FBI Negotiation Tactics with AI Roleplay

Practice 9 FBI negotiation tactics via AI-powered drills for a real conversation.

Introduction

30 Second Summary

Everyone has that one conversation they keep putting off. The raise you need to ask for. The vendor contract that needs renegotiating. The client who keeps pushing back on your terms. You know it is coming, but you have no plan for what to say when things get tense.

In this project, you will practice 9 FBI negotiation tactics from Never Split the Difference by Chris Voss through AI-powered roleplay drills using Claude Cowork. You will walk away with tested scripts, a pocket playbook, and an AI voice recording of your best lines.

What You'll Build

A complete preparation system for one real conversation in the next 14-30 days, built through 4 scored drills against an AI Coach that pushes back on weak responses.

By the end of this project, you'll have:

  • 4 scored scripts tested against an AI Counterpart that escalates pressure across rounds.
  • A Pocket Playbook with 9 personalized trigger phrases you can glance at 30 seconds before the real conversation.
  • A voice playback .mp3 of your strongest script in a calm, controlled tone for rehearsal on your commute.
  • Secret Mission: Re-run your weakest drill on harder mode (the other person changes personality mid-conversation) and teach one drill to someone else within 48 hours.

Are there any prerequisites?

You need an active Claude Pro or Max subscription ($20/month or higher) with the Claude Desktop app installed, Python 3.10+ on your machine, and a real conversation you need to have in the next 14-30 days.

No negotiation experience required. The drills teach the tactics through doing, not reading.

Run the Automated Setup

You have a real conversation coming up in the next 14-30 days. You need tested scripts, not theory. In this step, one PowerShell script builds your entire practice system in under 30 seconds.

The script creates your folder structure, installs the voice tool, and generates every drill prompt you will need. Then you point Claude Cowork at the folder and answer 8 questions about your real situation.

In this step, get ready to:
  • Run a setup script that builds all project files and installs the voice tool.
  • Create a Cowork Project with your AI Coach permanently loaded.
  • Describe your real situation and score yourself on the 9 tactics.
Run the setup script

One script does all the heavy lifting. You create it, run it, and your entire project folder appears with everything inside.

  • Press the Windows key to open your search bar.
  • Type PowerShell and press Enter to open it.
  • Move to your Desktop by running this command:
cd ~\Desktop
  • Create and open a new file called bootstrap.ps1 by running:
notepad bootstrap.ps1

What does this command do?

This opens Notepad and creates a new file called bootstrap.ps1 on your Desktop. Notepad will ask if you want to create the file. Click Yes.

  • Copy the entire script below and paste it into the empty Notepad window:
# FBI Negotiation Lab - Setup Script
# Run once. Creates everything you need.
# Requires: Python 3.10+ installed and on PATH

Write-Host "=== FBI Negotiation Lab - Setup ===" -ForegroundColor Cyan
Write-Host "Building your project..." -ForegroundColor Yellow

# Create folder structure
$root = "fbi-negotiation-lab"
New-Item -ItemType Directory -Force -Path "$root/drills" | Out-Null
New-Item -ItemType Directory -Force -Path "$root/output" | Out-Null

# Install the voice tool
Write-Host "Installing voice tool (edge-tts)..." -ForegroundColor Yellow
pip install edge-tts --quiet

# Generate Coach system prompt
$coachPrompt = @"
You are Coach, an FBI-style negotiation scoring judge modeled on Chris Voss's methods. You are calibrated, never warm. You score scripts, not feelings.

Voice register: Voss's calm late-night-FM-DJ tone. Tactical. Direct. No academic hedging, no warm self-help, no "win-win" softening. Short sentences. Declarative.

You operate in a negotiation simulation with two other personas you control:
- Counterpart: Rotates archetype per drill (Boss, Buyer, Vendor, Investor/Stakeholder). Holds a reservation point and a hidden agenda. Reveals only on a tactic match. Escalates pressure across beats (Beat 1 cooperative, Beat 2 adds constraint, Beat 3 adds time pressure or dismissal).
- Black Swan Witness: Invoked once, at the Capstone drill Beat 4 only. Surfaces a pre-loaded unknown-unknown only when the learner asks a question that probes for unknown-unknowns.

Scoring per beat (unless overridden by drill-specific criteria):
- Criterion 1: 0-2
- Criterion 2: 0-2
- Criterion 3: 0-2
- Total per beat: max 6

After each beat:
1. Score the beat (3 criteria, 0-2 each)
2. Reveal "What Were They Thinking?" (Counterpart's internal monologue: what shifted, what held, what the hidden agenda is doing)
3. Deliver anti-misreading correction ONLY if triggered (learner made the specific error)

Anti-misreading bank (deploy only when the specific error occurs):
- "Mirroring is not parroting. 1-3 words. Not the whole sentence."
- "Labeling is not therapy. Name the emotion to move it, not to dwell."
- "Calibrated Questions are not interrogation. No 'Why.' 'Why' creates an accusation."
- "'No' is not rejection. 'No' is the start of the negotiation."
- "Tactical Empathy is not agreement. Understand the emotion. Do not concede to it."

Predict-then-reveal protocol:
Before each beat, ask the learner: "What do you predict the Counterpart will say or do in response to your move?" Record their prediction. After the beat plays out, name the gap between prediction and reality. The gap is the learning signal.

Variant ladder:
- Default: standard escalation across 3 beats
- Lighter: 2 beats, cooperative throughout
- Harder: Counterpart switches archetype at Beat 2 without warning

Output formats:
- Script Card: the learner's best line from the drill, formatted as a single block quote they can read before their real conversation
- Pocket Playbook: exactly 9 lines, one per tactic, each a personalized trigger phrase
- Mermaid Tactic Decision Tree: a flowchart in Mermaid syntax showing "If Counterpart does X, deploy tactic Y"

9-Item Tactic Readiness Scorecard:
Rate 1-3 on each (1 = cannot define or deploy, 2 = can define but not yet deployed live, 3 = have deployed live with observable result):
1. Mirror
2. Label
3. Tactical Empathy
4. Accusation Audit
5. "That's Right" signal
6. Calibrated Questions
7. No Gambit
8. Bending Reality (Ackerman)
9. Black Swans
Total: /27
"@
$coachPrompt | Out-File -FilePath "$root/coach-system-prompt.md" -Encoding utf8

# Generate Scenario Intake prompt
$intakePrompt = @"
Run Scenario Intake.

Ask me the following questions one at a time. Wait for my answer before asking the next. Do not batch them.

1. Who is your Counterpart? (Role, relationship to you, power dynamic)
2. What is the negotiation about? (Specific outcome at stake)
3. When does this happen? (Date or window, next 14-30 days)
4. What is your ideal outcome? (Specific, measurable)
5. What is your BATNA? (Best alternative if this negotiation fails completely)
6. What do you think their reservation point is? (The worst deal they would still accept)
7. What do you think their hidden agenda is? (What they want that they have not stated)
8. What are you most afraid they will say? (Your biggest vulnerability in this conversation)

After I answer all 8, output a Scenario Brief formatted as a reusable block I can paste into future conversations. Then run the 9-Item Tactic Readiness Scorecard (pre-score).
"@
$intakePrompt | Out-File -FilePath "$root/scenario-intake.md" -Encoding utf8

# Generate drill prompts
$drillMirror = @"
Run Mirror drill. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Boss / or my real Counterpart's role]

Run 3 beats with escalation:
- Beat 1 (cooperative): Counterpart makes opening statement about the situation
- Beat 2 (constraint): Counterpart introduces a budget, timeline, or resource constraint
- Beat 3 (dismissal): Counterpart attempts to end the conversation prematurely

Scoring criteria per beat:
1. Precision (0-2): Did the learner mirror exactly 1-3 words? (0 = parroted whole sentence or said nothing, 1 = mirrored 4-5 words, 2 = mirrored 1-3 key words)
2. Tonality (0-2): Did the learner note upward inflection? (0 = flat statement, 1 = question mark but no inflection note, 2 = explicit upward inflection noted)
3. Silence (0-2): Did the learner mark a 4+ second pause after the mirror? (0 = no pause, 1 = brief pause noted, 2 = explicit "..." or "4 seconds of silence" marked)

Before each beat, ask me to predict. After each beat, reveal "What Were They Thinking?" and score.

After all 3 beats, output the Mirror Script Card: my single best mirror line from this drill, formatted as a block quote.
"@
$drillMirror | Out-File -FilePath "$root/drills/drill-mirror.md" -Encoding utf8

$drillLabel = @"
Run Label + Tactical Empathy drill. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Buyer / or my real Counterpart's role]

Phase 1 - Accusation Audit:
Ask me to list 3-5 accusations my Counterpart might be thinking about me. Score the audit on completeness (did I cover the real fears?) and tone (fearless, speculative, not defensive).

Phase 2 - Label Sequence (3 beats targeting "That's Right"):
- Beat 1: Counterpart presents position with visible frustration. I deploy a Label.
- Beat 2: Counterpart softens but has not committed. I label the deeper emotion.
- Beat 3: I aim for "That's Right." Reveal whether I hit identity-level or surface-level.

Scoring criteria per beat:
1. Stem accuracy (0-2): Used "It seems/sounds/feels like..." (0 = "I think..." or "You are...", 1 = close but imprecise stem, 2 = exact Voss stem)
2. Emotion specificity (0-2): Named a real, specific emotion (0 = generic "upset"/"frustrated", 1 = somewhat specific, 2 = precise emotion matched to context)
3. Pause discipline (0-2): Marked 4-second pause after the label (0 = no pause, 1 = brief, 2 = explicit 4-sec noted)

Before each beat, ask me to predict. After each beat, reveal "What Were They Thinking?" and score.

After all beats, output the Label Script Card: my Accusation Audit opening + best label sequence.
"@
$drillLabel | Out-File -FilePath "$root/drills/drill-label.md" -Encoding utf8

$drillQuestions = @"
Run Calibrated Questions drill. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Vendor / or my real Counterpart's role]

CONSTRAINT (enforce strictly): The only legal moves in this drill are a Calibrated Question, a Mirror, or a Label. If I counter-offer, anchor a number, or ask "Why," immediately stop the beat, flag the violation, and make me redo.

Run 3 beats with forced problem-solving:
- Beat 1 (aggressive anchor): Counterpart throws a hard anchor. I must respond without counter-anchoring.
- Beat 2 (pushback): Counterpart says "That's all we have." I deploy another Calibrated Question.
- Beat 3 (deadline): Counterpart adds time pressure. I deploy Calibrated Question + Label combo.

Scoring criteria per beat:
1. Question construction (0-2): "How"/"What" opener, open-ended (0 = closed/yes-no question, 1 = open but used "Why" or "Can you", 2 = clean "How"/"What" calibrated question)
2. No-anchor discipline (0-2): Did not counter-offer or reveal a number (0 = counter-offered, 1 = hinted at a range, 2 = zero numbers spoken)
3. Forced problem-solving (0-2): Counterpart started offering solutions to my problem (0 = Counterpart held firm, 1 = Counterpart acknowledged but didn't solve, 2 = Counterpart actively proposed alternatives)

Before each beat, ask me to predict. After each beat, reveal "What Were They Thinking?" and score.

After all beats, output the Questions Script Card: my 3-5 best calibrated questions personalized to this scenario.
"@
$drillQuestions | Out-File -FilePath "$root/drills/drill-calibrated-questions.md" -Encoding utf8

$drillCapstone = @"
Run Capstone drill: No Gambit + Ackerman + Black Swans. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Investor/Stakeholder / or my real Counterpart's role]

My Ackerman numbers (pre-calculated on paper):
- 65%: [fill in]
- 85%: [fill in]
- 95%: [fill in]
- 100% (non-round): [fill in]

Sequencing: Defensive first, then offensive.

Run 4 beats:
- Beat 1 (Strategic No): Counterpart makes aggressive opening. I deploy a No-Oriented Question.
- Beat 2 (Fresh Accusation Audit): Counterpart pushes back harder. I open a fresh Accusation Audit to clear the air before going on offense.
- Beat 3 (Ackerman Sequence): Counterpart re-engages. I deploy my 65% number. Counterpart rejects. I move to 85% with a Calibrated Question. Score on discipline: did I jump to 100%?
- Beat 4 (Black Swan): Invoke the Black Swan Witness. A hidden piece of information surfaces ONLY if I ask an unknown-unknowns question ("What am I missing?" / "What's the one thing that would change everything here?"). If I do not ask, the Black Swan stays hidden and you note the miss.

Scoring criteria per beat:
1. Beat 1 - No Gambit: No-Oriented Question deployed correctly (0-2)
2. Beat 2 - Accusation Audit: Fearless, exhaustive, correct tone (0-2)
3. Beat 3 - Ackerman Discipline: Held the sequence, did not jump to final number (0-2)
4. Beat 4 - Black Swan Discovery: Asked an unknown-unknowns question that surfaced the hidden information (0-2)

Capstone threshold: 8/10 across 4 sub-tactics.

After all 4 beats, run the full After-Action Review:
1. Score the capstone (X/8, pass/fail against threshold)
2. Re-run the 9-Item Tactic Readiness Scorecard (post-score)
3. Compare pre-score vs. post-score, name the delta
4. Generate the Pocket Playbook (9 lines, one trigger phrase per tactic, personalized to my scenario)
5. Generate the Mermaid Tactic Decision Tree
6. Output Capstone Script Card

Then instruct me to run the generate-playback.ps1 script to create the AI voice .mp3 of my strongest script card.
"@
$drillCapstone | Out-File -FilePath "$root/drills/drill-capstone.md" -Encoding utf8

$harderVariant = @"
Run [drill name] HARDER.

[Paste Scenario Brief here]

Same drill structure as the default, but at Beat 2 the Counterpart switches archetype without warning:
- If original archetype was Boss, switch to emotionally-charged Spouse tone
- If original archetype was Buyer, switch to cold-transactional Investor tone
- If original archetype was Vendor, switch to confrontational Boss tone
- If original archetype was Investor, switch to warm-but-evasive Spouse tone

Add a 4th scoring criterion for the switch beat only:
4. Adaptation Speed (0-2): Did the learner recognize the shift and adjust their tactic selection within 1 response? (0 = continued same approach obliviously, 1 = noticed but fumbled, 2 = recognized and adapted cleanly)

Run the drill. Score all beats. Output a revised Script Card reflecting the adapted approach.
"@
$harderVariant | Out-File -FilePath "$root/drills/harder-variant-trigger.md" -Encoding utf8

# Generate the voice helper script
$playbackScript = @"
# Generates your AI voice playback
# Save your best script text to output/best-script.txt first, then run this.

`$scriptPath = "fbi-negotiation-lab/output/best-script.txt"
`$outputPath = "fbi-negotiation-lab/output/playback.mp3"

if (-not (Test-Path `$scriptPath)) {
    Write-Host "ERROR: No best-script.txt found." -ForegroundColor Red
    Write-Host "Save your strongest script card text to: `$scriptPath" -ForegroundColor Yellow
    exit 1
}

`$text = Get-Content `$scriptPath -Raw
Write-Host "Generating voice playback..." -ForegroundColor Cyan
edge-tts --voice en-US-AndrewNeural --rate=-15% --pitch=-2Hz --text "`$text" --write-media `$outputPath

if (Test-Path `$outputPath) {
    Write-Host "Done. Playing..." -ForegroundColor Green
    Start-Process `$outputPath
} else {
    Write-Host "ERROR: Voice generation failed. Check your internet connection." -ForegroundColor Red
}
"@
$playbackScript | Out-File -FilePath "$root/generate-playback.ps1" -Encoding utf8

# Test the voice
Write-Host "Testing voice..." -ForegroundColor Yellow
edge-tts --voice en-US-AndrewNeural --rate=-15% --pitch=-2Hz --text "Mirror. Label. Calibrate. Close." --write-media "$root/output/test-voice.mp3"

if (Test-Path "$root/output/test-voice.mp3") {
    Write-Host ""
    Write-Host "=== SETUP COMPLETE ===" -ForegroundColor Green
    Write-Host ""
    Write-Host "Your project:" -ForegroundColor Cyan
    Write-Host "  $root/coach-system-prompt.md    <- Paste into Cowork Project instructions"
    Write-Host "  $root/scenario-intake.md        <- First drill prompt to run in Cowork"
    Write-Host "  $root/drills/drill-mirror.md"
    Write-Host "  $root/drills/drill-label.md"
    Write-Host "  $root/drills/drill-calibrated-questions.md"
    Write-Host "  $root/drills/drill-capstone.md"
    Write-Host "  $root/drills/harder-variant-trigger.md"
    Write-Host "  $root/generate-playback.ps1     <- Creates your voice .mp3 after drills"
    Write-Host "  $root/output/test-voice.mp3     <- Should be playing now"
    Write-Host ""
    Write-Host "Next step: Open Claude Desktop, switch to Cowork, create a Project, attach this folder." -ForegroundColor Yellow
    Start-Process "$root/output/test-voice.mp3"
} else {
    Write-Host "WARNING: Voice test failed. Check that Python and pip are on your PATH." -ForegroundColor Red
}
  • Press Ctrl+S to save the file in Notepad, then close Notepad.
  • Back in PowerShell, run the script:
.\bootstrap.ps1

What does this script do?

The script handles everything automatically so you spend zero time on file management:

  • Creates the fbi-negotiation-lab/drills/ and fbi-negotiation-lab/output/ folders.
  • Installs edge-tts, a free Python package that generates AI voice from text.
  • Generates all drill prompt files and the Coach personality file.
  • Creates a voice helper script for later.
  • Tests the voice and auto-plays a sample audio clip.

Wait about 30 seconds. You should hear a calm male voice say "Mirror. Label. Calibrate. Close." and see === SETUP COMPLETE === in green text.

  • Confirm the folders and files exist by running:
ls fbi-negotiation-lab

You should see coach-system-prompt.md, scenario-intake.md, generate-playback.ps1, and the drills and output folders listed.

Script won't run or shows an error?

If PowerShell says scripts are disabled on this system, you need to allow script execution. Run this command, then try .\bootstrap.ps1 again:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

If you see a warning about pip not being recognized, Python is not on your PATH. Reinstall Python and make sure to check "Add Python to PATH" during installation.

If the voice test fails but everything else worked, check your internet connection. The edge-tts tool needs internet access to generate audio.

Help me troubleshoot my bootstrap.ps1 script. I'm on Windows and getting an error when I run it.

✔️ Awesome, I've got everything!

Great. Make sure you heard the test audio play. Your project folder is ready for the next substep.

ⓧ I'd like to double check the full code

Here is the complete bootstrap.ps1 file. Compare yours against this:

# FBI Negotiation Lab - Setup Script
# Run once. Creates everything you need.
# Requires: Python 3.10+ installed and on PATH

Write-Host "=== FBI Negotiation Lab - Setup ===" -ForegroundColor Cyan
Write-Host "Building your project..." -ForegroundColor Yellow

# Create folder structure
$root = "fbi-negotiation-lab"
New-Item -ItemType Directory -Force -Path "$root/drills" | Out-Null
New-Item -ItemType Directory -Force -Path "$root/output" | Out-Null

# Install the voice tool
Write-Host "Installing voice tool (edge-tts)..." -ForegroundColor Yellow
pip install edge-tts --quiet

# Generate Coach system prompt
$coachPrompt = @"
You are Coach, an FBI-style negotiation scoring judge modeled on Chris Voss's methods. You are calibrated, never warm. You score scripts, not feelings.

Voice register: Voss's calm late-night-FM-DJ tone. Tactical. Direct. No academic hedging, no warm self-help, no "win-win" softening. Short sentences. Declarative.

You operate in a negotiation simulation with two other personas you control:
- Counterpart: Rotates archetype per drill (Boss, Buyer, Vendor, Investor/Stakeholder). Holds a reservation point and a hidden agenda. Reveals only on a tactic match. Escalates pressure across beats (Beat 1 cooperative, Beat 2 adds constraint, Beat 3 adds time pressure or dismissal).
- Black Swan Witness: Invoked once, at the Capstone drill Beat 4 only. Surfaces a pre-loaded unknown-unknown only when the learner asks a question that probes for unknown-unknowns.

Scoring per beat (unless overridden by drill-specific criteria):
- Criterion 1: 0-2
- Criterion 2: 0-2
- Criterion 3: 0-2
- Total per beat: max 6

After each beat:
1. Score the beat (3 criteria, 0-2 each)
2. Reveal "What Were They Thinking?" (Counterpart's internal monologue: what shifted, what held, what the hidden agenda is doing)
3. Deliver anti-misreading correction ONLY if triggered (learner made the specific error)

Anti-misreading bank (deploy only when the specific error occurs):
- "Mirroring is not parroting. 1-3 words. Not the whole sentence."
- "Labeling is not therapy. Name the emotion to move it, not to dwell."
- "Calibrated Questions are not interrogation. No 'Why.' 'Why' creates an accusation."
- "'No' is not rejection. 'No' is the start of the negotiation."
- "Tactical Empathy is not agreement. Understand the emotion. Do not concede to it."

Predict-then-reveal protocol:
Before each beat, ask the learner: "What do you predict the Counterpart will say or do in response to your move?" Record their prediction. After the beat plays out, name the gap between prediction and reality. The gap is the learning signal.

Variant ladder:
- Default: standard escalation across 3 beats
- Lighter: 2 beats, cooperative throughout
- Harder: Counterpart switches archetype at Beat 2 without warning

Output formats:
- Script Card: the learner's best line from the drill, formatted as a single block quote they can read before their real conversation
- Pocket Playbook: exactly 9 lines, one per tactic, each a personalized trigger phrase
- Mermaid Tactic Decision Tree: a flowchart in Mermaid syntax showing "If Counterpart does X, deploy tactic Y"

9-Item Tactic Readiness Scorecard:
Rate 1-3 on each (1 = cannot define or deploy, 2 = can define but not yet deployed live, 3 = have deployed live with observable result):
1. Mirror
2. Label
3. Tactical Empathy
4. Accusation Audit
5. "That's Right" signal
6. Calibrated Questions
7. No Gambit
8. Bending Reality (Ackerman)
9. Black Swans
Total: /27
"@
$coachPrompt | Out-File -FilePath "$root/coach-system-prompt.md" -Encoding utf8

# Generate Scenario Intake prompt
$intakePrompt = @"
Run Scenario Intake.

Ask me the following questions one at a time. Wait for my answer before asking the next. Do not batch them.

1. Who is your Counterpart? (Role, relationship to you, power dynamic)
2. What is the negotiation about? (Specific outcome at stake)
3. When does this happen? (Date or window, next 14-30 days)
4. What is your ideal outcome? (Specific, measurable)
5. What is your BATNA? (Best alternative if this negotiation fails completely)
6. What do you think their reservation point is? (The worst deal they would still accept)
7. What do you think their hidden agenda is? (What they want that they have not stated)
8. What are you most afraid they will say? (Your biggest vulnerability in this conversation)

After I answer all 8, output a Scenario Brief formatted as a reusable block I can paste into future conversations. Then run the 9-Item Tactic Readiness Scorecard (pre-score).
"@
$intakePrompt | Out-File -FilePath "$root/scenario-intake.md" -Encoding utf8

# Generate drill prompts
$drillMirror = @"
Run Mirror drill. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Boss / or my real Counterpart's role]

Run 3 beats with escalation:
- Beat 1 (cooperative): Counterpart makes opening statement about the situation
- Beat 2 (constraint): Counterpart introduces a budget, timeline, or resource constraint
- Beat 3 (dismissal): Counterpart attempts to end the conversation prematurely

Scoring criteria per beat:
1. Precision (0-2): Did the learner mirror exactly 1-3 words? (0 = parroted whole sentence or said nothing, 1 = mirrored 4-5 words, 2 = mirrored 1-3 key words)
2. Tonality (0-2): Did the learner note upward inflection? (0 = flat statement, 1 = question mark but no inflection note, 2 = explicit upward inflection noted)
3. Silence (0-2): Did the learner mark a 4+ second pause after the mirror? (0 = no pause, 1 = brief pause noted, 2 = explicit "..." or "4 seconds of silence" marked)

Before each beat, ask me to predict. After each beat, reveal "What Were They Thinking?" and score.

After all 3 beats, output the Mirror Script Card: my single best mirror line from this drill, formatted as a block quote.
"@
$drillMirror | Out-File -FilePath "$root/drills/drill-mirror.md" -Encoding utf8

$drillLabel = @"
Run Label + Tactical Empathy drill. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Buyer / or my real Counterpart's role]

Phase 1 - Accusation Audit:
Ask me to list 3-5 accusations my Counterpart might be thinking about me. Score the audit on completeness (did I cover the real fears?) and tone (fearless, speculative, not defensive).

Phase 2 - Label Sequence (3 beats targeting "That's Right"):
- Beat 1: Counterpart presents position with visible frustration. I deploy a Label.
- Beat 2: Counterpart softens but has not committed. I label the deeper emotion.
- Beat 3: I aim for "That's Right." Reveal whether I hit identity-level or surface-level.

Scoring criteria per beat:
1. Stem accuracy (0-2): Used "It seems/sounds/feels like..." (0 = "I think..." or "You are...", 1 = close but imprecise stem, 2 = exact Voss stem)
2. Emotion specificity (0-2): Named a real, specific emotion (0 = generic "upset"/"frustrated", 1 = somewhat specific, 2 = precise emotion matched to context)
3. Pause discipline (0-2): Marked 4-second pause after the label (0 = no pause, 1 = brief, 2 = explicit 4-sec noted)

Before each beat, ask me to predict. After each beat, reveal "What Were They Thinking?" and score.

After all beats, output the Label Script Card: my Accusation Audit opening + best label sequence.
"@
$drillLabel | Out-File -FilePath "$root/drills/drill-label.md" -Encoding utf8

$drillQuestions = @"
Run Calibrated Questions drill. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Vendor / or my real Counterpart's role]

CONSTRAINT (enforce strictly): The only legal moves in this drill are a Calibrated Question, a Mirror, or a Label. If I counter-offer, anchor a number, or ask "Why," immediately stop the beat, flag the violation, and make me redo.

Run 3 beats with forced problem-solving:
- Beat 1 (aggressive anchor): Counterpart throws a hard anchor. I must respond without counter-anchoring.
- Beat 2 (pushback): Counterpart says "That's all we have." I deploy another Calibrated Question.
- Beat 3 (deadline): Counterpart adds time pressure. I deploy Calibrated Question + Label combo.

Scoring criteria per beat:
1. Question construction (0-2): "How"/"What" opener, open-ended (0 = closed/yes-no question, 1 = open but used "Why" or "Can you", 2 = clean "How"/"What" calibrated question)
2. No-anchor discipline (0-2): Did not counter-offer or reveal a number (0 = counter-offered, 1 = hinted at a range, 2 = zero numbers spoken)
3. Forced problem-solving (0-2): Counterpart started offering solutions to my problem (0 = Counterpart held firm, 1 = Counterpart acknowledged but didn't solve, 2 = Counterpart actively proposed alternatives)

Before each beat, ask me to predict. After each beat, reveal "What Were They Thinking?" and score.

After all beats, output the Questions Script Card: my 3-5 best calibrated questions personalized to this scenario.
"@
$drillQuestions | Out-File -FilePath "$root/drills/drill-calibrated-questions.md" -Encoding utf8

$drillCapstone = @"
Run Capstone drill: No Gambit + Ackerman + Black Swans. Default difficulty.

[Paste Scenario Brief here]

Counterpart archetype: [Investor/Stakeholder / or my real Counterpart's role]

My Ackerman numbers (pre-calculated on paper):
- 65%: [fill in]
- 85%: [fill in]
- 95%: [fill in]
- 100% (non-round): [fill in]

Sequencing: Defensive first, then offensive.

Run 4 beats:
- Beat 1 (Strategic No): Counterpart makes aggressive opening. I deploy a No-Oriented Question.
- Beat 2 (Fresh Accusation Audit): Counterpart pushes back harder. I open a fresh Accusation Audit to clear the air before going on offense.
- Beat 3 (Ackerman Sequence): Counterpart re-engages. I deploy my 65% number. Counterpart rejects. I move to 85% with a Calibrated Question. Score on discipline: did I jump to 100%?
- Beat 4 (Black Swan): Invoke the Black Swan Witness. A hidden piece of information surfaces ONLY if I ask an unknown-unknowns question ("What am I missing?" / "What's the one thing that would change everything here?"). If I do not ask, the Black Swan stays hidden and you note the miss.

Scoring criteria per beat:
1. Beat 1 - No Gambit: No-Oriented Question deployed correctly (0-2)
2. Beat 2 - Accusation Audit: Fearless, exhaustive, correct tone (0-2)
3. Beat 3 - Ackerman Discipline: Held the sequence, did not jump to final number (0-2)
4. Beat 4 - Black Swan Discovery: Asked an unknown-unknowns question that surfaced the hidden information (0-2)

Capstone threshold: 8/10 across 4 sub-tactics.

After all 4 beats, run the full After-Action Review:
1. Score the capstone (X/8, pass/fail against threshold)
2. Re-run the 9-Item Tactic Readiness Scorecard (post-score)
3. Compare pre-score vs. post-score, name the delta
4. Generate the Pocket Playbook (9 lines, one trigger phrase per tactic, personalized to my scenario)
5. Generate the Mermaid Tactic Decision Tree
6. Output Capstone Script Card

Then instruct me to run the generate-playback.ps1 script to create the AI voice .mp3 of my strongest script card.
"@
$drillCapstone | Out-File -FilePath "$root/drills/drill-capstone.md" -Encoding utf8

$harderVariant = @"
Run [drill name] HARDER.

[Paste Scenario Brief here]

Same drill structure as the default, but at Beat 2 the Counterpart switches archetype without warning:
- If original archetype was Boss, switch to emotionally-charged Spouse tone
- If original archetype was Buyer, switch to cold-transactional Investor tone
- If original archetype was Vendor, switch to confrontational Boss tone
- If original archetype was Investor, switch to warm-but-evasive Spouse tone

Add a 4th scoring criterion for the switch beat only:
4. Adaptation Speed (0-2): Did the learner recognize the shift and adjust their tactic selection within 1 response? (0 = continued same approach obliviously, 1 = noticed but fumbled, 2 = recognized and adapted cleanly)

Run the drill. Score all beats. Output a revised Script Card reflecting the adapted approach.
"@
$harderVariant | Out-File -FilePath "$root/drills/harder-variant-trigger.md" -Encoding utf8

# Generate the voice helper script
$playbackScript = @"
# Generates your AI voice playback
# Save your best script text to output/best-script.txt first, then run this.

`$scriptPath = "fbi-negotiation-lab/output/best-script.txt"
`$outputPath = "fbi-negotiation-lab/output/playback.mp3"

if (-not (Test-Path `$scriptPath)) {
    Write-Host "ERROR: No best-script.txt found." -ForegroundColor Red
    Write-Host "Save your strongest script card text to: `$scriptPath" -ForegroundColor Yellow
    exit 1
}

`$text = Get-Content `$scriptPath -Raw
Write-Host "Generating voice playback..." -ForegroundColor Cyan
edge-tts --voice en-US-AndrewNeural --rate=-15% --pitch=-2Hz --text "`$text" --write-media `$outputPath

if (Test-Path `$outputPath) {
    Write-Host "Done. Playing..." -ForegroundColor Green
    Start-Process `$outputPath
} else {
    Write-Host "ERROR: Voice generation failed. Check your internet connection." -ForegroundColor Red
}
"@
$playbackScript | Out-File -FilePath "$root/generate-playback.ps1" -Encoding utf8

# Test the voice
Write-Host "Testing voice..." -ForegroundColor Yellow
edge-tts --voice en-US-AndrewNeural --rate=-15% --pitch=-2Hz --text "Mirror. Label. Calibrate. Close." --write-media "$root/output/test-voice.mp3"

if (Test-Path "$root/output/test-voice.mp3") {
    Write-Host ""
    Write-Host "=== SETUP COMPLETE ===" -ForegroundColor Green
    Write-Host ""
    Write-Host "Your project:" -ForegroundColor Cyan
    Write-Host "  $root/coach-system-prompt.md    <- Paste into Cowork Project instructions"
    Write-Host "  $root/scenario-intake.md        <- First drill prompt to run in Cowork"
    Write-Host "  $root/drills/drill-mirror.md"
    Write-Host "  $root/drills/drill-label.md"
    Write-Host "  $root/drills/drill-calibrated-questions.md"
    Write-Host "  $root/drills/drill-capstone.md"
    Write-Host "  $root/drills/harder-variant-trigger.md"
    Write-Host "  $root/generate-playback.ps1     <- Creates your voice .mp3 after drills"
    Write-Host "  $root/output/test-voice.mp3     <- Should be playing now"
    Write-Host ""
    Write-Host "Next step: Open Claude Desktop, switch to Cowork, create a Project, attach this folder." -ForegroundColor Yellow
    Start-Process "$root/output/test-voice.mp3"
} else {
    Write-Host "WARNING: Voice test failed. Check that Python and pip are on your PATH." -ForegroundColor Red
}
Set up the Cowork Project

The Coach needs a permanent home. A Cowork Project bundles your folder, standing instructions, and memory so Claude already has context every time you start a session.

  • Open the Claude Desktop app on your computer.
  • At the top of the window, click Cowork in the mode picker to switch to Cowork mode.
  • In the left sidebar, click Projects, then click the + button to create a new project.
  • Select Use an existing folder and navigate to the fbi-negotiation-lab folder on your Desktop.
  • Name the project FBI Negotiation Lab.

Now you need to load the Coach personality into this project's Instructions.

  • Back in PowerShell, open the Coach file by running:
notepad fbi-negotiation-lab\coach-system-prompt.md
  • Select all the text in Notepad (Ctrl+A) and copy it (Ctrl+C).
  • Back in Claude Desktop, click into your FBI Negotiation Lab project to open its settings.
  • Find the Instructions field in the project settings.
  • Paste the entire Coach prompt into the Instructions field (Ctrl+V).
  • Click Save.

Why use a Cowork Project?

The Coach personality is over 1,000 words long. Without a Project, you would need to paste it at the start of every session.

A Cowork Project keeps the Instructions permanently loaded. Claude also reads your local drill files directly from the attached folder, so you never re-upload anything.

Your Cowork Project is set up. The Coach is now loaded permanently for every session you start inside this project.

Don't see Cowork mode?

Cowork requires a paid Claude plan (Pro, Max, Team, or Enterprise). Confirm your subscription is active under Settings in the Claude Desktop app. Make sure you have the latest version of Claude Desktop installed.

Help me find the Cowork mode in Claude Desktop. I have a paid plan but don't see it.

Describe your real situation and score yourself

The Coach needs to understand your specific negotiation before it can run drills. You will answer 8 questions about your real upcoming conversation, one at a time.

  • Inside your FBI Negotiation Lab project in Claude Cowork, start a new session.
  • Tell Claude to read the intake file by typing:
Read scenario-intake.md and run the Scenario Intake process.
  • Claude reads the file directly from your project folder and starts asking the intake questions.

What happens next?

The Coach will ask you 8 questions, one at a time. It will not batch them. Answer each one honestly about your real upcoming conversation. The questions cover who the other person is, what is at stake, your ideal outcome, your worst-case alternative, what you think they want, and what you are most afraid they will say.

After all 8 answers, the Coach outputs two things: a Scenario Brief (a block of text you reuse in every drill) and a 9-item pre-score (your self-rating 1-3 on each tactic, max 27).

  • Answer each of the 8 questions as the Coach presents them. Take your time with each answer.
  • When the Coach outputs your Scenario Brief, select and copy the entire brief text.
  • Save your Scenario Brief by running this in PowerShell:
notepad fbi-negotiation-lab\output\scenario-brief.md
  • Paste your Scenario Brief into the file (Ctrl+V).
  • Below the brief, paste your pre-score (the 9-item scorecard with your total out of 27).
  • Press Ctrl+S to save the file.

Your Scenario Brief is saved. This is the context block you will paste into every drill from here on. Your pre-score tells you where you are starting.

Coach batched all questions at once?

If the Coach asked all 8 questions in one message instead of one at a time, start a new session inside the same Cowork Project and ask Claude to read the intake file again. The Instructions should keep the Coach in character. If it still batches, add "Remember: one question at a time" at the end of your message.

The Coach asked all 8 questions at once instead of one at a time. Help me fix this.

Your environment is locked and loaded. You have a Coach that knows your situation, a starting score to beat, and a Scenario Brief ready to paste. Next up, you will use your first FBI tactic against a simulated version of your real Counterpart.

Mirror Under Pressure

Your lab is built and your Scenario Brief captures the real conversation ahead. Now it is time to practice the simplest, most underrated tool in negotiation: repeating the last 1-3 words someone said, then going completely silent.

Mirroring sounds absurd on paper. It works because silence makes people fill the gap. In this step, you will run 3 rounds of mirror practice against your AI Counterpart, with pressure increasing each round.

In this step, get ready to:
  • Run 3 rounds of mirror practice with escalating pressure.
  • Score your performance on precision, tonality, and silence.
  • Save your Mirror Script Card for the real conversation.
Three rounds, getting harder

Each round, the Coach asks you to predict what the Counterpart will say before you deploy your mirror. After you respond, the Coach reveals what actually happened inside their head. The gap between your prediction and reality is where the learning lives.

  • Return to your FBI Negotiation Lab project in Claude Cowork.
  • Start a new session within the project.
  • Tell Claude to read the drill file by typing: "Read drills/drill-mirror.md and run the Mirror drill."
  • When Claude asks for your Scenario Brief, tell it to read output/scenario-brief.md from the project folder.

What happens next?

The Coach takes over. It plays your Counterpart and runs you through 3 beats of increasing difficulty. Before each beat, it asks you to predict what the Counterpart will say. After you respond, it scores you and reveals the Counterpart's internal monologue.

Here is what each round looks like:

  • Round 1 (cooperative): The Counterpart makes an opening statement about the situation. You predict what they will say, then mirror their last 1-3 words. The Coach shows you what shifted inside their head.
  • Round 2 (constraint added): They introduce a budget, timeline, or resource constraint. You mirror again. The Coach scores you and reveals the impact.
  • Round 3 (shutdown attempt): They try to end the conversation prematurely. You mirror their words and hold silence. The Coach shows you what the silence did.

The mirror formula

Pick only 1-3 words from what they just said (usually the last few). Use an upward inflection (like a question). Then mark a 4-second silence. That is it. Do not add commentary, do not paraphrase, do not explain.

  • Work through all 3 rounds. Follow the Coach's prompts, make your predictions, and deploy your mirrors.

Coach cutting off or not scoring?

Make sure you started a new session inside the Project, not a general Claude chat. The Coach persona only loads inside the FBI Negotiation Lab project in Cowork.

If the Coach skips the predict step, remind it: "Run the predict-then-reveal protocol before each beat."

Help me get my Coach to follow the drill structure.

Get your score and script

After all 3 rounds, the Coach gives you a final score and your Mirror Script Card. The scoring tells you exactly where your mirroring broke down (or held up).

How does the scoring work?

Each round is scored 0-6 on three criteria:

  • Precision (0-2): Did you pick only 1-3 words? Repeating a whole sentence scores 0.
  • Tonality (0-2): Did you note the upward inflection? A flat statement scores 0.
  • Silence (0-2): Did you mark a 4-second pause after the mirror? No pause scores 0.

💡 What if the Coach corrects me?

If you repeated too many words, the Coach fires an anti-misreading correction: "Mirroring is not parroting. 1-3 words. Not the whole sentence." This only triggers when you make the specific mistake. It is not a lecture. It is a signal to adjust.

  • Review your scores across all 3 rounds. Note which criterion was weakest.
  • The Coach outputs your Mirror Script Card at the end: your single best mirror line from this drill, formatted as a block quote.
  • Copy the Script Card text from your Cowork session.
  • Save it to fbi-negotiation-lab/output/script-card-mirror.md on your machine.

Not sure what to save?

The Script Card is a short block quote (2-5 lines) containing your best mirror line plus context on when you used it. If the Coach did not output one explicitly, paste this at the end of your conversation: "Output my Mirror Script Card now."

Help me get my Mirror Script Card from the Coach.

Confirm your Script Card is saved by checking that fbi-negotiation-lab/output/script-card-mirror.md exists and contains your best mirror line.

You now have a tested mirror line ready to deploy. Next up, you will practice labeling emotions and earning the most powerful two words in negotiation: "That's right."

Label the Emotion, Earn "That's Right"

Your Mirror drill proved that repeating someone's words makes them keep talking. Now you need the tool that makes them feel understood. Labeling is naming the other person's emotion out loud without agreeing with it.

The target response is "That's right." Not "You're right" (that means they want you to stop talking). "That's right" means you actually understand their position at an identity level. In this drill, you practice naming emotions until you earn that signal.

In this step, get ready to:
  • Draft an Accusation Audit listing what the other person probably thinks about you.
  • Run three labeling rounds with escalating difficulty, targeting "That's right."
  • Save your Label Script Card with your Accusation Audit opening and best label sequence.
Open with what they are probably thinking about you

Before you label emotions in real-time, you start with an Accusation Audit. This means listing every negative thing the other person might be thinking about you. You say them out loud, first, before they can.

This clears the air and shows you are not afraid of their judgment. The key is being fearless, not defensive.

  • Start a new session in your FBI Negotiation Lab Cowork Project.
  • Tell Claude to read the drill file: "Read drills/drill-label.md and run the Label + Tactical Empathy drill."
  • When Claude asks for your Scenario Brief, tell it to read output/scenario-brief.md from the project folder.

The Coach asks you to list 3-5 accusations the other person might be thinking about you right now, even if they have not said them out loud.

  • Draft your opening using stems like "You're probably thinking..." or "It might seem like..." to name each accusation.
  • Aim for the real fears, not surface-level annoyances.
  • Submit your Accusation Audit when you have 3-5 items listed.

What is an Accusation Audit?

An Accusation Audit is when you take a step back and name every possible negative thing the other side might be thinking about you. You say these out loud, first, before they can use them against you.

The tone matters as much as the content. You deliver it fearlessly and speculatively. Not "I know you think I'm difficult." Instead: "You're probably thinking I'm being unreasonable about timeline." The speculative framing gives them room to correct you, which keeps the conversation moving.

The Coach scores your Accusation Audit on two things: whether you covered the real fears (not softballs) and whether your tone was fearless rather than defensive.

Run three rounds of labeling

After the Accusation Audit, the drill shifts into three live rounds. Each round, the Counterpart shows a different emotional state. Your job is to label that emotion accurately enough that they eventually say "That's right."

  • Round 1: The Counterpart shows frustration. Label it using a stem like "It seems like..." or "It sounds like..." followed by a specific emotion. Then mark a 4-second silence.
  • Round 2: The Counterpart softens but has not committed. Label the deeper emotion underneath their words.
  • Round 3: Go for "That's right." This means hitting the identity level. Name what this situation means to them personally, not just how they feel about the logistics.

What makes a good label?

A label has three parts: the correct stem ("It seems like...", "It sounds like...", "It looks like..."), a specific named emotion, and 4 seconds of silence after you deliver it.

The stem matters because "I think you're frustrated" puts YOU at the center. "It seems like there's frustration around timeline" puts the EMOTION at the center. That small shift removes defensiveness.

Before each round, the Coach asks you to predict what the Counterpart will say or do. After you deploy your label, the Coach reveals "What Were They Thinking?" and scores your attempt.

Surface vs. identity

"It seems like you're frustrated about the budget" is surface-level. "It sounds like you want to be seen as the person who built this the right way" is identity-level. "That's right" only comes when you hit the identity layer.

Get your score and save the script card

After all three rounds, the Coach gives you a final score and generates your Label Script Card.

  • Review your scores. Each round is scored 0-6 on three criteria: stem accuracy, emotion specificity, and pause discipline.
  • Read any corrections the Coach delivers. Two common ones: "Labeling is not therapy. Name the emotion to move it, not to dwell." and "Tactical Empathy is not agreement. Understand it. Do not give in to it."
  • Copy the Label Script Card the Coach generates (your Accusation Audit opening + your best label sequence).
  • Save it to fbi-negotiation-lab/output/script-card-label.md.

What is the difference between "That's right" and "You're right"?

"That's right" is what people say when they feel completely understood. It means alignment. "You're right" is what people say to get you to stop talking. It means compliance, not agreement.

If you got "You're right" in your drill, you labeled at the surface level. Go deeper. Name what the situation means to them as a person, not just the logistics they are dealing with.

🙋‍♀️ Coach cut the conversation short?

If the Coach ended the drill before all three rounds, your Scenario Brief might be missing from the prompt. Go back to your message and confirm you replaced [Paste Scenario Brief here] with the actual text from scenario-brief.md.

If the Coach is not scoring per beat or not asking for predictions, start a new session and ask Claude to re-read the drill file. The Coach persona resets with each new session in the Cowork Project.

Help me troubleshoot my Label drill session.

You now have two script cards: one for getting people talking (Mirror) and one for making them feel understood (Label). Next, you flip the dynamic entirely. Instead of responding to their emotions, you make them solve your problem for you using only questions.

Make Them Solve Your Problem

You got them talking with a Mirror. You earned their trust with a Label. Now they throw a hard number at you. Your instinct screams to counter-offer. That instinct is wrong.

In this step, you practice Calibrated Questions. These are open-ended "How" and "What" questions that flip the dynamic. Instead of you fighting their number, they start solving your problem for you.

In this step, get ready to:
  • Run the Calibrated Questions drill with 3 rounds of escalating pressure.
  • Hold a hard constraint: no counter-offers, no numbers, no "Why."
  • Save your Questions Script Card with 3-5 personalized questions.
Learn the rule before you start

This drill has one strict constraint the Coach enforces. You can only use three moves: a "How" or "What" question, a Mirror, or a Label. If you counter-offer, throw a number, or ask "Why," the Coach stops you immediately and makes you redo the beat.

Why can't I ask "Why"?

"Why" makes people defensive. It sounds like an accusation ("Why did you do that?" reads as "Justify yourself"). "How" and "What" sound collaborative. "How am I supposed to make that work?" puts the problem in their lap without triggering their fight response.

  • Start a new session in your FBI Negotiation Lab Cowork Project.
  • Tell Claude to read the drill file: "Read drills/drill-calibrated-questions.md and run the Calibrated Questions drill."
  • When Claude asks for your Scenario Brief, tell it to read output/scenario-brief.md from the project folder.

Same flow as your earlier drills

This is the same rhythm you used for the Mirror and Label drills. New session in Cowork, tell Claude to read the drill file, and the Coach takes over.

Run three rounds as they anchor and you redirect

The Coach runs 3 beats. Each one escalates pressure. Your job is to resist the urge to counter-offer and instead ask a question that makes them do the problem-solving.

  • Round 1 (aggressive anchor): The Counterpart throws a hard number at you ("The budget is X, take it or leave it"). Your instinct is to counter. Instead, deploy a Calibrated Question like "How am I supposed to make that work?" The Coach shows what happened inside their head.
  • Round 2 (pushback): They dig in ("That's all we have"). You ask another question. The Coach scores you.
  • Round 3 (deadline): They add time pressure. You combine a Calibrated Question with a Label. The Coach reveals whether they started offering solutions to YOUR problem.

What does the Coach score you on?

Each beat is scored 0-6 across three criteria:

  • Question construction (0-2): Did you open with "How" or "What"? Is it genuinely open-ended?
  • No-anchor discipline (0-2): Did you speak zero numbers? No counter-offers, no ranges, no hints.
  • Forced problem-solving (0-2): Did the Counterpart start proposing alternatives or solutions for you?

Before each beat, the Coach asks you to predict what the Counterpart will say. After your move plays out, the Coach reveals the gap between your prediction and what actually happened.

What if the Coach stops you mid-beat?

If you slip and counter-offer, throw a number, or ask "Why," the Coach fires the correction: "Calibrated Questions are not interrogation. No 'Why.' 'Why' makes people feel accused." It then resets the beat for you to try again.

Getting stopped is not a failure. It means the constraint is working. Redo the beat with a clean "How" or "What" question.

Get your score and save the Questions Script Card

After all 3 beats, the Coach outputs your Questions Script Card. This is a set of 3-5 Calibrated Questions personalized to your specific scenario. These are the questions you will use in your real conversation.

  • Review the scores the Coach gave you for each beat.
  • Copy the Questions Script Card that the Coach outputs at the end of the drill.
  • Save it to fbi-negotiation-lab/output/script-card-questions.md.

What does a good Questions Script Card look like?

Your card should have 3-5 questions that are specific to your real scenario. Generic questions like "How can we work together?" are too soft. Strong questions force a concrete response: "How am I supposed to make that work with the current headcount?" or "What does a successful outcome look like from your side?"

🙋‍♀️ Coach didn't output a Script Card?

If the conversation ended without a formatted Script Card, type: "Output my Questions Script Card now: my 3-5 best calibrated questions from this drill, formatted as a block quote." The Coach will generate it.

Help me get my Questions Script Card from the Coach.

You now have three Script Cards in your arsenal: Mirror, Label, and Calibrated Questions. Next up is the final boss. You bring all the tactics together, hold your offer numbers under pressure, and hunt for the hidden information that changes everything.

Close It — No, Ackerman, Black Swan

You have three separate muscles now. Mirroring gets them talking. Labeling earns trust. Calibrated Questions redirect pressure. This capstone drill tests whether you can chain all of them together when the stakes escalate across four consecutive rounds.

The Coach deploys its toughest sequence here: your Counterpart opens aggressively, pushes harder, re-engages on numbers, and then guards a piece of hidden information. You pass by holding your pre-calculated numbers under pressure and asking the one question that unlocks what they are not telling you.

In this step, get ready to:
  • Deploy a strategic "No" question, an Accusation Audit, your Ackerman numbers, and a Black Swan question across 4 escalating rounds.
  • Score your capstone and compare your post-score to where you started.
  • Generate your Pocket Playbook, Tactic Decision Tree, and AI voice playback of your strongest script.
Confirm your Ackerman numbers are ready

The Ackerman system uses a specific sequence of offers to anchor the negotiation without revealing your final number too early. You calculated these on paper before opening Claude Cowork.

  • Check that you have 4 numbers written on your paper:
  • Your first offer: 65% of your target.
  • Second offer: 85% of your target.
  • Third offer: 95% of your target.
  • Final offer: 100% of your target, written as a non-round number (e.g. $97,300 not $97,000).

Why non-round numbers?

A precise number like $97,300 signals that you calculated it carefully. Round numbers feel arbitrary and invite counter-offers. The Coach will test whether you hold this sequence under pressure.

Run the capstone drill: four rounds, defensive then offensive

This drill chains four tactics in sequence. You start defensive (resetting the dynamic), then go on offense (holding numbers and hunting for hidden information).

  • Switch back to your FBI Negotiation Lab project in Claude Cowork.
  • Start a new session.
  • Tell Claude to read the drill file: "Read drills/drill-capstone.md and run the Capstone drill."
  • When Claude asks for your Scenario Brief, tell it to read output/scenario-brief.md from the project folder.
  • Fill in your 4 Ackerman numbers when Claude asks (65%, 85%, 95%, 100% non-round).

What happens across the 4 rounds?

The Coach asks you to predict the Counterpart's response before each round, then reveals what actually happened. Here is what each round tests:

  • Round 1 (Strategic No): They open aggressively. You deploy a No-Oriented Question like "Have you given up on making this work?" This resets the power dynamic.
  • Round 2 (Accusation Audit): They push harder. You name everything they might be thinking about you. Out loud. First. This clears the air before you go on offense.
  • Round 3 (Ackerman): They re-engage. You drop your 65% number. They reject. You move to 85% paired with a Calibrated Question. The Coach watches whether you panic and jump straight to 100%.
  • Round 4 (Black Swan): Hidden information exists that changes everything. It only surfaces if you ask a question that probes what you do not know: "What am I missing?" or "What is the one thing that would change everything here?"
  • Work through all 4 rounds. For each round, write your prediction first when the Coach asks, then deploy your tactic.
  • Pay attention to the "What Were They Thinking?" reveal after each round. The gap between your prediction and reality is where the learning happens.

Hold your numbers on paper

Glance at your written numbers before Round 3 starts. Calculating under pressure always ends with you caving to your final number immediately. The paper keeps you honest.

🙋‍♀️ Coach stopped the round?

If the Coach flags that you counter-offered, threw a number too early, or asked "Why," it means you broke the constraint for that round. Re-read the Coach's correction, then redo the round using the right tactic.

help me understand why the Coach stopped my round and what tactic I should have used instead

Save your outputs and generate the voice playback

After all 4 rounds, the Coach runs a full after-action review. It scores your capstone against the 8/10 threshold, re-runs your 9-item scorecard (post-score), compares it to your pre-score, and generates three final deliverables.

  • Wait for the Coach to finish the after-action review. It generates your Pocket Playbook (9 lines), Tactic Decision Tree (Mermaid flowchart), and Capstone Script Card.
  • Copy the Capstone Script Card and save it to fbi-negotiation-lab/output/script-card-capstone.md.
  • Copy the Pocket Playbook (including your post-score and delta at the bottom) and save it to fbi-negotiation-lab/output/playbook.md.
  • Copy the Mermaid Tactic Decision Tree and save it to fbi-negotiation-lab/output/tactic-tree.md.

What is in the Pocket Playbook?

Exactly 9 lines. One trigger phrase per tactic, personalized to your scenario. This is what you glance at 30 seconds before the real conversation starts. The post-score at the bottom shows how far you have come from where you started.

Now pick your strongest script card from any of the 4 drills you completed. This is the line that felt most natural and most likely to land in your real conversation.

  • Look back at your 4 script cards: script-card-mirror.md, script-card-label.md, script-card-questions.md, and script-card-capstone.md.
  • Pick the one that feels most ready to use. Copy just the script text (the line you would actually say).
  • Save it to fbi-negotiation-lab/output/best-script.txt as plain text.

The voice generation script reads this file and creates an AI audio version you can replay before your real conversation. It uses the edge-tts tool with a calm, slowed-down voice to train your ear for pacing and tone.

  • Return to your PowerShell terminal and generate the voice playback by running this command:
.\fbi-negotiation-lab\generate-playback.ps1

What does this script do?

It reads your best-script.txt file and feeds it to edge-tts with the en-US-AndrewNeural voice slowed by 15% and pitched down 2Hz. The result is a calm, controlled version of your script saved as playback.mp3. It auto-plays when generation finishes.

  • Listen to the playback. You should hear your script read in a calm, low-pitched voice at a pace you can mirror in the real conversation.
  • Confirm that fbi-negotiation-lab/output/playback.mp3 exists in your output folder.

Seeing "ERROR: No best-script.txt found"?

The script could not find your text file. Make sure you saved your strongest script card text to fbi-negotiation-lab/output/best-script.txt (not inside a subfolder or with a different filename).

If you see "Voice generation failed," check your internet connection. The edge-tts tool requires a network connection to generate audio.

help me troubleshoot why my generate-playback script is failing

💡 Listen on your commute

This audio is not a presentation. It is a pacing metronome. Play it in your car, on your walk, or on the train before the real conversation. Listen for the rhythm and calm tone, not just the words.

Secret mission

Harder Mode + Teach Someone Else

The difficulty goes up. The other person changes personality mid-conversation. They start professional, then suddenly get emotional (or the reverse). You have to recognize the shift and adjust your tactic in one response.

Clean Up Your Resources

Clean Up Your Resources

This project runs entirely locally and uses your existing Claude subscription. There are no ongoing costs beyond your paid plan.

Decide whether to keep your resources available for future conversations, archive them to revisit later, or delete them entirely.

Resources you used:

  • Cowork Project "FBI Negotiation Lab" in Claude Desktop (stores your Coach persona and drill sessions locally).
  • Local folder fbi-negotiation-lab/ on your Desktop (all drill files, output scripts, and audio).
  • Python package edge-tts (installed via pip, generates voice audio).

Keep everything running

No action needed. Choose this if you plan to reuse the system for a future conversation.

  • Use this project again on Day 30 for a re-score against your real conversation's outcome.
  • On Day 90, save a new Scenario Brief for a completely different conversation and run different drills.
  • The edge-tts package stays installed and available for future voice generation at no cost.
  • Your Cowork Project keeps the Coach persona loaded permanently. Open a new session any time to run a fresh drill.

Pause - I'll come back to this later

Archive the Cowork Project to get it out of your sidebar without losing anything.

  • Open the Claude Desktop app and switch to Cowork mode.
  • Find "FBI Negotiation Lab" in the Projects sidebar.
  • Open the project's menu and select Archive.
  • The project disappears from your active list but the local fbi-negotiation-lab/ folder stays untouched on your Desktop.

Delete - I don't want to use this again

Remove all project resources and start fresh.

  • Archive the Cowork Project (same steps as the Pause tab above). Archiving removes the project metadata from Claude Desktop.
  • Delete the local project folder by running this in PowerShell:
cd ~/Desktop
Remove-Item -Recurse -Force fbi-negotiation-lab
  • Uninstall the voice tool by running:
pip uninstall edge-tts -y
  • Confirm the folder is gone by running:
ls ~/Desktop
  • You should no longer see fbi-negotiation-lab in the output.

That's a Wrap!

That's a Wrap!

Nice work. You just built a complete conversation preparation system for a real negotiation in the next 14-30 days, powered by AI roleplay coaching and 9 FBI-trained tactics.

You've learned how to:

  • Run 4 scored drills covering 9 negotiation tactics (Mirroring, Labeling, Tactical Empathy, Accusation Audit, Calibrated Questions, No Gambit, Ackerman, and Black Swans) against an AI Counterpart that escalated pressure across each round.
  • Produce a Pocket Playbook, a Tactic Decision Tree, and an AI voice playback (.mp3) of your strongest script for rehearsal before the real conversation.
  • Measure your improvement using a 9-item Tactic Readiness Scorecard (pre-score vs. post-score), giving you a concrete number showing how far you came.
  • Secret Mission: Re-ran your weakest drill on harder mode (Counterpart switched archetype mid-conversation) and taught one drill to someone else within 48 hours.

Ready to quiz yourself?