fabrik-codek
ikchain 作
Coding Agents & IDEs概要
Cognitive architecture for developers: hyper-personalization engine (personal profiling, competence model, adaptive.
説明
このスキルでできること
Fabrik-Codek is a personal cognitive architecture that runs locally with any Ollama model. It builds a knowledge graph from your work sessions, profiles your expertise per topic, and routes queries to the right retrieval strategy. Responses feed back into the system to adjust future retrieval.
Unlike generic RAG setups, it tracks your actual expertise level per topic and adjusts retrieval depth and model selection accordingly, without requiring manual feedback.
使いどころ
- Finding past code patterns by semantic meaning
- Getting context-aware answers based on your own session history
- Checking how concepts in your stack relate to each other
- Searching for exact terms across accumulated knowledge
- Monitoring which topics you have deep vs shallow coverage on
ワークフロー例
User asks: 'How should I handle database connection pooling in my stack?'
- 1
Task Router classifies the query as an architecture/design task and detects the database topic
- 2
Competence Model looks up the user's expertise score for that topic
- 3
Retrieval strategy is selected: hybrid vector + graph search given moderate competence
- 4
Knowledge graph is traversed for entities related to connection pooling in the user's history
- 5
A personalized system prompt is assembled and sent to the local Ollama model
A context-aware answer drawing on the user's own past decisions and patterns, with retrieval parameters logged for outcome tracking
要件
- Fabrik-Codek installed from source: git clone https://github.com/ikchain/Fabrik-Codek + pip install -e ".[dev]"
- Ollama installed and running locally (ollama.ai)
- Ollama model pulled locally, e.g. ollama pull qwen2.5-coder:7b
- Meilisearch running locally (optional, for full-text search only)
インストール
clawhub install ikchain/fabrik-codekCLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。