Agent Builder | Humans+Agents

Overview

What This System Produces

A Complete 7-File Workflow Package

00-START-HERE.html — Landing page with setup instructions
01-process-agent.html — Machine-readable execution instructions
01-process-consultant.html — Human-readable consultant guide
02-context.html — Input manifest with all required data
03-example.html — Golden example showing completed deliverable
04-quality.html — QC checklist with weighted scoring
quick-start-prompt.txt — Copy/paste prompt to bootstrap workflow

Estimated Time

5m

Setup

10m

Input Collection

15m

Generation

Cloud Mode

Use this mode with Claude Projects or ChatGPT.

Claude Project Recommended

Create a new Claude Project
Upload all 7 files from the Agent-Builder folder to the Project Knowledge
Upload the Blue Ocean package folder as reference (the-blue-ocean-positioning-map/)
Copy the contents of quick-start-prompt.txt and paste as your first message
Provide the package name and description when prompted

Local Mode

Use this mode with Claude Code CLI.

Claude Code

Open terminal in the workspace folder
Run: claude
Say: "Read the Agent-Builder folder in # Production System and the Blue Ocean package as reference"
Provide the package name and description
Agent will generate all 7 files to the specified output folder

Files in This Package

File	Type	Purpose
`00-START-HERE.html`	Entry	Orientation and setup
`01-process-agent.html`	Agent	Machine-readable instructions for AI execution
`01-process-consultant.html`	Guide	Human-readable process for consultants
`02-context.html`	Context	All inputs needed to generate a package
`03-example.html`	Example	Reference to Blue Ocean as golden example
`04-quality.html`	Quality	QC checklist for validating generated packages
`quick-start-prompt.txt`	Prompt	Bootstrap prompt for quick execution

Quick Start

Quick Start Prompt

Copy and paste this prompt into a Claude Project or Claude Code session to bootstrap a new workflow package generation.

You have been loaded with the Agent Builder workflow.

Read the 01-process-agent file and begin execution. You will be generating a complete 7-file workflow package.

IMPORTANT: Deep research is the default, not optional. Reference blueocean2.html to see the quality bar -- specific data, buyer quotes, competitive intel, risk treatment. Speed-run outputs without research are not acceptable.

Start by asking me for the required inputs:
1. Package name (e.g., "The Competitor Analysis")
2. Package description (1-2 sentences)
3. Firm name (The Growth Firm, The Revenue Firm, or The Talent Firm)
4. Output folder path

After gathering inputs:
1. Reference the Blue Ocean package (the-blue-ocean-positioning-map) for structure
2. Reference blueocean2.html for the quality bar on deliverable outputs
3. Generate all 7 files with research-grade content
4. Run QC validation (specificity weighted at 40%)
5. Report the final score before delivering

Agent Process — Section 00

Context Reset

Re-Read Before Every Package

Before starting ANY new package, re-read these instructions from the beginning. Context drift accumulates across packages — frameworks from Package A bleed into Package B. Fresh read = fresh start.

Mandatory Pre-Flight Checklist

Re-read this entire process — Do not rely on memory from previous packages
Re-read the PRD — Refresh the system architecture and golden examples
Clear mental slate — The new package has its own domain, frameworks, and requirements
Web search for THIS deliverable — Do not reuse framework research from other packages

This prevents the #1 failure mode: copying patterns from the last package instead of building what's correct for this one.

Agent Process — Section 01

Role & Objective

Identity

You are a Package Generator Agent
You create workflow packages that enable consultants and AI agents to produce client-ready deliverables
You follow the 7-file package structure exactly

Goal

Generate a complete 7-file workflow package for the specified deliverable
All files follow design system and quality standards
Package passes QC checklist at 90%+ score

Output Format

7 HTML files + 1 TXT file per package
Saved to: /[output-path]/[firm-name]/the-[package-name]/
File naming: numbered prefixes (00-, 01-, 02-, etc.)

Agent Process — Section 02

Execution Context

Reference Files

Before generating, read these files for structure and quality reference:

File	Purpose
`02-context.html`	Input manifest — what data you need to collect
`03-example.html`	Golden reference — Blue Ocean package structure
`04-quality.html`	QC criteria — validation checklist

Output File Structure

the-[package-name]/ ├── 00-START-HERE.html ├── 01-process-agent.html ├── 01-process-consultant.html ├── 02-context.html ├── 03-example.html ├── 04-quality.html └── quick-start-prompt.txt

Agent Process — Section 03

Gather Inputs

Collect the following inputs before generating. Ask user one section at a time.

Package Definition

Package Name Required — Format: "The [Name]" (e.g., "The Competitor Analysis")
Package Description Required — 1-2 sentences describing deliverable purpose
Firm Name Required — "The [Name] Firm" (e.g., "The Growth Firm")

Output Configuration

Output Folder Path Required — Base path for saving (e.g., "/Outputs")
Design System Override Optional — Custom colors/fonts if not using default

Content Guidance

Example Company Type Recommended — Industry/type for fictional example (e.g., "B2B SaaS")
Key Sections Recommended — Main sections the deliverable should include
Similar Deliverables Optional — Reference packages to study for structure

Agent Process — Section 04

Example-First Generation

Critical: Build the example FIRST, then derive all other files from it. This prevents framework drift.

Why Example-First

Without this sequence, you'll copy structure from unrelated packages and apply wrong frameworks. ERRC belongs in Blue Ocean, not Lean Six Sigma. DMAIC belongs in process improvement, not market analysis. Build the example with correct domain frameworks, then extract everything else from it.

Step 1: Web Search for Domain Frameworks

Before Creating Any Files

Search: "[deliverable type] framework methodology best practices"
Identify 3-5 core frameworks practitioners actually use
Note key terminology, standard structures, expected outputs
Flag industry-specific variations
DO NOT borrow frameworks from other packages — find what's correct for THIS deliverable

Step 2: Create 03-example.html FIRST

Build the Golden Example Before Anything Else

Apply the correct domain frameworks you identified in Step 1
Use a fictional but realistic client scenario
Include specific data, numbers, quotes to show research-backed quality
Make it complete enough to serve as a template
Validate: would a practitioner recognize this as correct methodology?

Step 3: Derive Process Files from Example

Reverse-Engineer from What You Built

Create 01-process-agent.html: What steps produced the example?
Create 01-process-consultant.html: Human-readable walkthrough of same process
Document the frameworks actually used (from Step 1, visible in Step 2)
Identify the logical sequence of analysis

Step 4: Derive Context from Example

What Inputs Made the Example Possible?

Create 02-context.html by examining what data the example needed
Required: What was essential to produce the example?
Recommended: What would have made it better?
Optional: What adds depth when available?

Step 5: Define Quality from Example

What Makes the Example Good?

Create 04-quality.html by analyzing success criteria
Specificity (40%): What specific data does the example contain?
Completeness (25%): Which sections are non-negotiable?
Validation (20%): What evidence supports the recommendations?
Presentation (15%): What formatting standards apply?

Step 6: Write Entry Files Last

Now You Know What the Package Does

Create 00-START-HERE.html: Explain when to use, what it produces, how to begin
Create quick-start-prompt.txt: Bootstrap prompt pointing to process file
Test: fresh LLM session with quick-start should work without clarification

Agent Process — Section 05

File Specifications

Detailed requirements for each file type.

File 1: 00-START-HERE.html

Required Sections

What This Workflow Produces — bullet list of outputs
Estimated Time — 3-column grid (Setup, Input Collection, Generation)
Cloud Mode — step-by-step for Claude/ChatGPT Projects
Local Mode — step-by-step for Claude Code CLI
Files in This Package — table with links and badges
Footer navigation linking all package files

File 2: 01-process-agent.html

Required Sections

Section 01: Role & Objective — identity, goal, output format
Section 02: Execution Context — reference files, output paths
Section 03: Gather Inputs — full manifest with priority tags
Section 04: Generate Deliverable — section specs, format rules
Section 05: Quality Validation — QC process, pass threshold
Section 06: Output Delivery — save location, delivery summary

File 3: 01-process-consultant.html

Required Sections

Time Estimates — banner with total time + phase breakdown
3-Step Process — Gather Context → Build Output → Quality Control
Interview vs Research Modes — when to use each
Sample Prompts — copy/paste prompts for each phase
What If Context Is Incomplete — handling gaps, minimum viable
Package Files — linked list with badges

File 4: 02-context.html

Required Sections

5-6 logical input sections (varies by deliverable type)
Each input has: name, priority tag, description, format hint, example
Priority tags: Required (red), Recommended (orange), Optional (gray)
25-35 total inputs typical

File 5: 03-example.html

Required Characteristics

Complete deliverable — all sections fully populated, NO placeholders
Fictional but realistic company — appropriate to deliverable type
Matches design system — same styling agents will produce
Shows quality bar — level of specificity and depth expected

Critical

The example file IS the template. Agents replicate its structure exactly. Every section, every data point must be present and realistic.

File 6: 04-quality.html

Required Sections

Scoring dashboard — sticky header with section scores + total
4 weighted dimensions: Completeness (40%), Specificity (30%), Validation (20%), Presentation (10%)
25-30 checkable criteria with Pass/Fail buttons
Interactive JavaScript for score calculation
Verdict thresholds: Approved (90%+), Minor Revisions (75-89%), Rejected (<75%)

File 7: quick-start-prompt.txt

Template

You have been loaded with the [Package Name] workflow.

Read the 01-process-agent file and begin execution. Start by asking me for the required context inputs, one section at a time. After gathering all inputs, generate the complete [deliverable name] following the 03-example structure. Run QC validation against the 04-quality criteria and report the final score before delivering.

My client is: [CLIENT NAME]

Agent Process — Section 06

Quality Validation

After generating all files, validate against the QC checklist in 04-quality.html.

Pass Threshold

Approved (90%+): Proceed to delivery
Minor Revisions (75-89%): Fix failed criteria, re-run QC. Max 2 cycles.
Rejected (<75%): Regenerate entire package. If rejected twice, escalate.

Hard Rule

Do not deliver a package that scores below 90%. Iterate until Approved or escalate to human review.

Agent Process — Section 07

Output Delivery

Save Location

/[output-path]/[firm-name]/the-[package-name]/

Delivery Summary Format

Report Template

## Package Generated: [Package Name]

**Location:** /[path]/the-[package-name]/
**Files Created:** 7
**QC Score:** [XX]%
**Verdict:** Approved

### Files
- 00-START-HERE.html
- 01-process-agent.html
- 01-process-consultant.html
- 02-context.html
- 03-example.html
- 04-quality.html
- quick-start-prompt.txt

Consultant Guide

3-Step Process

Human-readable process for creating workflow packages. Use this guide when running the Agent Builder manually or supervising AI execution. Total time: approximately 30 minutes.

1

Gather Context

Collect all inputs needed to generate the package. Reference the Context Loading section below for the full input manifest.

Package name and description
Target firm (Growth, Revenue, Talent, etc.)
Output folder path
Example company type (B2B SaaS, professional services, etc.)
Key sections the deliverable should include

2

Generate Package

Use AI to generate all 7 files. Reference the Blue Ocean package for structure.

Load the Agent Builder context into Claude/ChatGPT
Provide the gathered inputs
Review each file as it's generated
Request revisions for any issues before moving to next file

3

Quality Control

Validate the generated package against the QC checklist.

Run through all criteria (Completeness, Specificity, Validation, Presentation)
Score must reach 90%+ to approve
Fix any failed criteria and re-validate
Document the final score in delivery summary

Consultant Guide

Interview vs. Research Mode

Interview Mode

When you have access to a subject matter expert who can answer questions.

Faster — get answers directly
More accurate — SME knows the domain
Best for: new deliverable types, complex domains

Research Mode

When you need to research best practices independently.

Self-directed — no dependencies
Scalable — can batch multiple packages
Best for: common deliverables, well-documented domains

Consultant Guide

Sample Prompts

Initial Setup

Copy This Prompt

Read the Agent-Builder folder in # Production System. Then read the Blue Ocean package (the-blue-ocean-positioning-map) as your structural reference. You'll be generating a new workflow package.

Package Generation

Copy This Prompt

Generate a workflow package called "The [Package Name]" for The [Firm] Firm.

Description: [1-2 sentences describing the deliverable]

Save to: /Outputs/[firm-name]/the-[package-name]/

Use a fictional [industry type] company for the example. Follow the Blue Ocean structure exactly.

QC Validation

Copy This Prompt

Run the QC checklist from 04-quality.html against the package you just generated. Report the score for each dimension and the overall verdict. If any criteria fail, explain what needs to be fixed.

Consultant Guide

What If Context Is Incomplete?

Sometimes you won't have all the inputs. Here's how to handle gaps:

Missing Input	Workaround
Package description unclear	Research similar deliverables, propose structure, get approval
No example company type specified	Default to B2B SaaS for Growth/Revenue Firm, professional services for Talent Firm
Key sections unknown	Research best practices for that deliverable type, propose sections
Output path not specified	Use default: /Outputs/[firm-name]/the-[package-name]/

Minimum Viable Package

At minimum, you need:

Package name
1-sentence description
Target firm

Everything else can be researched or defaulted.

Context Loading

Context Manifest

All inputs needed to generate a workflow package. Gather these before running the Agent Builder.

Section 1: Package Definition

Core information about what package to create.

Package Name Required

The name of the deliverable package to create. Should follow "The [Deliverable Name]" convention.

Format

The [Name]

Example

The Blue Ocean Positioning Map

Package Description Required

A 1-2 sentence description of what this deliverable is and what it produces.

Format

1-2 sentences describing purpose and outputs

Example

A competitive positioning analysis that maps your offering against competitors across key value dimensions, identifying blue ocean opportunities.

Firm Name Required

Which firm this package belongs to. Determines folder structure and header branding.

Format

The [Name] Firm

Options

The Growth Firm, The Revenue Firm, The Talent Firm

Section 2: Output Configuration

Where and how to save the generated files.

Output Folder Path Required

The base path where package folders should be created. Packages save to /[path]/[firm-name]/the-[package-name]/.

Format

Absolute or relative file path

Example

/5. Outputs/Firms

Section 3: Content Guidance

Direction for the deliverable's content and structure.

Example Company Type Recommended

The type of fictional company to use in the golden example. Should be appropriate for the deliverable type.

Format

Industry/business model description

Example

B2B SaaS company selling to mid-market

Key Sections Recommended

Main sections the deliverable should include. If not provided, agent will research best practices.

Format

List of section names

Example

Value Dimensions, Competitor Profiles, Positioning Map, Strategic Recommendations

Reference Packages Optional

Existing packages to study for structure or style inspiration.

Format

List of package names or paths

Example

the-blue-ocean-positioning-map, the-ideal-customer-profile

Section 4: Design System

Visual identity specifications. Defaults apply if not overridden.

Color Palette Override Optional

Custom colors if not using the default palette.

Format

Named colors with hex values

Default

Noir #0a0a0a, Ivory #f5f4f0, White #ffffff, Charcoal #1a1a1a, Stone #8a8680

Typography Override Optional

Custom fonts if not using the default stack.

Format

Font names for headings, body, code

Default

Headings: DM Serif Display, Body: DM Sans (300), Code: JetBrains Mono

Section 5: Execution Options

How the Agent Builder should run.

Execution Mode Optional

Whether to run in Claude Code (local) or Claude/ChatGPT Project (cloud).

Options

local or cloud

Default

local (Claude Code)

Generate Reports Optional

Whether to generate Quality Report and System Improvement Report after package creation.

Format

true or false

Default

true

Example Output

Golden Example — The Blue Ocean Positioning Map

The Blue Ocean Positioning Map package serves as the gold standard reference. The blueocean2.html output demonstrates research-backed quality — specific data, buyer quotes, competitive intel, risk treatment.

This package demonstrates the complete 7-file structure with all elements properly implemented. Use it as your structural template when generating new packages.

Why This Package?

Blue Ocean demonstrates the upgraded 7-file structure with numbered prefixes, dual process docs (agent + consultant), and quick-start prompt. The blueocean2.html output shows what deep research produces vs. a speed-run: specific stats ($166B market, 3% vs 90% completion), direct buyer quotes, competitive intel with sources, and risk scenarios with mitigations.

Structure Overview

Every package follows this exact file structure:

the-[package-name]/ ├── 00-START-HERE.html # Entry point, setup guide ├── 01-process-agent.html # Machine-readable instructions ├── 01-process-consultant.html # Human-readable guide ├── 02-context.html # Input manifest ├── 03-example.html # Completed deliverable ├── 04-quality.html # QC checklist └── quick-start-prompt.txt # Bootstrap prompt

7

Files

6

Sections per Process

25+

Context Inputs

25+

QC Criteria

What to Replicate

When generating a new package, ensure these elements match the Blue Ocean reference:

00-START-HERE.html

Hero with package title and description
"What This Workflow Produces" — bulleted output list
Estimated Time — 3-column grid
Cloud Mode setup steps
Local Mode setup steps
Files in This Package — linked table with badges
Footer with all file links

01-process-agent.html

Section 01: Role & Objective — identity, goal, output format
Section 02: Execution Context — reference files, paths
Section 03: Gather Inputs — full manifest with priority tags
Section 04: Generate Deliverable — section specs
Section 05: Quality Validation — QC process
Section 06: Output Delivery — save location, summary

01-process-consultant.html

Time banner with total + phase breakdown
3-step process cards (Gather → Build → QC)
Interview vs Research mode comparison
Sample prompts for each phase
What If Context Is Incomplete section
Package files list with badges

02-context.html

5-6 logical input sections
Each input: name, priority tag, description, format, example
Priority tags: Required (red), Recommended (orange), Optional (gray)
25-35 total inputs

03-example.html (or direct deliverable output)

Complete deliverable — NO placeholders
Research-backed content: specific data, quotes, competitive intel
Buyer voice: direct quotes from real sources
Risk treatment: what could go wrong + mitigations
All sections fully populated with evidence
Matches design system exactly

04-quality.html

Sticky score dashboard
4 dimensions: Completeness (40%), Specificity (30%), Validation (20%), Presentation (10%)
25-30 checkable criteria
Interactive Pass/Fail buttons
JavaScript score calculation
Verdict thresholds (90%+ Approved)

quick-start-prompt.txt

References 01-process-agent file
Specifies execution sequence
Includes [CLIENT NAME] placeholder
Under 100 words

Design System Reference

All packages must follow this visual identity:

Colors

Noir: #0a0a0a — headers, badges, code blocks
Ivory: #f5f4f0 — page background
White: #ffffff — cards, content areas
Charcoal: #1a1a1a — body text
Stone: #8a8680 — secondary text, labels
Border: #d8d4cc — dividers

Typography

Headings: DM Serif Display, serif
Body: DM Sans, font-weight: 300
Code: JetBrains Mono, monospace

Badge Colors by File Type

Start Here: Orange (#d97706)
Process (Agent/Consultant): Purple (#8b5cf6)
Context: Blue (#1d4ed8)
Example: Charcoal (#1a1a1a)
Quality: Green (#16a34a)

Quality

Quality Scoring

Quality criteria for validating a generated 7-file workflow package. The gate check must pass before weighted scoring begins.

Gate Check — Framework Fit

This gate must pass before weighted scoring begins. If any item fails, STOP and rebuild the 03-example.html using the correct frameworks for this deliverable type.

Gate Check Items

All three must pass to unlock weighted scoring:

1. Web search was performed to identify correct domain frameworks
2. 03-example uses only frameworks appropriate to this deliverable domain
3. No cross-domain framework contamination (e.g., ERRC in Six Sigma, DMAIC in Strategy)

Gate Failed — Required Action

STOP all further work. Return to Step 1 of the example-first workflow. Perform a web search to identify correct frameworks for this deliverable type, then rebuild 03-example.html from scratch using the correct methodology.

Failure Response Protocol

1st

Fix & Revalidate

Address specific failures, make targeted fixes, run QC again on failed sections only

2nd

Rebuild Section

Same section fails twice → rebuild that file from scratch using the 03-example as reference

3rd

Human Review

Same section fails 3x → escalate to human review, likely a systemic issue with the package design

Dimension 1: Completeness (40% weight)

All 7 files were created (00-START-HERE, 01-process-agent, 01-process-consultant, 02-context, 03-example, 04-quality, quick-start-prompt)

Files saved to correct folder: /[firm-name]/the-[package-name]/

File naming uses numbered prefixes (00-, 01-, 02-, etc.)

00-START-HERE has: output preview, time estimates, cloud mode, local mode, file table

01-process-agent has all 6 sections (Role, Context, Inputs, Generate, QC, Delivery)

01-process-consultant has: time banner, 3-step process, modes, prompts, gap handling

02-context has 25+ inputs with priority tags across 5+ sections

03-example is a complete deliverable with NO placeholders

04-quality has 25+ criteria across 4 weighted dimensions

quick-start-prompt.txt is present and under 100 words

Dimension 2: Specificity (20% weight)

03-example uses a realistic fictional company appropriate to the deliverable type

03-example data is plausible (realistic numbers, metrics, company details)

02-context inputs are specific to this deliverable (not generic)

02-context includes realistic example values for each input

04-quality criteria are relevant to this specific deliverable type

01-process-agent instructions are actionable (not vague)

01-process-consultant sample prompts are copy/paste ready

Dimension 3: Fit for Purpose (20% weight)

03-example uses frameworks appropriate to this deliverable type (not borrowed from unrelated packages)

A domain practitioner would recognize the methodology as correct for this deliverable

Framework terminology matches industry standards (e.g., DMAIC for Six Sigma, not ERRC)

01-process-agent instructions align with how practitioners actually execute this deliverable

02-context inputs request data that's actually needed for this methodology

Dimension 4: Validation (10% weight)

All internal links between files work correctly

Footer navigation present on all HTML files

04-quality JavaScript calculates scores correctly

Files render correctly in browser (no broken styles)

Dimension 5: Presentation (10% weight)

Header shows correct firm name

Doc badges use correct colors (Start: orange, Process: purple, Context: blue, Example: charcoal, Quality: green)

Color palette matches design system (noir, ivory, white, charcoal, stone)

Typography correct (DM Serif Display headings, DM Sans 300 body, JetBrains Mono code)

Voice is professional, confident, and direct (no hedging)

Verdict Thresholds

Score	Verdict	Action
90%+	Approved	Proceed to delivery
75-89%	Minor Revisions	Fix failed criteria, re-run QC. Max 2 cycles.
<75%	Rejected	Regenerate entire package. If rejected twice, escalate to human review.