name: openclaw-knowledge-coach description: Build and operate an OpenClaw-based local knowledge assistant that imports personal/local documents into a knowledge base and creates practice exercises during import. Use when users ask to set up OpenClaw knowledge workflows, ingest local notes/files, structure chunks and tags, or generate retrieval practice (quiz, flashcards, recall prompts) to master stored knowledge.

OpenClaw Knowledge Coach

Create a local knowledge workflow in OpenClaw where importing knowledge also produces practice material for retention. OpenPraxis is on PyPI: use pip install openpraxis to get the praxis CLI.

CLI First

Use OpenPraxis CLI as the default execution path.

Install from PyPI (recommended):

pip install openpraxis
praxis --help

Or install from source for development:

git clone https://github.com/Sibo-Zhao/OpenPraxis.git
cd OpenPraxis
pip install -e ".[dev]"
praxis --help

Configure provider/model/API key before ingestion/practice:

praxis llm setup
praxis llm show

Use environment variables when needed (higher priority than config file):

export OPENAI_API_KEY="your_key_here"
# or ARK_API_KEY / MOONSHOT_API_KEY / DEEPSEEK_API_KEY based on provider

Core Workflow

1. Confirm scope and source
2. Confirm knowledge domains, source folders, and accepted file types.

3. Confirm whether to preserve existing metadata (tags, dates, project names).

4. Define import contract

5. Normalize each source into a record with doc_id, title, source_path, tags, created_at, and content.

6. Split long content into chunks with stable IDs such as doc_id#chunk-001.

7. Import into OpenClaw

8. Ingest normalized records into the local OpenClaw knowledge base.

9. Keep a deterministic mapping between source file and imported IDs for later updates.

10. Generate exercises at import time

11. For each chunk, create at least one retrieval exercise.
12. Prefer three exercise types:
13. free-recall: ask the user to explain from memory.
14. qa: ask direct question-answer pairs.
15. application: ask scenario-based transfer questions.

16. Save answer keys and concise grading rubrics.

17. Build review queue

18. Group exercises by topic and difficulty.

19. Schedule spaced review windows (for example: day 1, day 3, day 7, day 14).

20. Validate quality

21. Reject exercises that can be answered without the imported knowledge.
22. Reject ambiguous or duplicate questions.
23. Ensure every exercise points back to doc_id and chunk_id.

CLI Command Playbook

Run this sequence when the user asks to import local knowledge and create practice:

1. Add a local file

praxis add "/absolute/path/to/note.md" --type report

1. List recent inputs and capture target input_id

praxis list --limit 20

1. Force-generate a new practice scene for an existing input

praxis practice <input_id>

1. Submit answer by file (preferred for deterministic runs)

praxis answer <scene_id> --file "/absolute/path/to/answer.md"

1. Inspect pipeline results and insight cards

praxis show <input_id>
praxis insight <input_id>

1. Export insights to Markdown/JSON

praxis export --format md --output "/absolute/path/to/insights.md"
praxis export --format json --output "/absolute/path/to/insights.json"

Agent Execution Rules

• Prefer praxis add for import and initial exercise generation.
• Parse IDs from CLI output, then chain praxis practice and praxis answer.
• Use praxis answer --file instead of interactive stdin in automation flows.
• If duplicate content is skipped, rerun with praxis add ... --force when user wants reprocessing.
• Use one-shot runtime model override only when requested:

praxis --provider openai --model gpt-4.1-mini add "/absolute/path/to/note.md"

• For image notes, pass image file path directly to praxis add; OCR extraction is built in.
• Always finish with praxis show plus praxis insight or praxis export so user gets concrete output artifacts.

Output Contract

When executing tasks with this skill, always provide these outputs:

• Import summary: files processed, chunks created, failures.
• Exercise summary: counts by type/topic/difficulty.
• Review plan: next due batches and estimated workload.
• Traceability map: source -> doc_id -> chunk_id -> exercise_id.

Exercise Format

Use this compact JSON-like structure per exercise:

{
  "exercise_id": "ex-...",
  "doc_id": "...",
  "chunk_id": "...",
  "type": "free-recall | qa | application",
  "question": "...",
  "answer_key": "...",
  "rubric": ["point 1", "point 2"],
  "difficulty": "easy | medium | hard",
  "next_review": "YYYY-MM-DD"
}

For more generation patterns, read references/exercise-patterns.md.

这个 Skill 质量中等偏上。它将知识存储和复习练习结合起来的设计思路不错，文档层次分明，命令说明也比较详细。但它依赖一个叫 OpenPraxis 的外部工具，且没有明确说明这个工具怎么安装、有什么限制，使用时可能遇到问题。另外文档缺少实用例子和错误处理指南，新手上手可能有难度。如果你需要管理本地文档并生成练习，它的核心理念可行，但需要一定的技术基础来填补文档未覆盖的部分。