القائمة
基于用户画像和向量化笔记提供个性化对话。当用户需要个性化交流、上下文感知的回应，或希望 AI 记住并引用其之前的想法和笔记时使用。
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | ai-partner-chat |
| description | 基于用户画像和向量化笔记提供个性化对话。当用户需要个性化交流、上下文感知的回应，或希望 AI 记住并引用其之前的想法和笔记时使用。 |
Provide personalized, context-aware conversations by integrating user persona, AI persona, and vectorized personal notes. This skill enables AI to remember and reference the user's previous thoughts, preferences, and knowledge base, creating a more coherent and personalized interaction experience.
Before first use, complete these steps in order:
Create directory structure
mkdir -p config notes vector_db scripts
Set up Python environment
python3 -m venv venv
./venv/bin/pip install -r .claude/skills/ai-partner-chat/scripts/requirements.txt
Note: First run will download embedding model (~4.3GB)
Generate persona templates
Copy from .claude/skills/ai-partner-chat/assets/ to config/:
user-persona-template.md → config/user-persona.mdai-persona-template.md → config/ai-persona.mdUser adds notes
Place markdown notes in notes/ directory (any format/structure)
Initialize vector database (see section 1.2 below)
Now proceed to Core Workflow →
Before using this skill for the first time, complete the following setup:
Create two Markdown files to define interaction parameters:
User Persona (user-persona.md):
assets/user-persona-template.mdAI Persona (ai-persona.md):
assets/ai-persona-template.mdThis skill uses AI Agent approach for intelligent note chunking:
When you initialize the vector database, Claude Code will:
<project_root>/notes/ directorychunk_schema.Chunk format<project_root>/vector_db/Key advantages:
Chunk Format Requirement:
All chunks must conform to this schema (see scripts/chunk_schema.py):
{
'content': 'chunk text content',
'metadata': {
'filename': 'note.md', # Required
'filepath': '/path/to/file', # Required
'chunk_id': 0, # Required
'chunk_type': 'date_entry', # Required
'date': '2025-11-07', # Optional
'title': 'Section title', # Optional
}
}
Location: Create <project_root>/scripts/chunk_and_index.py
Required structure:
# Import provided utilities
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent / ".claude/skills/ai-partner-chat/scripts"))
from chunk_schema import Chunk, validate_chunk
from vector_indexer import VectorIndexer
def chunk_note_file(filepath: str) -> List[Chunk]:
"""
Analyze THIS file's format and generate appropriate chunks.
Each chunk must conform to chunk_schema.Chunk format:
{
'content': 'text',
'metadata': {
'filename': 'file.md',
'filepath': '/path/to/file',
'chunk_id': 0,
'chunk_type': 'your_label'
}
}
"""
# TODO: Analyze actual file format (NOT template-based)
# TODO: Generate chunks based on analysis
# TODO: Validate each chunk with validate_chunk()
pass
def main():
# Initialize vector database
indexer = VectorIndexer(db_path="./vector_db")
indexer.initialize_db()
# Process all note files
all_chunks = []
for note_file in Path("./notes").glob("**/*"):
if note_file.is_file():
chunks = chunk_note_file(str(note_file))
all_chunks.extend(chunks)
# Index chunks
indexer.index_chunks(all_chunks)
if __name__ == "__main__":
main()
Execute: ./venv/bin/python scripts/chunk_and_index.py
Key points:
chunk_note_file() function logic should be dynamically created based on analyzing actual file contentchunk_schema.ChunkFor each user query, follow this process:
Read both persona files to understand:
Query the vector database to find the top 5 most semantically similar notes:
from scripts.vector_utils import get_relevant_notes
# Query for relevant context
relevant_notes = get_relevant_notes(
query=user_query,
db_path="./vector_db",
top_k=5
)
Or use the command-line tool:
python scripts/query_notes.py "user query text" --top-k 5
Combine the following elements to inform the response:
Synthesize a response that:
When Referencing Notes:
Example Response Pattern:
[Acknowledge user's query in preferred communication style]
[Incorporate relevant note context naturally if applicable]
"I remember you mentioned [insight from note] - this connects well with..."
[Provide main response following AI persona guidelines]
[Optional: Ask follow-up question based on user's learning style]
When the user creates new notes, add them to the vector database:
python scripts/add_note.py /path/to/new_note.md
Personas can be updated anytime by editing the Markdown files. Changes take effect in the next conversation.
To completely rebuild the vector database:
python scripts/init_vector_db.py /path/to/notes --db-path ./vector_db
This will delete the existing database and re-index all notes.
User data is stored in project root, not inside the skill directory:
<project_root>/
├── notes/ # User's markdown notes
├── vector_db/ # ChromaDB vector database
├── venv/ # Python dependencies
├── config/
│ ├── user-persona.md # User persona definition
│ └── ai-persona.md # AI persona definition
└── .claude/skills/ai-partner-chat/ # Skill code (can be deleted/reinstalled)
├── SKILL.md
└── scripts/
├── chunk_schema.py # Chunk format specification
├── vector_indexer.py # Core indexing utilities
└── vector_utils.py # Query utilities
Design principles:
Philosophy: Instead of pre-written chunking strategies, Claude Code analyzes each note and generates optimal chunking code on the fly.
How it works:
chunk_schema.Chunk formatvector_indexer.pyBenefits:
<project_root>/vector_db/)chunk_schema.py: Defines required chunk format specificationvector_indexer.py: Core utilities for embedding generation and ChromaDB indexingvector_utils.py: Query utilities for retrieving relevant chunksrequirements.txt: Python dependencies (chromadb, sentence-transformers)Note: No pre-written chunking scripts. Chunking is done by Claude Code dynamically.
<project_root>/
├── notes/ # User's notes (managed by user)
│ └── *.md
├── vector_db/ # Vector database (auto-generated)
├── venv/ # Python environment
├── config/ # User configuration
│ ├── user-persona.md
│ └── ai-persona.md
└── .claude/skills/ai-partner-chat/
├── SKILL.md # This file
├── scripts/
│ ├── chunk_schema.py # Chunk format spec
│ ├── vector_indexer.py # Indexing utilities
│ ├── vector_utils.py # Query utilities
│ └── requirements.txt # Dependencies
└── assets/
├── user-persona-template.md
└── ai-persona-template.md
<project_root>/notes/ anytimeDatabase Connection Errors:
<project_root>/vector_db/ directory exists and is writablePoor Retrieval Quality:
top_k value for more contextChunking Issues: