context-graph

majiayu000

Updated Today

1 views

Otherai

About

The context-graph skill enables Claude to store, search, and learn from past decision traces using semantic search with Voyage AI embeddings and ChromaDB vector storage. Use it when you need to query historical precedents, implement learning loops, or maintain persistent memory across conversations. It provides tools for storing traces, performing semantic searches, and updating outcomes to improve future decisions.

Quick Install

Claude Code

Recommended

Plugin CommandRecommended

/plugin add https://github.com/majiayu000/claude-skill-registry

Git CloneAlternative

git clone https://github.com/majiayu000/claude-skill-registry.git ~/.claude/skills/context-graph

Copy and paste this command in Claude Code to install this skill

Documentation

Context Graph

Living records of decision traces with semantic search. Find similar past decisions by meaning, not keywords.

Setup

MCP Server (recommended):

The context-graph MCP server provides the same functionality via tools:

context_store_trace - Store decisions with embeddings
context_query_traces - Semantic search
context_get_trace - Get by ID
context_update_outcome - Mark success/failure
context_list_traces - List with pagination
context_list_categories - Category breakdown

Configure in .claude/mcp.json:

{
  "mcpServers": {
    "context-graph": {
      "command": "uv",
      "args": ["--directory", "context-graph-mcp", "run", "python", "server.py"],
      "env": {"VOYAGE_API_KEY": "your_key_here"}
    }
  }
}

CLI Scripts (alternative):

# 1. Install dependencies
pip install voyageai chromadb

# 2. Set Voyage AI key
export VOYAGE_API_KEY="your_key_here"

# 3. Store/query traces
python scripts/store-trace.py "DECISION"
python scripts/query-traces.py "similar situation"

Instructions

Store trace after decisions with category + outcome
Query precedents when facing similar situations
Update outcome to success/failure after validation

Quick Commands (MCP)

context_store_trace(decision="Chose FastAPI for async", category="framework")
context_query_traces(query="web framework choice", limit=5)
context_update_outcome(trace_id="trace_abc...", outcome="success")

Quick Commands (CLI)

# Store a decision trace
python scripts/store-trace.py "Chose FastAPI over Flask for async support" --category framework

# Find similar past decisions
python scripts/query-traces.py "web framework selection"

# Query by category
python scripts/query-traces.py "database choice" --category architecture --limit 3

# Output JSON for parsing
python scripts/query-traces.py "error handling" --json

Trace Schema

Field	Description
`id`	Unique trace identifier
`timestamp`	When stored
`category`	Grouping (framework, api, error, etc.)
`decision`	What was decided (text)
`outcome`	pending / success / failure
`state`	State machine state when decided
`feature_id`	Related feature (if any)
`embedding`	1024-dim vector (Voyage AI)

When to Use

Situation	Action
Made a technical decision	Store trace with category
Facing similar problem	Query traces before deciding
Session complete	Query category → extract patterns
Repeating error	Query traces for that error

GitHub Repository

majiayu000/claude-skill-registry

Path: skills/context-graph

Related Skills

sglang

evaluating-llms-harness

Testing

This Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.

View skill

langchain

llamaguard

Other

LlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.