Sibo-Zhao/OpenPraxis
An OpenClaw-native knowledge retention skill that turns raw inputs into structured practice so you can use what you know, not just store it.
What it solves
People collect notes, articles, reflections, but often fail to retain and apply them. OpenPraxis transforms raw inputs into practice loops — generate questions, answer them, get scored, and produce insight cards — so knowledge becomes usable.
How it works
You add a file (text, markdown, or image) with praxis add. For images, a vision-capable model extracts text first. The system classifies the input by type (report, interview, reflection, idea) and stores it. You can then generate a practice scene with praxis practice, optionally specifying a role, context, and time limit. The scene is a question or exercise based on the source. You answer it with praxis answer, optionally in an editor or from a file. The answer is evaluated and scored. You can continue with follow-ups until evaluation completes. The conversation is saved. You can also generate "insight cards" from your inputs, which summarize key takeaways. All data is stored locally; search and list commands run without an LLM call. When running as an OpenClaw skill, LLM calls use the host agent's model with zero API key setup. Standalone CLI users configure their own provider (openai, doubao, kimi, deepseek). Architecture separates business workflow from LLM calls via a backend abstraction.
Who it’s for
OpenClaw users who want to retain knowledge from their agent conversations, and standalone CLI users who want a structured practice system for notes and documents. It's for learners, professionals, and teams who want to turn information into actionable skills.
Highlights
- Zero-config integration with OpenClaw: uses the host agent's model.
- Accepts text and image inputs, with vision-based text extraction.
- Generates practice scenes with role, context, and time constraints.
- Evaluates answers with scores and saves full conversation history.
- Local search and list commands that don't call an LLM.
- Supports multiple LLM providers: OpenAI, Doubao, Kimi, DeepSeek.
- Export to markdown or JSON.
- Pending sessions persist across CLI restarts.
Related
- Project
- Project
- Project
- Project