Turn raw community conversations into validated customer intelligence.
CustomerMind is a production-ready, three-stage reasoning pipeline engineered to ingest raw, unstructured customer conversations—Reddit threads, Discord chat logs, App Store review dumps, and support tickets—and transform them into structured, evidence-anchored product and marketing assets.
CustomerMind is structured around three sequential reasoning modules supported by a permanent Knowledge Layer containing customer psychology and direct-response copywriting frameworks, a Reasoning Layer ensuring strict evidence-anchoring, and a Decision Layer executing confidence calibration.
graph TD
%% Input Node
Input[/"Raw Community Data <br> (Reddit, Reviews, Support Tickets)"/] --> Prep["Data Pre-processor <br> (Text Normalization & Sarcasm Filter)"]
%% Module 1
subgraph M1 ["Module 1: Sentiment Extractor"]
Prep --> Noise["Noise Filter <br> (Strip Spam / Bots)"]
Noise --> Sentiment["Sentiment Tagging <br> (Frustrated, Negative, Aspirational, Positive, Neutral)"]
Sentiment --> Clustering["Topic Clustering <br> (>60% Keyword Overlap Merging)"]
end
%% Module 2
subgraph M2 ["Module 2: Pain Intelligence Engine"]
Clustering --> Pains["Root-Cause Analysis <br> (Friction, Hidden Costs, Workarounds)"]
Pains --> JTBD["Jobs-To-Be-Done Mapping <br> (Functional, Emotional, Social Jobs)"]
JTBD --> Rank["Pain Weighting <br> (Frequency × Intensity)"]
end
%% Module 3
subgraph M3 ["Module 3: Voice & Messaging Generator"]
Rank --> Glossary["Vocab Extractor <br> (Voice of the Customer Glossary)"]
Glossary --> CopyGen["Copywriting Synthesis <br> (PAS Headlines, Hooks, Value Prop)"]
CopyGen --> Recs["Strategic Recommendations <br> (Product, Marketing, Competitive)"]
end
%% Layers and Audits
Knowledge[("Knowledge Layer <br> (15 Master Frameworks)")] -.-> |Inject Rules| M2
Knowledge -.-> |Inject Rules| M3
Recs --> Audit{"Self-Review Protocol <br> (Rubric Score >= 4/5?)"}
%% Output
Audit -->|Pass| OutputReport[\"Structured Markdown Report <br> + Interactive HTML Dashboard + PDF\"\]
Audit -->|Fail| Prep
%% Styling
style Input fill:#e1f5fe,stroke:#03a9f4,stroke-width:2px;
style M1 fill:#efebe9,stroke:#8d6e63,stroke-width:2px;
style M2 fill:#fff8e1,stroke:#ffb300,stroke-width:2px;
style M3 fill:#e8f5e9,stroke:#4caf50,stroke-width:2px;
style Knowledge fill:#f3e5f5,stroke:#9c27b0,stroke-width:2px;
style Audit fill:#ffebee,stroke:#e91e63,stroke-width:2px;
style OutputReport fill:#e0f2f1,stroke:#009688,stroke-width:3px;
The CustomerMind Claude Skill and CLI installer were designed, calibrated, and validated using a structured database of 199 distinct academic, product engineering, and linguistic sources across 5 research modules.
199 RESEARCH SOURCES
│
┌──────────────────┬─────────────────────┼─────────────────────┬──────────────────┐
▼ ▼ ▼ ▼ ▼
[Mod 1] [Mod 2] [Mod 3] [Mod 4] [Mod 5]
Claude Skills EDA Product Psychology & Natural Language Linguistic Brain
Architecture Blueprints JTBD Frameworks Engineering (AI-Dev Slang)
39 Sources 42 Sources 21 Sources 30 Sources 67 Sources
| Research Area | Source Count | Core Methodology Integrated | Key Focus Areas |
|---|---|---|---|
| 1. Architecture & Knowledge Design | 39 | Progressive disclosure, cognitive routing, agentic subagent orchestration, and memory context limit management. | Model Context Protocol (MCP), Claude Code workflows, .claude folder structures, and LLM reasoning benchmarks. |
| 2. Enterprise Deliverable Architect (EDA) | 42 | Modular capability design for structured programmatic outputs (PDF, DOCX, XLSX, PPTX). | Secure enterprise MCP servers, API gateway security, and Salesforce/Recruitly GTM automation integrations. |
| 3. Customer Psychology & JTBD | 21 | The Jobs-To-Be-Done (JTBD) matrix, buying trances theory, and consumer psychology. | Mapping functional, emotional, and social dimensions of customer intent from raw text. |
| 4. Natural Language Generation (NLG) | 30 | Syntax diversity algorithms, active voice enforcement, and AI cliche identification. | The Forbidden Language Database designed to eradicate synthetic LLM cliches ("delve", "tapestry", "skyrocket"). |
| 5. The Customer Language Brain | 67 | Exhaustive linguistic and sentiment analysis of modern AI-assisted engineering communities. | Tracking real-world developer frictions, token wastes, vibe coding pitfalls, and editor migrations (bolt.new, Cursor Composer, Claude Code). |
CustomerMind/
├── README.md ← You are here
├── LICENSE ← MIT License
├── CHANGELOG.md ← Version history tracking
├── CONTRIBUTING.md ← Development rules
├── .customermind.json ← Integrity config manifest
│
├── skill/ ← The Claude Skill Brain
│ ├── system_prompt.md ← 1108-line complete production system prompt
│ ├── architecture.md ← Pipeline routing & dependency specifications
│ ├── module_reference.md ← Module I/O specifications
│ ├── reasoning_rules.md ← Evidence-anchoring and opportunity scoring rules
│ ├── output_contract.md ← Standardized 10-section layout template
│ └── guardrails.md ← Forbidden language & edge case protocols
│
├── knowledge/ ← Permanent Skill Knowledge Base
│ ├── 01_customer_psychology.md ← Buying trances and cognitive biases
│ ├── 02_pain_intelligence.md ← Root cause extraction models
│ ├── 03_customer_language.md ← Linguistic markers of intense friction
│ ├── 04_ai_cliches.md ← Syntactic patterns of AI slop
│ ├── 05_brand_voice.md ← Conversational voice guidelines
│ ├── ... ← Additional framework documentation
│ └── 15_glossary.md ← Customer vocabulary baseline glossary
│
├── packages/ ← Monorepo Workspace Packages
│ ├── cli/ ← Node.js CLI & Interactive Installer
│ │ ├── src/commands/analyze/ ← Analyze command (Reddit API + LLM pipeline)
│ │ ├── src/utils/filesystem/ ← Transactional filesystem manager (with rollback)
│ │ └── tests/ ← Test suite for installer, parser, and command actions
│ └── dashboard/ ← Next.js Dashboard Web App
│ ├── app/ ← React/Tailwind frontend layout
│ └── public/ ← Static charts and dashboard assets
│
├── research/ ← Analysis Workspaces
│ ├── raw/ ← Raw community inputs (pasted comments, Reddit exports)
│ └── reports/ ← Validated markdown reports
│
├── scripts/ ← Python Validation & Utility scripts
│ ├── validate_outputs.py ← Validates markdown reports against output contract
│ ├── score_quality.py ← Applies the 7-dimension evaluation rubric
│ └── generate_reports.py ← Batch prepares analysis files
│
└── tests/ ← Benchmark & Evaluation Suites
├── benchmark_cases/ ← Reference evaluation cases
├── edge_cases/ ← anomalous datasets (spam, sarcasm heavy)
└── regression_tests/ ← Core regression suites
- Open
skill/system_prompt.mdand copy its entire content. - Paste it as a System Prompt inside a Claude Project or as the initial instruction in a chat thread.
- Paste your raw, unstructured text (Reddit threads, App Store reviews, customer chats).
- CustomerMind will execute and return a verified Markdown report.
Navigate to the CLI workspace and install dependencies:
cd packages/cli
npm install
npm run buildLink the package globally to access the executable:
npm linkInstall CustomerMind configurations, directories, and knowledge files into any workspace:
npx create-customermindThe installer automatically detects your operating system, Node environment, and copies the permanent skill assets into your workspace with full transactional rollback safety.
Query Reddit discussions programmatically and trigger the LLM analysis pipeline:
customermind analyze "Cursor AI" --posts 100 --time monthThis single command:
- Crawls Reddit for conversations about the specified topic.
- Saves the raw dataset to
research/raw/cursor_ai_input.txt. - Runs the raw text through the Claude/Gemini API using the system prompt.
- Generates a fully formatted markdown report at
research/reports/cursor_ai_report.md. - Compiles a PDF copy at
reports/dashboard/report.pdf. - Generates an interactive, visual HTML dashboard at
reports/dashboard/index.html.
To ensure professional-grade, human-sounding deliverables, CustomerMind applies rigid guardrails:
Important
No Hallucinations: Every customer pain point, quote, or phrase featured in the reports must map to a verbatim snippet in the raw text database. If no quote exists, the claim is rejected.
Warning
AI Slop Eradication: All generated copy is verified against the Forbidden Language Database containing 50+ banned words ("delve", "tapestry", "revolutionize") and structural rules (e.g. sentence length variation, active voice).
Tip
Data Quality Protocols: When raw input is sparse (<5 comments) or consists mostly of noise, CustomerMind skips the report generation step and activates the Elicitation Protocol, warning you of insufficient data quality.
You can verify the integrity of any generated Markdown report using the provided Python scripts:
Run Output Contract Validation:
python scripts/validate_outputs.py research/reports/scenario1_report.mdRun Rubric Quality Scorer:
python scripts/score_quality.py research/reports/scenario1_report.mdRun the CLI unit/integration tests:
cd packages/cli
npm test