S01E21: Most will stop here... don't

The Prompt Library

Let’s begin.

Three weeks in, you have something most reps don't.

You have the Main Branch: five files that tell your AI who your buyer is, what your product does, how to handle objections, what your process looks like, and how your team sounds.

You have the Personal Branch: four files that carry your account history, your voice, your deals, your read on your territory.

Every chat you open starts from a foundation now.

That's real.

That took work.

And it's more than most people in your org will ever build.

But here's what's still happening.

Every morning you open a new chat, load your files, and then type a prompt from scratch. Different instruction every time. Nothing saved. Nothing that builds. You prep for a call by describing what you need in fresh language. You write a follow-up opener starting from zero. You ask for help with an objection and frame it differently every time.

The codebase is the memory. The prompt is the question. You built the memory. The questions are still one-offs.

That's the half most reps leave unfinished.

And it's not obvious until you've been running the system for a few weeks and realize:

The Difference That Compounds

A one-off prompt works once. When the chat closes, it's gone. Tomorrow you write something similar from memory. Or you spend three minutes trying to recreate the phrasing that worked last time. Either way, nothing improves. There's no baseline. No iteration. Just the same starting line every day.

A prompt asset is different. It's a saved file with a name.

It has:

• The right context load at the top

• A specific instruction

• A defined output format

• A one-line quality bar that tells you when it's working

You run it. You read what comes back. If it needs more than three edits, you update the file. Then you run it again next time. The output gets better. Not because the AI learns. Because you're improving the prompt. That's the whole thing.

Your codebase is the memory. Your prompt library is the muscle memory.

One without the other is a half-built system.

What Goes in the Library

Not every prompt belongs here.

The rule is simple:

Seven prompts are worth building for most reps:

1. Pre-call prep
   
   Turns your account context into what you need to know before you walk in.

2. Cold opener
   
   Generates a personalized first touch in your voice based on a specific trigger.

3. Follow-up
   
   Drafts the post-conversation email that references what was actually said.

4. Post-call notes
   
   Captures what changed and what needs to go in accounts.md or deals.md before you close the tab.

5. Objection response
   
   Handles the specific objection in the context of this deal and this stage.

6. Deal review
   
   Gives you a Friday honest assessment of where things actually stand.

7. Pipeline digest
   
   Creates the end-of-week summary you'd send to your manager.

Seven prompts.

That's the whole library for most reps starting out.

Build them well once and you won't rewrite them again.

You'll only refine them.

What One Actually Looks Like

Every prompt asset has four parts.

1. Context Load

The context load tells the AI which files to read before it does anything. Teal files from the Main Branch. Amber files from the Personal Branch. Without this, the prompt is just a better one-off. With it, the output starts from everything you've built.

2. Instruction

The instruction is what you want done. The more specific it is, the less you edit the output. "Write a pre-call prep note" is too vague. "Give me three things to know about this account, two things to confirm on this call, and one thing to listen for" is specific enough to produce something usable.

3. Output Format

The output format tells it how to structure the response. Bullets only. No preamble. No "great question, here's what I think." Format instructions cut your editing by half.

4. Quality Bar

The quality bar is one sentence that defines what done looks like. If the output references the specific account by name and situation, it's working. If it sounds generic, the context load isn't specific enough. That one sentence gives you a test you can run every time.

How It Compounds

First run:

You edit six things. You note what didn't land. You update the instruction.

By run five:

You edit three things. The format is right. The tone still drifts occasionally.

By run twenty:

You add voice.md to the context load. Now it reads like you wrote it. One or two small edits.

By run fifty:

You read it once and send it. You don't remember the last time you changed something. It's yours. The prompt didn't get smarter on its own. You made it better. That's the difference between a prompt library and a folder of notes. Notes stay still. Prompt assets compound every time you show up.

What the Library Looks Like When It's Done

Seven files. Each one named for what it does. Each one with all four parts filled in.

The paid section below has the actual starter templates. Copy them, fill in the brackets, save them in a folder.

You don't need to write these from scratch. The structure is already here.

What makes them yours is the context load. Which files you point them at. Which accounts. Which deals. That's it.

The templates are the scaffolding. You fill them with your codebase and they become your prompts.

Rep Action This Week

Build two prompts:

• Pre-call prep

• Follow-up

Four parts each:

• Context load

• Instruction

• Output format

• Quality bar

Save them as files in a folder called: prompt-library/

Run them on a real account this week. After each run, note what you'd change. Update the file. That's the entire practice.

Two prompts. One folder. One week. That's the library.

Consultants spend 10+ hours a week in PowerPoint. SlideHub gives your team one searchable library of approved slides. Find slides, build with AI, and update decks without leaving Microsoft 365.

See SlideHub in action

Wispr Flow sits at the system level — no plugins, no extensions, no setup per app. Speak your prompts anywhere you type and get clean, paste-ready text. 4x faster than typing. Start flowing free.

The Starter Templates

Below are fill-in templates for all seven prompt library files.

Same format as the previous templates.

Copy them, fill in the brackets, delete the placeholder text.

One note:

The context load line is the most important part of each template.

Don't skip it or leave it vague.

The more precisely you specify which files to load, the less you'll need to edit the output.

The context load is what makes a prompt asset different from a saved note.

prompt-library/pre-call-prep.md

# PRE-CALL PREP

## CONTEXT LOAD
Load from main-branch/: buyer.md, product.md, objections.md
Load from personal-branch/: accounts/[account-name].md, deals.md (entry for [company] only)

## INSTRUCTION
I have a call with [contact name], [title] at [company], on [date].

Generate:
- 3 things I need to know walking in from account history and deal context
- 2 things I need to confirm on this call
- 1 thing to listen for that would change how I position

## OUTPUT FORMAT
Bullets only. No preamble. Under 150 words total.

## QUALITY BAR
If the output references this specific account's history and current deal stage, it's working.
If it sounds like it could apply to any account, the context load needs to be more specific.

prompt-library/cold-opener.md

# COLD OPENER

## CONTEXT LOAD
Load from main-branch/: buyer.md, product.md
Load from personal-branch/: voice.md

## INSTRUCTION
Write a cold opener to [prospect name], [title] at [company].

Trigger: [what made you reach out, such as job posting, funding, news, referral]

The opener should be 3-5 sentences.

Direct.

No setup.

Lead with their situation, not our product.

## OUTPUT FORMAT
Email only. No subject line. No sign-off. Just the body of the opener.

## QUALITY BAR
If you'd send it without changing the first sentence, the voice is right.
If it starts with "I", rewrite it.

prompt-library/follow-up.md

# FOLLOW-UP

## CONTEXT LOAD
Load from personal-branch/: accounts/[account-name].md, voice.md, deals.md (entry for [company] only)

## INSTRUCTION
Write a follow-up email after [describe the interaction, such as call, email exchange, demo].

Key thing that was said or decided: [what happened]

Specific next step we agreed on: [what's next]

The email should move us toward that next step without being pushy.

## OUTPUT FORMAT
Email body only. Under 100 words. My voice, not the team's.

## QUALITY BAR
If the email references what actually happened and ends with a clear action, it's done.

prompt-library/post-call-notes.md

# POST-CALL NOTES

## CONTEXT LOAD
Load from personal-branch/: accounts/[account-name].md, deals.md (entry for [company] only)
Load from main-branch/: process.md

## INSTRUCTION
I just finished a call with [contact] at [company].

Here's what happened:

[Paste or describe the key moments, what was said, what changed, what you heard]

Generate:
- What changed about this account, as an update for accounts.md
- What changed about this deal, as an update for deals.md
- Specific next move and date

## OUTPUT FORMAT
Three labeled sections. Bullet points. Ready to paste into the files directly.

## QUALITY BAR
If I can paste the output directly into my files without rewriting it, it's working.

prompt-library/objection-response.md

# OBJECTION RESPONSE

## CONTEXT LOAD
Load from main-branch/: objections.md, product.md
Load from personal-branch/: accounts/[account-name].md, deals.md (entry for [company] only), voice.md

## INSTRUCTION
I'm facing this objection from [contact] at [company]:

[paste exact objection or close paraphrase]

Context:

We are at [stage] in the deal.

[One sentence on their specific situation.]

Write a response that addresses the real concern, uses our documented handle from objections.md, and fits where this deal actually is.

## OUTPUT FORMAT
Written response only. 3-5 sentences. My voice. No "great question" or "I understand where you're coming from."

## QUALITY BAR
If the response acknowledges their specific situation and doesn't feel like a canned handle, it's working.

prompt-library/deal-review.md

# DEAL REVIEW

## CONTEXT LOAD
Load from personal-branch/: deals.md (all active entries)
Load from main-branch/: process.md

## INSTRUCTION
Run a deal health check across my active pipeline.

For each deal, tell me:
- Where it actually stands, not the CRM stage
- What's the real risk to close
- What needs to move this week specifically

## OUTPUT FORMAT
One section per deal. Deal name as header. Three bullet points each. Under 200 words total.

## QUALITY BAR
If it tells me something I didn't already know or hadn't said out loud, it's earning its place.

prompt-library/pipeline-digest.md

# PIPELINE DIGEST

## CONTEXT LOAD
Load from personal-branch/: deals.md (all entries), territory.md

## INSTRUCTION
Write a weekly pipeline summary I could send to my manager.

Cover:
- What closed or moved this week
- What's at risk
- What's on track to close next week
- One market observation from my territory that's relevant right now

## OUTPUT FORMAT
4 short sections with headers. Professional but in my voice. Under 250 words total.

## QUALITY BAR
If it reads like something I'd send without editing the opener, done.

The Prompt Anatomy Template

Here's the blank template for any prompt you build on your own.

Fill in all four parts every time. If you skip one, the prompt won't hold up over repeated runs.

# [PROMPT NAME]

## CONTEXT LOAD
Load from main-branch/: [which files]
Load from personal-branch/: [which files, which specific entries if needed]

## INSTRUCTION
[What you want done. Be specific. Name the output. Name the format. Name the scope.]

## OUTPUT FORMAT
[How the response should be structured. Bullets, sections, length, what to omit.]

## QUALITY BAR
[One sentence: what does "good output" look like for this prompt?
If output passes this test, the prompt is working. If not, update the instruction.]

The Improvement Protocol

After every prompt run:

• Read the output

• Count the edits you made

• If 0-2 edits: the prompt is working, no change needed

• If 3-4 edits: note what was off, then update one thing in the instruction or output format

• If 5+ edits: the context load is probably wrong. Check which files you're loading and whether they're current
  

That's it. You don't need a system. You need a habit. Run it. Read it. Update it if needed. Close the file.

The prompt gets better every time you do this. Skip it once and nothing breaks. Skip it consistently and the prompts drift back toward generic.

One update per run. Ten runs. You won't recognize the prompt you started with.

Next week marks the end of the BYOC series. I am very excited about S01E023 (two weeks away)