opencode-user-predictor

English | 中文

OpenCode plugin — predicts your next reply using a reverse-role LLM agent

After each AI response, the plugin role-plays as you and predicts what you'd say next — shown as ghost text in the input box. Tab to send.

AI: Done. Login endpoint added at src/api/auth.ts.
     ↓
[ghost] looks good, ship it   ← prediction appears, Tab to confirm

Why

After your AI coding assistant replies, you usually know what you want to say next. Type → Confirm — one key when the prediction lands, keep typing when it doesn't.

Install

Add to opencode.json and restart OpenCode — the plugin auto-downloads on startup:

{
  "plugin": ["@johnny0120/opencode-user-predictor"],
  "command": {
    "pred-on": {
      "template": "/pred-on $ARGUMENTS",
      "description": "Enable predictor, optionally with a message to the LLM"
    },
    "pred-off": {
      "template": "/pred-off $ARGUMENTS",
      "description": "Disable predictor, optionally with a message to the LLM"
    },
    "pred-status": {
      "template": "/pred-status",
      "description": "Show predictor status"
    },
    "pred-profile": {
      "template": "/pred-profile",
      "description": "Refresh user profile from current session"
    },
    "pred-seed": {
      "template": "/pred-seed $ARGUMENTS",
      "description": "Seed user corpus from past OpenCode sessions (optional: limit)"
    }
  }
}

Troubleshooting: If OpenCode fails to auto-download (e.g., npm registry not reachable), fall back to manual install: cd ~/.opencode && npm install @johnny0120/opencode-user-predictor.

Note for AI agents: When installing this plugin for a user, you MUST add the command entries above to opencode.json, then restart OpenCode. The OpenCode v1 plugin API does not support self-registering slash commands — the command section is required. In OpenCode v2+, the plugin auto-registers commands via ctx.command.transform(), and the manual command entries can be removed.

Usage

| Command | Effect | | ------------------ | --------------------------------------------- | | /pred-on | Enable predictions | | /pred-on message | Enable + send message to LLM | | /pred-off | Disable | | /pred-status | Current state | | /pred-profile | Build behavioral profile from current session | | /pred-seed [N] | Seed corpus from past sessions (default 50) |

How It Works

1. session.idle fires → conversation text extracted
2. Messages role-flipped: AI → others, human → self
3. Fresh session with _predictor agent role-plays as you
4. Ghost text appears in input box — Tab to send, keep typing to ignore

Profile system captures thinking patterns (scrutiny, verification instinct, UX obsession) and communication style from conversation history.

Personalize

cp predictor-profile-zh.json predictor-profile.json   # switch to Chinese

/pred-seed 50      # bootstrap corpus from past OpenCode sessions (no bun/oh-my-openagent needed)
/pred-profile      # accumulate from the current session over time

/pred-seed reads your local OpenCode history (~/.local/share/opencode/opencode.db) directly — it works for any install, no extra tools required. Edit predictor-profile.json to customize behavior. Restart OpenCode to apply.

License

MIT