> ## Documentation Index
> Fetch the complete documentation index at: https://docs.chainstream.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Skills

> Structured AI capability packages for on-chain data and DeFi execution

## What Are Agent Skills

Agent Skills are structured instruction packages (`SKILL.md` files) that teach AI coding assistants how to use ChainStream's on-chain data and DeFi capabilities. Unlike raw API docs, skills provide **decision trees, workflows, safety rules, and error recovery** — everything an AI agent needs to operate autonomously.

<CardGroup cols={3}>
  <Card title="chainstream-data" icon="magnifying-glass" color="#4D9CFF">
    **Tool pattern** — read-only on-chain data: token analytics, market trends, wallet profiling, WebSocket streams
  </Card>

  <Card title="chainstream-graphql" icon="diagram-project" color="#10B981">
    **Tool pattern** — custom GraphQL analytics over 27 cubes: cross-cube JOINs, aggregations, time-series
  </Card>

  <Card title="chainstream-defi" icon="right-left" color="#9333EA">
    **Process pattern** — irreversible DeFi execution: swap, launchpad, transaction broadcast
  </Card>
</CardGroup>

## Skills vs MCP vs SDK

| Layer            | What It Is                                                                                | Best For                                          |
| ---------------- | ----------------------------------------------------------------------------------------- | ------------------------------------------------- |
| **Agent Skills** | High-level AI instruction set (SKILL.md) with decision trees, workflows, and safety rules | AI coding assistants (Cursor, Claude Code, Codex) |
| **MCP Server**   | Model Context Protocol — 17 tools callable by AI models                                   | AI chat assistants (Claude Desktop, ChatGPT)      |
| **CLI**          | Command-line tool with wallet and x402 payment                                            | Scripts, CI/CD, AI agents needing DeFi            |
| **SDK**          | TypeScript/Python/Go/Rust client library                                                  | Custom applications                               |

Skills sit at the **highest abstraction layer** — they reference MCP tools and CLI commands internally, routing the AI agent to the right tool for each task.

## Routing Decision Tree

```mermaid theme={null}
flowchart TD
  Start["User request"] --> IsExec{"Executing transaction?"}
  IsExec -->|Yes| DefiSkill["chainstream-defi"]
  IsExec -->|No| IsData{"Reading data?"}
  IsData -->|No| NotCovered["Not a ChainStream task"]
  IsData -->|Yes| Custom{"Needs JOINs / custom<br/>aggregation / time-series?"}
  Custom -->|Yes| GraphqlSkill["chainstream-graphql"]
  Custom -->|No| DataSkill["chainstream-data"]

  DataSkill --> HasMCP{"MCP configured?"}
  HasMCP -->|Yes| UseMCP["Use MCP tools"]
  HasMCP -->|No| UseCLI["Use CLI commands"]

  GraphqlSkill --> Schema["graphql schema --summary<br/>graphql query --file ..."]

  DefiSkill --> HasWallet{"Wallet available?"}
  HasWallet -->|Yes| UseCLIDeFi["Use CLI for signing"]
  HasWallet -->|No| NeedWallet["Prompt: chainstream login"]
```

## Skill Comparison

| Aspect            | chainstream-data                           | chainstream-graphql                         | chainstream-defi                                       |
| ----------------- | ------------------------------------------ | ------------------------------------------- | ------------------------------------------------------ |
| Pattern           | Tool (read-only)                           | Tool (read-only)                            | Process (execute)                                      |
| Risk Level        | Low                                        | Low                                         | High (irreversible)                                    |
| Wallet Required   | No (API Key sufficient)                    | No (API Key sufficient)                     | Yes (signing needed)                                   |
| MCP Support       | Full (17 tools)                            | CLI-driven                                  | Tools available, but execution requires wallet on host |
| User Confirmation | Not required                               | Not required                                | **Mandatory** before every transaction                 |
| Typical Actions   | Search, analyze, track, stream             | JOIN, aggregate, time-series, complex WHERE | Swap, create token, broadcast                          |
| Best For          | Standard analytics via pre-built endpoints | Custom analytics that REST doesn't expose   | Trading, launching tokens, signing                     |

## Shared Resources

All skills share common reference documents:

| Resource           | Content                                                          |
| ------------------ | ---------------------------------------------------------------- |
| **Authentication** | Four auth paths (API Key, wallet login, raw key, Tempo MPP)      |
| **x402 Payment**   | x402 and MPP payment protocols, plan selection flow              |
| **Error Handling** | HTTP status codes, retry strategies, DeFi-specific errors        |
| **Chains**         | Supported chains matrix, native token addresses, block explorers |

## Supported Platforms

Skills work with any AI coding assistant that supports `SKILL.md` files:

| Platform    | Installation Method                   |
| ----------- | ------------------------------------- |
| Cursor      | Auto-discovered via `.cursor-plugin/` |
| Claude Code | `/plugin install chainstream`         |
| Codex       | Clone + symlink                       |
| OpenCode    | Clone + symlink                       |
| Gemini CLI  | `gemini extensions install`           |

See [Installation Guide](/en/docs/ai-agents/agent-skills/installation) for setup instructions.

## Next Steps

<CardGroup cols={2}>
  <Card title="Installation" icon="download" href="/en/docs/ai-agents/agent-skills/installation">
    Set up skills on your platform
  </Card>

  <Card title="chainstream-data" icon="magnifying-glass" href="/en/docs/ai-agents/agent-skills/chainstream-data">
    Standard data queries and analytics
  </Card>

  <Card title="chainstream-graphql" icon="diagram-project" href="/en/docs/ai-agents/agent-skills/chainstream-graphql">
    Custom GraphQL analytics, JOINs, aggregations
  </Card>

  <Card title="chainstream-defi" icon="right-left" href="/en/docs/ai-agents/agent-skills/chainstream-defi">
    DeFi execution workflows
  </Card>

  <Card title="MCP Server" icon="plug" href="/en/docs/ai-agents/mcp-server/introduction">
    Underlying MCP protocol
  </Card>
</CardGroup>