opencode-request-logger

An OpenCode plugin that logs raw LLM request bodies to disk for debugging and analysis.

What it does

Intercepts outgoing fetch calls to LLM providers (Anthropic, OpenAI, Google, Bedrock, etc.) and writes each request body as a timestamped JSON file. Logs are organized by session ID.

Log location: ~/.config/opencode/logs/request-logger/<sessionId>/<timestamp>_<seq>.json

Each file contains:

{
  "timestamp": "2026-03-04T12:00:00.000Z",
  "url": "https://api.anthropic.com/v1/messages",
  "body": { "...raw request body..." }
}

How it works

Two interception layers ensure coverage across all providers:

1. globalThis.fetch wrapper -- catches most providers since the AI SDK routes through fetch
2. chat.params hook -- wraps the per-request fetch passed to streamText, catching providers that bypass globalThis.fetch (e.g. AWS Bedrock)

Only requests with an AI-shaped body (containing messages, input, or contents arrays) are logged.

Installation

# From the plugin directory
npm install
npm run build

Then add to your OpenCode config (~/.config/opencode/config.json):

{
  "plugin": {
    "opencode-request-logger": {
      "path": "~/.config/opencode/plugin/opencode-request-logger"
    }
  }
}