opencode-gpt-talk-normal

OpenCode server plugin — automatically injects the talk-normal system prompt when using GPT models.

What it does

When model.id or model.modelID contains gpt (case-insensitive), this plugin prepends the talk-normal prompt to the system message before it reaches the LLM. Non-GPT models are completely unaffected.

The prompt source is talk-normal v0.6.1 by hexiecs.

Install

# via opencode CLI (recommended)
opencode plugin opencode-gpt-talk-normal

# or pnpm
pnpm add -D opencode-gpt-talk-normal

# or npm
npm install -D opencode-gpt-talk-normal

Configure

Add to your opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-gpt-talk-normal"]
}

Or with a local path for development:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["/absolute/path/to/opencode-gpt-talk-normal"]
}

Behavior

| Condition | Result | |---|---| | model.id contains gpt | talk-normal prompt prepended to system[] | | model is Claude, Gemini, etc. | no change | | prompt already injected | skipped (idempotent) |

E2E verified

Tested against real openai/gpt-5.4 through OpenCode's plugin hook (experimental.chat.system.transform). The captured request confirmed talk-normal appears as the first system message, while a control run with claude-sonnet-4.6 showed no injection.

Test

node --test

License

MIT