dbskill: The Wittgensteinian Diagnostic Engine for Claude Code

How 12,307 "Knowledge Atoms" turned an AI agent into a ruthless business strategist.

• View on GitHub • More from dontbesilent2025

A giant multi-layered funnel suspended in a void. Messy cloud-like business ideas are poured into the top. Each layer of the funnel is a sharp metallic mesh. Only tiny solid geometric atomic facts fall out the bottom. This illustrates the dissolution process where vague ideas are refined into stark logical truths. — The dbskill framework acts as a logic funnel, stripping away vague business buzzwords until only actionable facts remain.

从 12,307 条推文中提炼方法论，做成 Claude Code skill。

dontbesilent, Project Creator/Maintainer · dontbesilent2025/dbskill

Key Takeaways

The dbskill framework uses a Wittgensteinian filter to challenge and dissolve flawed business premises before generating solutions.
The system derives its logic from a structured database of over 4,000 "knowledge atoms" distilled from thousands of social media posts.
A multi-agent orchestration layer simulates debates between historical economic personas to provide diverse strategic perspectives.
The tool identifies and halts "thinking masturbation" by confronting users with the psychological reasons they are avoiding execution.

The Art of Saying "No"

Most AI tools try to be helpful. They take your prompt and immediately start generating solutions, code, or copy. The dbskill repository takes the opposite approach. It assumes your initial premise is probably flawed. Built as a "Skill Provider" for Anthropic's Claude Code CLI, the project acts as a diagnostic gatekeeper rather than a generative assistant.

The core of this system is the dbs-diagnosis workflow. It employs a philosophy of "Dissolution over Solution." Before offering any business advice, the AI runs the user's query through a Wittgensteinian filter to identify language traps and assumption errors. If you ask for a marketing strategy for a "lifestyle brand," the AI will stop and force you to define what "lifestyle" actually means in your specific economic context.

Portrait of dontbesilent2025, creator of dbskill.

The 5-layer diagnosis pipeline intercepts user requests and forces conceptual clarity before generating any solutions.

4,176 Atoms of Experience

To give the AI this level of rigor, the developer had to build a specialized knowledge base. The project is a rare example of "Social Mining." The creator took over 12,000 unstructured social media posts and distilled them into a highly structured JSONL database.

A massive Victorian-style mechanical drill boring into a mountain made of glowing bird-like icons. Behind the drill, a mechanical arm stacks perfectly uniform numbered bricks. — The project treats social media feeds as raw ore, mining them to extract structured, high-confidence knowledge atoms.

This "Atomic Library" contains 4,176 distinct entries. Each atom is tagged with a confidence score and a specific type, such as an anti-pattern or a fundamental principle. This metadata allows the AI to weigh conflicting advice and prioritize negative constraints over positive suggestions.

{"id": "2024Q4_020", "type": "anti-pattern", "confidence": 0.95, "knowledge": "Pricing is the Product. Do not separate the two.", "original": "...", "skills": ["dbs-diagnosis"]}

Hayek vs. Mises in the Terminal

The repository pushes the boundaries of Claude Code by introducing multi-agent orchestration within a single skill. The /chatroom-austrian component simulates a debate between Austrian School economists Friedrich Hayek and Ludwig von Mises.

The Austrian chatroom orchestrates a multi-agent debate, using a 'Judge' persona to synthesize differing economic viewpoints.

Instead of a single AI trying to balance opposing views, the system uses parallel tool calls to trigger distinct personas with different thinking styles. A third "Judge" persona monitors the output to prevent endless philosophical roleplay and forces the agents to deliver actionable business takeaways.

Solving the "Thinking Masturbation" Problem

While the diagnosis skills handle the business logic, the dbs-action skill targets the human bottleneck. It recognizes that users often procrastinate by asking the AI for more research or endless planning frameworks.

A close-up of a human figure holding a shield made of to-do lists and research papers to block a single small shining arrow labeled Action. The shield is engraved with the words Thinking Masturbation. — The framework treats endless planning and research as a defensive shield used to avoid actual execution.

Grounded in Adlerian psychology, the system treats this hesitation as a teleological choice made to protect the user's ego. If the AI detects "Execution Simulator" behavior, it halts the business analysis and confronts the user about their psychological blocks.

Feature	Standard AI Assistance	dbskill Diagnosis
Primary Goal	Generate solutions quickly	Dissolve flawed premises
Interaction Style	Agreeable and expansive	Reductive and confrontational
Knowledge Source	General LLM pre-training	Structured JSONL atoms
Handling Delays	Provides more planning templates	Diagnoses human psychological bottlenecks

By treating philosophical strategy and psychological intervention as version-controlled code, dbskill points toward a future where AI agents are valued less for what they can write, and more for what they refuse to accept.