Build Digital Twins with the immortal-skill Framework

Beyond Simple Chatbots: The Era of Digital Twins

Most AI builders struggle with the 'vibe' problem. You can prompt a model to 'act like an expert,' but it lacks the specific procedural nuances, memories, and communication patterns that define a real individual. Whether you are building an AI mentor that actually teaches like your favorite professor or a digital avatar of yourself to handle routine Slack inquiries, generic prompting isn't enough.

Enter immortal-skill: a professional-grade framework for digital immortality and persona distillation. Developed by agenmod, this modular skill provides a structured pipeline for extracting, refining, and packaging the essence of a person—based on real evidence—into a deployable AI building block.

Seven Blueprints for Any Use Case

The framework doesn't treat every person the same. It recognizes that a "Colleague" twin should focus on work style and procedures, while a "Family Member" twin focuses on shared memories and life wisdom. The skill provides seven distinct role templates:

1. Self: A full-dimensional digital twin.
2. Colleague: Workflow, communication style, and professional decision-making.
3. Mentor: Pedagogy and intellectual frameworks.
4. Family: Heritage, shared stories, and core values.
5. Partner/Ex: Relationship dynamics and emotional memory (strictly for private, ethical use).
6. Friend: Inside jokes and shared history.
7. Public Figure: Methodologies based on public records and open source data.

How to Install and Initialize

Getting started with immortal-skill is straightforward through the Lovable interface. You can add the framework to your project environment with a single command:

lovable add immortal-skill

Once added, the skill initializes a directory structure designed to handle the multi-phase distillation process, from raw data intake to the final SKILL.md assembly.

The Multi-Phase Distillation Pipeline

Building a high-fidelity digital twin requires more than a simple RAG (Retrieval-Augmented Generation) setup. immortal-skill follows a rigorous 7-phase process:

Phase 1: Ethical Guardrails

Before a single byte of data is processed, the framework requires ethical confirmation. This ensures that personal information is handled with consent and that the resulting AI isn't used for harassment or deceptive impersonation. It includes specific protocols for different roles, such as ensuring data de-sensitization for work colleagues.

Phase 2: Multi-Platform Data Intake

The framework provides a unified CLI to pull data from over 12 platforms. It supports:

• SaaS Platforms: Slack, Discord, Telegram, and Email.
• Local Databases: WeChat (via SQLite) and iMessage (macOS).
• Archives: WhatsApp exports, Twitter/X archives, and Google Takeout.

Phase 3-4: Feature Extraction and Conflict Resolution

This is where the magic happens. The framework mines data for four specific dimensions:

• Procedural: How they solve problems.
• Interaction: Their linguistic quirks and conversational rhythm.
• Memory: Key events and chronological facts.
• Personality: Core traits and emotional baselines.

Each extracted point is tagged with an "Evidence Level" (verbatim, artifact, or impression), ensuring that the AI knows the difference between a direct quote and an inferred trait. If the data contains contradictions—like a person changing their opinion over five years—the merge-policy identifies these as conflicts for the user to resolve.

Example: Building a Digital Mentor

Imagine you want to capture the teaching style of a senior developer. You would run the collection command:

python3 ./kit/immortal_cli.py collect --platform slack --channel #coding-mentorship

The system then processes these messages using the procedural-extractor.md prompt. Instead of just knowing facts, the resulting AI skill will adopt the developer's specific way of explaining complex concepts and their unique debugging logic.

Best Practices for Builders

• Verify Evidence: Aim for a high ratio of verbatim (exact quotes) to impression. This prevents "AI hallucinations" where the twin acts like a generic assistant rather than the target individual.
• Evolutionary Updates: Use the version_tool.py to take snapshots of your persona. As you gather more data or receive feedback, you can iterate on the digital twin without losing previous progress.
• Role-Specific Pruning: Don't feed a "Colleague" twin personal family data. Keep the data source relevant to the template to maintain professional utility.

Explore more and start distilling your first persona at /skill/immortal-skill.