Plugin Guide

AWI Plugin for Claude Code

The AWI plugin adds automatic pre-flight source checks to Claude Code. Before your agent fetches web content, AWI assesses the source and reports what it observed.

Installation

Terminal
# Install the plugin
claude plugin add agenticwebindex/awi/awi-plugin

# Set your API key
export AWI_API_KEY=awi_sk_...

Skills

source-checkAuto-invoked

Automatically runs before any web fetch. Extracts the domain, calls awi.lookup, and reports the posture assessment. If the source is scored below 0.4, it warns before proceeding.

source-evalManual (/awi:source-eval)

Full dimensional breakdown of a source. Shows all 5 dimensions, trajectory, recommendations, and interaction pattern. Use when you want detailed intelligence before a critical decision.

scan-unknownManual (/awi:scan-unknown)

Triggers an on-demand scan for domains not yet in the AWI index. Returns scan results with posture assessment when complete.

How it works

1. Agent prepares to fetch a URL (e.g., https://example.com/api/data)

2. The source-check skill auto-invokes, extracting the domain

3. Calls awi.lookup via MCP to get the source assessment

4. Reports the posture (safe/proceed_with_caution/restrict/avoid) and any key observations

5. If the domain is unknown, offers to trigger an on-demand assessment

MCP Configuration

.mcp.json
{
  "mcpServers": {
    "awi": {
      "type": "http",
      "url": "https://mcp.agenticwebindex.com/mcp",
      "headers": {
        "Authorization": "Bearer ${AWI_API_KEY}"
      }
    }
  }
}

Current limitations

  • Plugin is advisory only — it warns but does not block access
  • No local cache — every lookup hits the AWI server
  • Thresholds are currently hardcoded in skill definitions
  • Only web fetch is checked; MCP install and skill load are not yet intercepted