md, the matching part of offer.mdWrites one message, runs the copy check, then the gate helper. Has a reply mode for stuck dealsOne entry in messaging/queue/<date>.md, marked PASSED or NEEDS A PERSON
nurtureContacts marked messaged, connected or replied_not_nowPlans up to 10 light touches from posts a person pastes in: a page to follow, a post to like, a comment draft. A person does each oneA “Nurture” list in the same queue file, and next_action_date in contacts.csv
sales-reportcontacts.csv, the queue, loop-state.jsonCounts the week from files only, with no estimatesdeliverables/reports/<date>.md for Friday
Chapter 18

Sources and credits.

The setup is our sales system template. The ideas behind it credit the people below. Outside figures link to their source, with a date or check date.

A note on weight. Survey figures are as reported, forecasts are marked, and Part 3 uses made-up companies. Credit. The harness, loop, connector, company brain and stuck-deal ideas come from Charlie Hills’ free resources, linked below. Lines quoted from him are in quote marks. The folder, rules, examples, costs, plan and templates are ours. Thank you, Charlie.

  • S1Anthropic, Claude Code docs, How Claude remembers your project: “Claude treats them as context, not enforced configuration. To block an action regardless of what Claude decides, use a PreToolUse hook instead.” Checked 17 Nov 2026. code.claude.com/docs/en/memory
  • S2Kyle Poyar and Maja Voje, The 2026 State of B2B GTM Report. Growth Unhinged and GTM Strategist, October 2026. Survey of 300 B2B software and AI companies, September 2026. The return-odds ratings are on p.23, where N=275. Page numbers refer to the report PDF. stateofb2bgtm.com
  • S3Gartner press release, Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights. 20 May 2026. Survey of 645 B2B buyers, Aug to Sep 2025. gartner.com/en/newsroom
  • S4Gartner press release, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. 25 Jun 2025. A forecast. gartner.com/en/newsroom
  • S5LinkedIn User Agreement, section 8.2 (Don’ts), effective 3 Nov 2025. Read 18 Oct 2026, page checked live 17 Nov 2026. linkedin.com/legal/user-agreement
  • S6Anthropic, Claude Code docs, How Claude remembers your project (CLAUDE.md files are read at the start of every session). Checked 17 Nov 2026. code.claude.com/docs/en/memory
  • S7Anthropic, Claude Code docs, Subagents (each runs in its own context window; a subagent can set its own model). Checked 17 Nov 2026. code.claude.com/docs/en/sub-agents
  • S8Anthropic, Claude Code docs, Connect Claude Code to tools via MCP (tool definitions load when needed). Checked 17 Nov 2026. code.claude.com/docs/en/mcp
  • S9Anthropic, Claude Code docs, Extend Claude with skills (a skill’s body loads only when it is used). Checked 17 Nov 2026. code.claude.com/docs/en/skills
  • S10Charlie Hills (MarTech AI), The Agent Harness. Free resource, read 19 Oct 2026. charliehills.substack.com
  • S11Charlie Hills, Loop Engineering: The Complete Guide, and The 4 Claude Loops. Free resources, read 19 Oct 2026. charliehills.notion.site/382e396e… and charliehills.notion.site/3cae396e…
  • S12Charlie Hills, How to Set Up Your Claude Connectors (MCP). Free resource, read 19 Oct 2026. charliehills.notion.site
  • S13Charlie Hills, Machiavelli for Claude. Free skill, read 19 Oct 2026. github.com/charlie947/machiavelli
  • S14Charlie Hills (MarTech AI), Build a Company Brain: the four files your AI reads first. Free resource, read 19 Oct 2026. charliehills.notion.site/3d1e396e06bb81da93acc528e3dfc3f8
  • S15Noam Nisand, “Humanizer Prompt”, a LinkedIn post read in October 2026. Our ten checks summarise it in our own words.
  • S16Vendor price pages, list prices in USD with monthly billing, read 17 Nov 2026: Anthropic (Claude plans), LinkedIn Sales Navigator, Smartlead (Base $39), Instantly, ZeroBounce (one-off pack), Attio and Google Workspace (Business Starter, Flexible Plan, $8.40 a user). Data and enrichment tools are left out of the totals because their pages did not show a stable monthly price. Claude API use is our estimate. Rows marked “about” are rounded.
  • S17Anthropic, Claude Code docs, Configure permissions: “Permission rules are enforced by Claude Code, not by the model.” Also the Bash rule limits and the sandbox advice. Checked 17 Nov 2026. code.claude.com/docs/en/permissions
  • S18Anthropic, Claude Code docs, Hooks reference (PreToolUse and its deny decision). Checked 17 Nov 2026. code.claude.com/docs/en/hooks

Run one batch this week.

Five contacts, drafts only, nothing sent. Read the queue and see how many drafts you would send as they are. If you want a second pair of eyes on it, message Saif on LinkedIn.

linkedin.com/in/saiiful-islam-saif

Honest by design. Boutique by choice.

Fable InLead. fableinlead.comAI sales automation, with a person in charge. November 2026.

Keep a copy.

The same playbook as a 32-page PDF, to read offline or share with your team.

We use your details to give you the download. See our privacy policy.