Context Builder

Context Builder

Generate interactive context-building prompts for consulting clients. These prompts are designed to be run in Claude Code -- they guide a team through structured questions using AskUserQuestion, generate output files per section, and compile everything into a reusable CLAUDE.md.

Workflow

Phase 1: Intake (AskUserQuestion)

Ask all intake questions using AskUserQuestion with closed-list options. Gather:

Question 1: Company identifier

• Options: "I have a website URL", "I have a company name", "I have both"
• Follow up to get the actual URL/name

Question 2: Who will use this prompt?

• Options: "Specific person (name + role)", "A team (no specific person)", "Unknown / TBD"
• If specific person: follow up for name and role

Question 3: Primary consulting focus (multiSelect)

• "AI automation of current operations"
• "Existential strategy (what survives AI)"
• "New business models / pivots"
• "Product development with AI"

Question 4: Industry

• "Marketing / Advertising"
• "Manufacturing / Construction"
• "SaaS / Software"
• "Professional Services / Consulting"
• (Other)

Question 5: Existing context in vault?

• "Yes, there's a call transcript"
• "Yes, there are notes/files"
• "No existing context"
• If yes: ask for filename or search term to locate it

Question 6: Session language

• "Russian (questions in Russian, output in English)"
• "English throughout"
• "Other"

Phase 2: Research (automated)

Run these research steps in parallel where possible:

1. Web research: Use WebSearch and WebFetch (via Task agent) to gather:

• What the company does, products/services
• Target market, company size, geography
• Tech stack, partnerships
• Recent news, funding, team info
• Competitive landscape

2. Vault search: Search the Obsidian vault for:

• Transcripts mentioning the company name (Grep in vault root and Daily/)
• People files for contacts at the company (People/ folder)
• Any existing notes or research

3. Transcript analysis (if found): Extract from call transcripts:

• Team members and their roles
• Current AI tool usage
• Pain points and concerns mentioned
• Specific processes described
• Questions raised by the team

Phase 3: Section Selection (AskUserQuestion)

Present a curated set of sections based on the consulting focus. Use AskUserQuestion with multiSelect to let the user pick which sections to include.

Section Library

Draw from references/section-library.md for the full section catalog. Default section sets by focus:

AI Automation focus: 1. Process Inventory, 2. Pain Points & Waste, 3. Current Tech Stack, 4. AI Opportunity Mapping, 5. People & Org, 6. Data Reality Check, 7. Quick Wins

Existential Strategy focus: 1. Revenue & Service Map, 2. The Existential Question, 3. Client Value Chain, 4. New Business Models, 5. Data & Knowledge Assets, 6. People & Org, 7. Quick Wins & Pilots

Full Assessment (both): All 10 sections from the library.

After section selection, ask:

Express mode grouping: Present a suggested grouping of selected sections into 4 Express mega-sections. Let user confirm or adjust.

Phase 4: Generation

Generate two files:

1. The Context-Builder Prompt

Save to: Claude-Drafts/{company-slug}-context-prompt.md

Structure (follow the template in references/prompt-template.md):

---
created_date: '[[YYYYMMDD]]'
type: draft
topic: consulting, AI transformation, {industry}
for: {contact person or team name}
---

# AI Transformation Context Builder -- {Company Name}

## About {Company}
  [Generated from research -- company description, size, market, positioning]

## Current State
  **What's working:** [from research + transcript]
  **The gap:** [from research + transcript]
  [If existential concerns found: **Existential context:**]

## Mode Selection
  [Express vs Deep Dive with section descriptions]

## How This Works
  [Standard interactive session instructions]

## Session Resumability
  [Standard resumability logic]

## Interactive Flow
  [Selected sections with tailored questions]

## Output Files
  [One file per section + final CLAUDE.md]

## Relevant Frameworks
  [Selected from references/frameworks.md based on focus]

2. Instruction File (optional)

If the prompt will be sent to someone external, generate a short instruction file: Claude-Drafts/{company-slug}-context-instructions.md

Containing:

• What this file is and how to use it
• Prerequisites (Claude Code or similar)
• The two modes explained simply
• What they'll get on output
• Privacy note (they can share as much or as little as they want)

Phase 5: Delivery (AskUserQuestion)

Question: What to do with the generated files?

• "Save to vault only"
• "Save and send via Telegram"
• "Save and let me review first"

If Telegram: ask for the recipient handle/name, then send using the telegram skill (intro message + file).

Key Principles

• Maximize closed-list questions: Every AskUserQuestion should have concrete options. Minimize free-text input.
• Research before asking: Don't ask the user things that can be found via web search or vault search.
• Tailor sections to context: If the transcript reveals specific concerns (e.g., existential fears, specific tech stack), customize the section questions to reference those specifics.
• Bake in discovered context: The generated prompt's "About" and "Current State" sections should be rich with researched details so the person running the prompt gets a warm start.
• Language awareness: If session language is Russian, all AskUserQuestion interactions during prompt execution should be in Russian, but output files in English.

Resources

references/

• section-library.md -- Full catalog of available sections with question templates
• prompt-template.md -- Structural template for the generated prompt
• frameworks.md -- Consulting frameworks to selectively include

Consulting Frameworks

Selectively include in generated prompts based on consulting focus. Not all frameworks apply to every engagement.

Always Include

BCG 10/20/70 Rule

• 10% algorithms/models, 20% technology/data, 70% people/processes/culture
• Go narrow and deep: pick few high-value workflows, rethink entirely

Andrew Ng's Playbook

• Start with pilots, not strategy. Strategy comes from experience.
• Pilot -> Build team -> Train broadly -> Develop strategy -> Communicate

Include for Existential Strategy Focus

The "But" Heuristic

• When someone says "AI can automate everything, BUT..." -- the "but" is where current human value lives
• These "buts" may shrink over time, but they're the most reliable near-term moat
• Build strategy around making your "buts" bigger, not just optimizing what's automatable

The "Metro Newspaper" Test

• Don't orient on your filtered bubble's headlines; orient on what mainstream people know
• Most people still don't know what ChatGPT is -- you're in the avant-garde
• The real adoption curve is much earlier than tech Twitter suggests
• Practically: your competitors are probably not as AI-advanced as their press releases claim

Curiosity > Fear

• Teams driven by interest/curiosity adopt AI 10x more effectively than those driven by fear
• Internal reward (I explored something cool) is more sustainable than external pressure
• Find and amplify team members who genuinely enjoy experimenting
• Don't build AI strategy around anxiety; build it around capability and opportunity

Palantir Integration Model

• Don't just sell services; become the integration layer between client data + AI agents
• Offer AI-augmented workflows that combine proprietary data with domain expertise
• Human-in-the-loop at high-value decision points, automation for everything else
• Clients pay for the system, not the headcount

Include for Automation Focus

Value Stream Mapping (AI-adapted)

• Map current state -> Identify bottlenecks -> Design future state -> Monitor
• AI tools automate data collection and real-time updating
• Focus on cycle time reduction and handoff elimination

AI Readiness Canvas

• Strategic Imperative (Why): Vision, Value Proposition, Use Cases
• Foundational Capabilities (How): Data, Infrastructure, Skills, Governance

Deloitte AI Maturity Levels

• Starters -> Pathseekers -> Transformers
• Avoid "Underachiever" trap: many deployments, low outcomes
• Focus on few high-impact deployments rather than spraying AI everywhere

Include for Product Development Focus

The 15-Person Company Thought Experiment

• "What would a 15-person version of this company generating the same revenue look like?"
• Forces thinking about what's truly core vs what's operational overhead
• AI + small team often beats large team without AI on speed and cost

Fine-Tuning Economics

• Fine-tuning small models on domain data is becoming cheap ($10s-$100s)
• A fine-tuned small model can outperform GPT-4 on narrow tasks
• Proprietary training data is the moat, not the model architecture
• Consider: what training data do you have that nobody else has?

Distribution Advantage

• "Knowing how to sell and distribute is the skill AI can't vibe-code"
• Companies with distribution expertise + AI tools beat pure-AI startups
• If you can reach customers, you can test products rapidly
• Marketing/sales knowledge compounds; technical knowledge depreciates faster

Include for Regulated Industries

Compliance-First Automation

• Map regulatory boundaries before automating
• Some automations require certification or regulatory approval
• Build audit trails into every automated process
• Human sign-off at compliance checkpoints, automation everywhere else

Prompt Template

Structural template for generated context-builder prompts. Replace all {placeholders} with actual values. Sections marked [CONDITIONAL] are included only when relevant.

---
created_date: '[[{YYYYMMDD}]]'
type: draft
topic: consulting, AI transformation, {industry_keywords}
for: {contact_person_or_team}
---

# AI Transformation Context Builder -- {Company Name}

You are helping {audience_description} explore AI-driven transformation. {audience} is working with Gleb Kalinin (AI consultant) to {consulting_goal_summary}.

## About {Company Name}

{company_name} is a {company_type} (~{headcount} people) that:
- {bullet_1_core_business}
- {bullet_2_market}
- {bullet_3_products_or_services}
- {bullet_4_differentiator}
- {bullet_5_additional_context}

## Current State

**What's working:**
- {working_1}
- {working_2}
- {working_3}

**The gap:**
- {gap_1}
- {gap_2}
- {gap_3}

[CONDITIONAL: Include if existential concerns surfaced in research/transcript]
**Existential context (from team discussion):**
- {existential_1}
- {existential_2}

## Mode Selection

**At the start of the session, ask the user which mode they want using AskUserQuestion:**

### Express Mode (~15-20 min)
- Covers 4 combined mega-sections instead of {total_sections}
- 2-3 questions per section, focused on highest-signal information
- Generates a single combined output file + CLAUDE.md
- Best for: first pass, time-constrained, or when the user wants to get something down quickly

**Express sections:**
1. **{express_1_title}** -- {express_1_description} (combines Sections {express_1_combines})
2. **{express_2_title}** -- {express_2_description} (combines Sections {express_2_combines})
3. **{express_3_title}** -- {express_3_description} (combines Sections {express_3_combines})
4. **{express_4_title}** -- {express_4_description} (combines Sections {express_4_combines})

Express output: `context-output/express-context.md` + `context-output/CLAUDE.md`

### Deep Dive Mode (~60-90 min across multiple sessions)
- All {total_sections} sections with thorough exploration
- Follow-up questions and detailed output files
- Best for: comprehensive assessment, building full transformation brief

## How This Works

This is an **interactive context-building session**. You will guide the team through structured questions using the `AskUserQuestion` tool. Do NOT dump all questions at once. Instead:

1. **Ask one topic at a time** using AskUserQuestion with multiple-choice options where possible
2. **Follow up** on answers with clarifying questions before moving to the next topic
3. **Summarize** what you've learned after each section and confirm understanding
4. **Generate output files** as you go -- after each section, write the corresponding output file

The goal is to build a comprehensive context that future Claude Code sessions can use. Think of it as creating a personalized CLAUDE.md for the company's AI transformation.

[CONDITIONAL: Include if session language differs from output language]
**Language note:** {language_instruction}

## Session Resumability

**IMPORTANT: At the start of every session, check which output files already exist in `context-output/`.**

{resumability_checklist}

When resuming:
1. Read all existing output files to restore context
2. Tell the user: "Welcome back! I see we've already completed [sections]. Let's pick up with [next section]."
3. Briefly summarize what was captured in previous sections (from the files) before continuing
4. If the user wants to revise a completed section, allow it and re-generate that file

This means the team can stop at any point and come back later without losing progress.

## Interactive Flow

Guide the team through these sections in order. For each section, use AskUserQuestion to present structured choices, then follow up with open-ended questions based on answers.

{generated_sections}

## Output Files

Generate these files as you progress through the conversation. Write each file after completing its corresponding section. Save all files in a `context-output/` folder relative to where this prompt is run.

### Files to Generate:

{output_file_specs}

{total_sections + 1}. **`context-output/CLAUDE.md`** -- Final output
    - Personalized context file combining all findings
    - Can be dropped into any project as a CLAUDE.md for future sessions
    - Includes: company context, tech stack, current challenges, strategic options, terminology, key decisions ahead

## Relevant Frameworks

{selected_frameworks}

Generation Notes

When generating the actual prompt from this template:

1. Replace all placeholders with researched values 2. Customize section questions based on what was discovered in research/transcripts (reference specific tools, people, processes by name) 3. Remove [CONDITIONAL] markers and either include or exclude those sections 4. Number output files sequentially based on selected sections 5. Build the resumability checklist dynamically from the selected sections and their output files 6. Select frameworks from frameworks.md based on consulting focus 7. Keep the prompt self-contained: someone running it should not need any other files

Section Library

Catalog of sections for context-builder prompts. Each section includes: purpose, default questions, and output file spec. Select and customize based on consulting focus.

Core Sections

Revenue & Service Map

ID: revenue-service-map Purpose: Understand the business model and what generates money. Best for: All assessments, especially existential strategy. Questions:

• What are the main revenue streams? (list with rough proportions)
• For each service line: client type, pricing model, margin
• Which services are growing vs plateauing vs declining?
• Client lifecycle: acquisition -> onboarding -> delivery -> expansion/churn
• Pricing model: % of spend, retainer, project-based, success-based?

Output: context-output/01-revenue-service-map.md Output spec: Revenue streams with proportions, service lines table, growth trajectory

Process Inventory

ID: process-inventory Purpose: Map current workflows end to end. Best for: Automation focus. Questions:

• What are the main workflows? (list all core processes)
• For each: what triggers it, what data feeds it, steps, who executes each step?
• Which processes are formalized/documented vs tribal knowledge?
• Where do handoffs happen between people, systems, or both?

Output: context-output/01-process-map.md Output spec: Structured workflow descriptions, triggers/inputs/steps/outputs, documented vs undocumented flags

Pain Points & Waste

ID: pain-points Purpose: Identify where time/money/quality is lost. Best for: All assessments. Questions:

• What takes longer than it should?
• Where do errors happen most?
• What decisions are bottlenecks?
• What manual work feels obviously automatable?
• What information is hard to find when needed?
• What prevents scaling with the same headcount?

Output: context-output/02-pain-points.md Output spec: Pain Point Matrix: issue | severity (1-5) | frequency | current workaround. Sorted by impact.

Current Tech Stack

ID: tech-stack Purpose: Map the full technology and data landscape. Best for: All assessments. Questions:

• What software systems are used daily?
• What internal/custom tools exist?
• How do systems connect? (APIs, manual export, spreadsheets)
• What data exists but isn't being used?
• What integrations are fragile or manual?

Output: context-output/03-tech-stack.md Output spec: All systems with purpose, integrations, data formats. Connection map. Gaps highlighted.

The Existential Question

ID: existential Purpose: Honest strategic assessment of what survives AI transformation. Best for: Existential strategy focus. Questions:

• Which parts of the business could AI fully automate in 1-2 years? 3-5 years?
• What does a human here genuinely do that AI cannot today? (be concrete, not abstract)
• If platforms built perfect self-serve AI tools, what would clients still need you for?
• Is the current service model viable as a standalone business in 5 years?
• What would a team at 20% of current size generating the same revenue look like?
• Where is the "but" -- the thing you tried to automate but couldn't?

Output: context-output/04-existential-assessment.md Output spec: Automation timeline, defensibility analysis, risk assessment, strategic options with trade-offs

AI Opportunity Mapping

ID: ai-opportunities Purpose: Identify where AI/agents could help existing operations. Best for: Automation focus. Questions (customize by industry):

• Process monitoring and anomaly detection
• Dynamic scheduling and replanning
• Automated quality checks
• Knowledge capture from experienced staff
• Document/compliance checking
• Reporting and narrative generation
• Client communication assistance

Output: context-output/05-ai-opportunities.md Output spec: Prioritized list: opportunity | impact | effort | dependencies | timeframe. Grouped: quick wins, medium-term, strategic.

New Business Models

ID: new-business-models Purpose: Explore what the company could become. Best for: Existential strategy, pivots. Questions:

• What adjacent markets could be served with existing expertise + AI?
• Could parts of the service become self-serve tools?
• Could proprietary data/models become products?
• What business model experiments would you run with 3 months and a small team?
• What would the company look like if rebuilt from scratch today?

Output: context-output/06-new-business-models.md Output spec: Business model canvas sketches for top 2-3 options, experiment designs, resource requirements

People & Organizational Dynamics

ID: people-org Purpose: Understand the human dimension of transformation. Best for: All assessments. Questions:

• Who are the AI champions? Who is skeptical, and why?
• What's the current skill distribution? (advanced AI users vs basic vs none)
• What would change in team structure if AI handles 50% of manual work?
• Which roles grow more important with AI?
• How does the team feel emotionally about AI transformation?
• Appetite for learning vs "just tell me what to do"?

Output: context-output/07-people-org.md Output spec: Team AI readiness map, skill gap analysis, recommended structure evolution

Client Value Chain Analysis

ID: client-value-chain Purpose: Understand how value flows to clients and what's at risk. Best for: Service businesses, existential strategy. Questions:

• What do clients actually value most? (cost, expertise, convenience, speed, results, relationship?)
• Which client segments are most at risk of going direct/self-serve?
• Which segments would pay more for AI-augmented services?
• What do clients complain about or wish was different?
• What would make a client say "I'd never leave"?
• Adjacent services clients need that aren't currently offered?

Output: context-output/08-client-value-chain.md Output spec: Client segmentation by AI-risk, value proposition mapping (current vs future), deepening opportunities

Data & Knowledge Assets

ID: data-assets Purpose: Audit what proprietary assets exist and their defensibility. Best for: Product development, existential strategy. Questions:

• What historical data exists? (volume, timespan, quality)
• What expert knowledge lives only in people's heads?
• If you wanted to fine-tune a model on your expertise, what would training data look like?
• What competitive intelligence do you have that others don't?
• What would be needed to turn data into a defensible AI product?

Output: context-output/09-data-knowledge-assets.md Output spec: Data inventory: type | volume | quality | defensibility. Knowledge capture priorities. Productization path.

Make vs Buy Assessment

ID: make-buy Purpose: For specific technology/platform decisions. Best for: When there's a concrete build-or-buy decision pending. Questions:

• What capabilities are non-negotiable?
• What's unique to the company's process that off-the-shelf can't handle?
• What's commodity functionality?
• Cost of building wrong vs buying wrong?
• Timeline pressure?

Output: context-output/04-make-buy-scorecard.md Output spec: Capability breakdown: capability | must-have? | unique? | vendor coverage | build effort

Constraints & Compliance

ID: constraints Purpose: Map hard regulatory and operational boundaries. Best for: Regulated industries (manufacturing, healthcare, finance). Questions:

• Industry regulations affecting automation
• Safety requirements
• Quality certifications/standards
• Data privacy/security (GDPR, etc.)
• Insurance/liability considerations

Output: context-output/07-constraints-compliance.md Output spec: Regulatory requirements table, safety implications, certifications to maintain

Budget, Timeline & Active Projects

ID: budget-timeline Purpose: Ground the conversation in reality. Best for: When concrete implementation is expected. Questions:

• Budget envelope for AI/automation investment
• Active projects or client commitments constraining experimentation
• Hiring plans or resource constraints
• Key decision deadlines

Output: context-output/08-budget-timeline.md Output spec: Budget constraints, active commitments, decision deadlines, resource availability

Data Reality Check

ID: data-reality Purpose: Assess actual data quality vs assumed. Best for: Automation focus, when AI deployment is planned. Questions:

• What data is reliably captured today vs assumed to exist?
• Where are gaps: missing, inconsistent, unreliable?
• How much is in spreadsheets, emails, or people's heads?
• What instrumentation is needed for desired automations?

Output: context-output/09-data-reality.md Output spec: Data inventory: type | source | quality | format. Gaps needing instrumentation. Quick data wins.

Quick Wins & Pilots

ID: quick-wins Purpose: Define immediate next steps (4-8 weeks). Best for: All assessments (always include as final section). Questions:

• What manual tasks could be automated this month with existing tools?
• What 1-2 small experiments could test a new hypothesis?
• How to move more of the team from basic ChatGPT to compound AI workflows?
• Is there a client engagement where AI-augmented delivery could be tested?
• How would you measure success?

Output: context-output/10-pilots-quickwins.md Output spec: 2-3 immediate automations with owner/timeline/impact. 1-2 strategic experiments with hypothesis/method/criteria. Team activation plan.

Section Numbering

Output file numbers are assigned based on the order sections appear in the selected set. When fewer than 10 sections are selected, number them sequentially (01, 02, 03...) in the order they appear in the prompt. The final CLAUDE.md is always the last file generated.

Express Mode Grouping

When building Express mode, group selected sections into exactly 4 mega-sections. Suggested groupings:

1. What you do -- Revenue/process/tech stack/team sections 2. What hurts -- Pain points/constraints/data reality sections 3. What's defensible -- Existential/client value/data assets/new models sections 4. What to build -- AI opportunities/quick wins/pilots sections

Adjust labels and contents based on the specific sections selected.